tclf90
commited on
Commit
•
732c58a
1
Parent(s):
4f089e5
'首次commit'
Browse files- LICENSE +84 -0
- README.md +61 -17
- added_tokens.json +16 -0
- config.json +74 -0
- configuration.json +4 -0
- configuration_chatglm.py +58 -0
- generation_config.json +10 -0
- model-00001-of-00003.safetensors +3 -0
- model-00002-of-00003.safetensors +3 -0
- model-00003-of-00003.safetensors +3 -0
- model.safetensors.index.json +891 -0
- modeling_chatglm.py +1218 -0
- special_tokens_map.json +32 -0
- tokenization_chatglm.py +323 -0
- tokenizer.model +3 -0
- tokenizer_config.json +146 -0
LICENSE
ADDED
|
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
The glm-4-9b License
|
| 2 |
+
|
| 3 |
+
1. 定义
|
| 4 |
+
|
| 5 |
+
“许可方”是指分发其软件的 glm-4-9b 模型团队。
|
| 6 |
+
“软件”是指根据本许可提供的 glm-4-9b 模型参数。
|
| 7 |
+
|
| 8 |
+
2. 许可授予
|
| 9 |
+
|
| 10 |
+
根据本许可的条款和条件，许可方特此授予您非排他性、全球性、不可转让、不可再许可、可撤销、免版税的版权许可。
|
| 11 |
+
本许可允许您免费使用本仓库中的所有开源模型进行学术研究，对于希望将模型用于商业目的的用户，需在[这里](https://open.bigmodel.cn/mla/form)完成登记。经过登记的用户可以免费使用本模型进行商业活动，但必须遵守本许可的所有条款和条件。
|
| 12 |
+
上述版权声明和本许可声明应包含在本软件的所有副本或重要部分中。
|
| 13 |
+
如果您分发或提供 THUDM / 智谱AI 关于 glm-4 开源模型的材料（或其任何衍生作品），或使用其中任何材料（包括 glm-4 系列的所有开源模型）的产品或服务，您应:
|
| 14 |
+
|
| 15 |
+
(A) 随任何此类 THUDM / 智谱AI 材料提供本协议的副本；
|
| 16 |
+
(B) 在相关网站、用户界面、博客文章、关于页面或产品文档上突出显示 “Built with glm-4”。
|
| 17 |
+
如果您使用 THUDM / 智谱AI的 glm-4 开源模型的材料来创建、训练、微调或以其他方式改进已分发或可用的 AI 模型，您还应在任何此类 AI 模型名称的开头添加 “glm-4”。
|
| 18 |
+
|
| 19 |
+
3. 限制
|
| 20 |
+
|
| 21 |
+
您不得出于任何军事或非法目的使用、复制、修改、合并、发布、分发、复制或创建本软件的全部或部分衍生作品。
|
| 22 |
+
您不得利用本软件从事任何危害国家安全和国家统一，危害社会公共利益及公序良俗，侵犯他人商业秘密、知识产权、名誉权、肖像权、财产权等权益的行为。
|
| 23 |
+
您在使用中应遵循使用地所适用的法律法规政策、道德规范等要求。
|
| 24 |
+
|
| 25 |
+
4. 免责声明
|
| 26 |
+
|
| 27 |
+
本软件“按原样”提供，不提供任何明示或暗示的保证，包括但不限于对适销性、特定用途的适用性和非侵权性的保证。
|
| 28 |
+
在任何情况下，作者或版权持有人均不对任何索赔、损害或其他责任负责，无论是在合同诉讼、侵权行为还是其他方面，由软件或软件的使用或其他交易引起、由软件引起或与之相关
|
| 29 |
+
软件。
|
| 30 |
+
|
| 31 |
+
5. 责任限制
|
| 32 |
+
|
| 33 |
+
除适用法律禁止的范围外，在任何情况下且根据任何法律理论，无论是基于侵权行为、疏忽、合同、责任或其他原因，任何许可方均不对您承担任何直接、间接、特殊、偶然、示范性、
|
| 34 |
+
或间接损害，或任何其他商业损失，即使许可人已被告知此类损害的可能性。
|
| 35 |
+
|
| 36 |
+
6. 争议解决
|
| 37 |
+
|
| 38 |
+
本许可受中华人民共和国法律管辖并按其解释。 因本许可引起的或与本许可有关的任何争议应提交北京市海淀区人民法院。
|
| 39 |
+
请注意，许可证可能会更新到更全面的版本。 有关许可和版权的任何问题，请通过 [email protected] 与我们联系。
|
| 40 |
+
|
| 41 |
+
1. Definitions
|
| 42 |
+
|
| 43 |
+
“Licensor” means the glm-4-9b Model Team that distributes its Software.
|
| 44 |
+
“Software” means the glm-4-9b model parameters made available under this license.
|
| 45 |
+
|
| 46 |
+
2. License
|
| 47 |
+
|
| 48 |
+
Subject to the terms and conditions of this License, Licensor hereby grants you a non-exclusive, worldwide, irrevocable, non-sublicensable, revocable, photo-free copyright license.
|
| 49 |
+
This license allows you to use all open source models in this repository for free for academic research. For users who wish to use the models for commercial purposes, please do so [here](https://open.bigmodel.cn/mla/form)
|
| 50 |
+
Complete registration. Registered users are free to use this model for commercial activities, but must comply with all terms and conditions of this license.
|
| 51 |
+
The copyright notice and this license notice shall be included in all copies or substantial portions of the Software.
|
| 52 |
+
If you distribute or provide THUDM / Zhipu AI materials on the glm-4 open source model (or any derivative works thereof), or products or services that use any materials therein (including all open source models of the glm-4 series), you should:
|
| 53 |
+
|
| 54 |
+
(A) Provide a copy of this Agreement with any such THUDM/Zhipu AI Materials;
|
| 55 |
+
(B) Prominently display "Built with glm-4" on the relevant website, user interface, blog post, related page or product documentation.
|
| 56 |
+
If you use materials from THUDM/Zhipu AI's glm-4 model to create, train, operate, or otherwise improve assigned or available AI models, you should also add "glm-4" to the beginning of any such AI model name.
|
| 57 |
+
|
| 58 |
+
3. Restrictions
|
| 59 |
+
|
| 60 |
+
You are not allowed to use, copy, modify, merge, publish, distribute, copy or create all or part of the derivative works of this software for any military or illegal purposes.
|
| 61 |
+
You are not allowed to use this software to engage in any behavior that endangers national security and unity, endangers social public interests and public order, infringes on the rights and interests of others such as trade secrets, intellectual property rights, reputation rights, portrait rights, and property rights.
|
| 62 |
+
You should comply with the applicable laws, regulations, policies, ethical standards, and other requirements in the place of use during use.
|
| 63 |
+
|
| 64 |
+
4. Disclaimer
|
| 65 |
+
|
| 66 |
+
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE
|
| 67 |
+
WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR
|
| 68 |
+
COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
|
| 69 |
+
OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
|
| 70 |
+
|
| 71 |
+
5. Limitation of Liability
|
| 72 |
+
|
| 73 |
+
EXCEPT TO THE EXTENT PROHIBITED BY APPLICABLE LAW, IN NO EVENT AND UNDER NO LEGAL THEORY, WHETHER BASED IN TORT,
|
| 74 |
+
NEGLIGENCE, CONTRACT, LIABILITY, OR OTHERWISE WILL ANY LICENSOR BE LIABLE TO YOU FOR ANY DIRECT, INDIRECT, SPECIAL,
|
| 75 |
+
INCIDENTAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES, OR ANY OTHER COMMERCIAL LOSSES, EVEN IF THE LICENSOR HAS BEEN ADVISED
|
| 76 |
+
OF THE POSSIBILITY OF SUCH DAMAGES.
|
| 77 |
+
|
| 78 |
+
6. Dispute Resolution
|
| 79 |
+
|
| 80 |
+
This license shall be governed and construed in accordance with the laws of People’s Republic of China. Any dispute
|
| 81 |
+
arising from or in connection with this License shall be submitted to Haidian District People's Court in Beijing.
|
| 82 |
+
|
| 83 |
+
Note that the license is subject to update to a more comprehensive version. For any questions related to the license and
|
| 84 |
+
copyright, please contact us at [email protected].
|
README.md
CHANGED
|
@@ -1,26 +1,70 @@
|
|
| 1 |
---
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 6 |
---
|
| 7 |
-
### 当前模型的贡献者未提供更加详细的模型介绍。模型文件和权重，可浏览“模型文件”页面获取。
|
| 8 |
-
#### 您可以通过如下git clone命令，或者ModelScope SDK来下载模型
|
| 9 |
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
```python
|
| 16 |
-
#SDK模型下载
|
| 17 |
from modelscope import snapshot_download
|
| 18 |
-
model_dir = snapshot_download('tclf90
|
| 19 |
```
|
| 20 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
```
|
| 22 |
-
|
| 23 |
-
git clone https://www.modelscope.cn/tclf90/glm-4-9b-chat-GPTQ-Int8.git
|
| 24 |
```
|
| 25 |
-
|
| 26 |
-
<p style="color: lightgrey;">如果您是本模型的贡献者，我们邀请您根据<a href="https://modelscope.cn/docs/ModelScope%E6%A8%A1%E5%9E%8B%E6%8E%A5%E5%85%A5%E6%B5%81%E7%A8%8B%E6%A6%82%E8%A7%88" style="color: lightgrey; text-decoration: underline;">模型贡献文档</a>，及时完善模型卡片内容。</p>
|
|
|
|
| 1 |
---
|
| 2 |
+
license: GLM-4
|
| 3 |
+
license_name: glm-4-9b
|
| 4 |
+
license_link: LICENSE
|
| 5 |
+
pipeline_tag: text-generation
|
| 6 |
+
tags:
|
| 7 |
+
- chatglm
|
| 8 |
+
- gptq
|
| 9 |
+
- int8
|
| 10 |
+
- 量化修复
|
| 11 |
+
- vLLM
|
| 12 |
---
|
|
|
|
|
|
|
| 13 |
|
| 14 |
+
# GLM-4-9B-Chat-GPTQ-Int8-量化修复
|
| 15 |
+
原模型 [ZhipuAI/glm-4-9b-chat](https://www.modelscope.cn/models/ZhipuAI/glm-4-9b-chat)
|
| 16 |
+
|
| 17 |
+
|
| 18 |
+
### 【模型更新日期】
|
| 19 |
+
``` 2024-06-05 23:50 ```
|
| 20 |
+
|
| 21 |
+
### 【模型大小】
|
| 22 |
+
`11.0GB`
|
| 23 |
+
|
| 24 |
+
### 【06-05 情况告知】
|
| 25 |
+
|
| 26 |
+
1. 目前需要用vllm entrypoint的方式来启动模型。
|
| 27 |
+
|
| 28 |
+
|
| 29 |
+
### 【更新日志】
|
| 30 |
+
|
| 31 |
+
```
|
| 32 |
+
2004-06-05 23:50
|
| 33 |
+
首次commit
|
| 34 |
+
|
| 35 |
```
|
| 36 |
+
|
| 37 |
+
### 【介绍】
|
| 38 |
+
|
| 39 |
+
GLM-4-9B 是智谱 AI 推出的最新一代预训练模型 GLM-4 系列中的开源版本。 在语义、数学、推理、代码和知识等多方面的数据集测评中，GLM-4-9B 及其人类偏好对齐的版本 GLM-4-9B-Chat 均表现出较高的性能。 除了能进行多轮对话，GLM-4-9B-Chat 还具备网页浏览、代码执行、自定义工具调用（Function Call）和长文本推理（支持最大 128K 上下文）等高级功能。 本代模型增加了多语言支持，支持包括日语，韩语，德语在内的 26 种语言。我们还推出了支持 1M 上下文长度（约 200 万中文字符）的模型。
|
| 40 |
+
|
| 41 |
+
[更多详情...](https://www.modelscope.cn/models/ZhipuAI/glm-4-9b-chat/summary)
|
| 42 |
+
|
| 43 |
+
### 【量化修复】
|
| 44 |
+
|
| 45 |
+
调优了现有 `AWQ` 与 `GPTQ` 量化算法的量化策略。带有`量化修复`标签的`Int3`模型，可以比肩默认`AWQ`与`GPTQ`算法的`Int8`模型的能力。
|
| 46 |
+
|
| 47 |
+
1. 量化修复可以极大减少模型的`1.乱吐字`、`2.无限循环`、`3.长文能力丢失`等量化损失造成的模型不可用的情况。
|
| 48 |
+
|
| 49 |
+
2. 调优后的量化模型，`AWQ`与`GPTQ`模型在能力上没有表现出明显区别。同时考虑到`GPTQ`的`vLLM`引擎的并发推理效率最好，所以不再制作`AWQ`模型。
|
| 50 |
+
|
| 51 |
+
3. 待工作完成后补充...
|
| 52 |
+
|
| 53 |
+
### 【同期量化修复模型】
|
| 54 |
+
待工作完成后补充...
|
| 55 |
+
|
| 56 |
+
### 【模型下载】
|
| 57 |
```python
|
|
|
|
| 58 |
from modelscope import snapshot_download
|
| 59 |
+
model_dir = snapshot_download('tclf90/模型名', cache_dir="本地路径")
|
| 60 |
```
|
| 61 |
+
|
| 62 |
+
### 【[vLLM](https://github.com/vllm-project/vllm)推理（目前仅限Linux）】
|
| 63 |
+
#### 1. Python 简易调试
|
| 64 |
+
|
| 65 |
+
待工作完成后补充...
|
| 66 |
+
|
| 67 |
+
#### 2. 类ChatGPT RESTFul API Server
|
| 68 |
```
|
| 69 |
+
>>> python -m vllm.entrypoints.openai.api_server --model 本地路径/tclf90/模型名称
|
|
|
|
| 70 |
```
|
|
|
|
|
|
added_tokens.json
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"<eop>": 151334,
|
| 3 |
+
"<sop>": 151333,
|
| 4 |
+
"<|assistant|>": 151337,
|
| 5 |
+
"<|begin_of_image|>": 151339,
|
| 6 |
+
"<|begin_of_video|>": 151341,
|
| 7 |
+
"<|end_of_image|>": 151340,
|
| 8 |
+
"<|end_of_video|>": 151342,
|
| 9 |
+
"<|endoftext|>": 151329,
|
| 10 |
+
"<|observation|>": 151338,
|
| 11 |
+
"<|system|>": 151335,
|
| 12 |
+
"<|user|>": 151336,
|
| 13 |
+
"[MASK]": 151330,
|
| 14 |
+
"[gMASK]": 151331,
|
| 15 |
+
"[sMASK]": 151332
|
| 16 |
+
}
|
config.json
ADDED
|
@@ -0,0 +1,74 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_name_or_path": "tclf90/glm-4-9b-chat-GPTQ-Int8",
|
| 3 |
+
"add_bias_linear": false,
|
| 4 |
+
"add_qkv_bias": true,
|
| 5 |
+
"apply_query_key_layer_scaling": true,
|
| 6 |
+
"apply_residual_connection_post_layernorm": false,
|
| 7 |
+
"architectures": [
|
| 8 |
+
"ChatGLMForConditionalGeneration"
|
| 9 |
+
],
|
| 10 |
+
"attention_dropout": 0.0,
|
| 11 |
+
"attention_softmax_in_fp32": true,
|
| 12 |
+
"auto_map": {
|
| 13 |
+
"AutoConfig": "configuration_chatglm.ChatGLMConfig",
|
| 14 |
+
"AutoModel": "modeling_chatglm.ChatGLMForConditionalGeneration",
|
| 15 |
+
"AutoModelForCausalLM": "modeling_chatglm.ChatGLMForConditionalGeneration",
|
| 16 |
+
"AutoModelForSeq2SeqLM": "modeling_chatglm.ChatGLMForConditionalGeneration",
|
| 17 |
+
"AutoModelForSequenceClassification": "modeling_chatglm.ChatGLMForSequenceClassification"
|
| 18 |
+
},
|
| 19 |
+
"bias_dropout_fusion": true,
|
| 20 |
+
"classifier_dropout": null,
|
| 21 |
+
"eos_token_id": [
|
| 22 |
+
151329,
|
| 23 |
+
151336,
|
| 24 |
+
151338
|
| 25 |
+
],
|
| 26 |
+
"ffn_hidden_size": 13696,
|
| 27 |
+
"fp32_residual_connection": false,
|
| 28 |
+
"hidden_dropout": 0.0,
|
| 29 |
+
"hidden_size": 4096,
|
| 30 |
+
"kv_channels": 128,
|
| 31 |
+
"layernorm_epsilon": 1e-06,
|
| 32 |
+
"model_type": "chatglm",
|
| 33 |
+
"multi_query_attention": true,
|
| 34 |
+
"multi_query_group_num": 2,
|
| 35 |
+
"num_attention_heads": 32,
|
| 36 |
+
"num_hidden_layers": 40,
|
| 37 |
+
"num_layers": 40,
|
| 38 |
+
"original_rope": true,
|
| 39 |
+
"pad_token_id": 151329,
|
| 40 |
+
"padded_vocab_size": 151552,
|
| 41 |
+
"post_layer_norm": true,
|
| 42 |
+
"quantization_config": {
|
| 43 |
+
"batch_size": 1,
|
| 44 |
+
"bits": 8,
|
| 45 |
+
"block_name_to_quantize": null,
|
| 46 |
+
"cache_block_outputs": true,
|
| 47 |
+
"damp_percent": 0.1,
|
| 48 |
+
"dataset": null,
|
| 49 |
+
"desc_act": false,
|
| 50 |
+
"exllama_config": {
|
| 51 |
+
"version": 1
|
| 52 |
+
},
|
| 53 |
+
"group_size": 64,
|
| 54 |
+
"max_input_length": null,
|
| 55 |
+
"model_seqlen": null,
|
| 56 |
+
"module_name_preceding_first_block": null,
|
| 57 |
+
"modules_in_block_to_quantize": null,
|
| 58 |
+
"pad_token_id": null,
|
| 59 |
+
"quant_method": "gptq",
|
| 60 |
+
"sym": true,
|
| 61 |
+
"tokenizer": null,
|
| 62 |
+
"true_sequential": true,
|
| 63 |
+
"use_cuda_fp16": true,
|
| 64 |
+
"use_exllama": true
|
| 65 |
+
},
|
| 66 |
+
"rmsnorm": true,
|
| 67 |
+
"rope_ratio": 500,
|
| 68 |
+
"seq_length": 131072,
|
| 69 |
+
"tie_word_embeddings": false,
|
| 70 |
+
"torch_dtype": "float16",
|
| 71 |
+
"transformers_version": "4.40.2",
|
| 72 |
+
"use_cache": true,
|
| 73 |
+
"vocab_size": 151552
|
| 74 |
+
}
|
configuration.json
ADDED
|
@@ -0,0 +1,4 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"framework": "Pytorch",
|
| 3 |
+
"task": "text-generation"
|
| 4 |
+
}
|
configuration_chatglm.py
ADDED
|
@@ -0,0 +1,58 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from transformers import PretrainedConfig
|
| 2 |
+
|
| 3 |
+
|
| 4 |
+
class ChatGLMConfig(PretrainedConfig):
|
| 5 |
+
model_type = "chatglm"
|
| 6 |
+
|
| 7 |
+
def __init__(
|
| 8 |
+
self,
|
| 9 |
+
num_layers=28,
|
| 10 |
+
padded_vocab_size=65024,
|
| 11 |
+
hidden_size=4096,
|
| 12 |
+
ffn_hidden_size=13696,
|
| 13 |
+
kv_channels=128,
|
| 14 |
+
num_attention_heads=32,
|
| 15 |
+
seq_length=2048,
|
| 16 |
+
hidden_dropout=0.0,
|
| 17 |
+
classifier_dropout=None,
|
| 18 |
+
attention_dropout=0.0,
|
| 19 |
+
layernorm_epsilon=1e-5,
|
| 20 |
+
rmsnorm=True,
|
| 21 |
+
apply_residual_connection_post_layernorm=False,
|
| 22 |
+
post_layer_norm=True,
|
| 23 |
+
add_bias_linear=False,
|
| 24 |
+
add_qkv_bias=False,
|
| 25 |
+
bias_dropout_fusion=True,
|
| 26 |
+
multi_query_attention=False,
|
| 27 |
+
multi_query_group_num=1,
|
| 28 |
+
rope_ratio=1,
|
| 29 |
+
apply_query_key_layer_scaling=True,
|
| 30 |
+
attention_softmax_in_fp32=True,
|
| 31 |
+
fp32_residual_connection=False,
|
| 32 |
+
**kwargs
|
| 33 |
+
):
|
| 34 |
+
self.num_layers = num_layers
|
| 35 |
+
self.vocab_size = padded_vocab_size
|
| 36 |
+
self.padded_vocab_size = padded_vocab_size
|
| 37 |
+
self.hidden_size = hidden_size
|
| 38 |
+
self.ffn_hidden_size = ffn_hidden_size
|
| 39 |
+
self.kv_channels = kv_channels
|
| 40 |
+
self.num_attention_heads = num_attention_heads
|
| 41 |
+
self.seq_length = seq_length
|
| 42 |
+
self.hidden_dropout = hidden_dropout
|
| 43 |
+
self.classifier_dropout = classifier_dropout
|
| 44 |
+
self.attention_dropout = attention_dropout
|
| 45 |
+
self.layernorm_epsilon = layernorm_epsilon
|
| 46 |
+
self.rmsnorm = rmsnorm
|
| 47 |
+
self.apply_residual_connection_post_layernorm = apply_residual_connection_post_layernorm
|
| 48 |
+
self.post_layer_norm = post_layer_norm
|
| 49 |
+
self.add_bias_linear = add_bias_linear
|
| 50 |
+
self.add_qkv_bias = add_qkv_bias
|
| 51 |
+
self.bias_dropout_fusion = bias_dropout_fusion
|
| 52 |
+
self.multi_query_attention = multi_query_attention
|
| 53 |
+
self.multi_query_group_num = multi_query_group_num
|
| 54 |
+
self.rope_ratio = rope_ratio
|
| 55 |
+
self.apply_query_key_layer_scaling = apply_query_key_layer_scaling
|
| 56 |
+
self.attention_softmax_in_fp32 = attention_softmax_in_fp32
|
| 57 |
+
self.fp32_residual_connection = fp32_residual_connection
|
| 58 |
+
super().__init__(**kwargs)
|
generation_config.json
ADDED
|
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_from_model_config": true,
|
| 3 |
+
"eos_token_id": [
|
| 4 |
+
151329,
|
| 5 |
+
151336,
|
| 6 |
+
151338
|
| 7 |
+
],
|
| 8 |
+
"pad_token_id": 151329,
|
| 9 |
+
"transformers_version": "4.40.2"
|
| 10 |
+
}
|
model-00001-of-00003.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:4c79e4ae12da0ae7b9b233f2fe8b6b3a46269c58f38c596dd030ab0ed7845505
|
| 3 |
+
size 4970796416
|
model-00002-of-00003.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c82ba03bd76f42c804706423fc4be54486cf53b255b86798acce637e318d14b4
|
| 3 |
+
size 4819195016
|
model-00003-of-00003.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:cf4b1f6373006239a705138c4c400bb9bf28703e3c90e70eb720a38e7a5e10b9
|
| 3 |
+
size 1241514128
|
model.safetensors.index.json
ADDED
|
@@ -0,0 +1,891 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"metadata": {
|
| 3 |
+
"total_size": 11031388224
|
| 4 |
+
},
|
| 5 |
+
"weight_map": {
|
| 6 |
+
"transformer.embedding.word_embeddings.weight": "model-00001-of-00003.safetensors",
|
| 7 |
+
"transformer.encoder.final_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 8 |
+
"transformer.encoder.layers.0.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 9 |
+
"transformer.encoder.layers.0.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 10 |
+
"transformer.encoder.layers.0.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 11 |
+
"transformer.encoder.layers.0.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 12 |
+
"transformer.encoder.layers.0.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 13 |
+
"transformer.encoder.layers.0.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 14 |
+
"transformer.encoder.layers.0.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 15 |
+
"transformer.encoder.layers.0.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 16 |
+
"transformer.encoder.layers.0.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 17 |
+
"transformer.encoder.layers.0.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 18 |
+
"transformer.encoder.layers.0.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 19 |
+
"transformer.encoder.layers.0.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 20 |
+
"transformer.encoder.layers.0.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 21 |
+
"transformer.encoder.layers.0.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 22 |
+
"transformer.encoder.layers.0.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 23 |
+
"transformer.encoder.layers.0.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 24 |
+
"transformer.encoder.layers.0.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 25 |
+
"transformer.encoder.layers.0.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 26 |
+
"transformer.encoder.layers.0.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 27 |
+
"transformer.encoder.layers.0.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 28 |
+
"transformer.encoder.layers.0.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 29 |
+
"transformer.encoder.layers.0.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 30 |
+
"transformer.encoder.layers.1.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 31 |
+
"transformer.encoder.layers.1.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 32 |
+
"transformer.encoder.layers.1.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 33 |
+
"transformer.encoder.layers.1.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 34 |
+
"transformer.encoder.layers.1.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 35 |
+
"transformer.encoder.layers.1.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 36 |
+
"transformer.encoder.layers.1.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 37 |
+
"transformer.encoder.layers.1.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 38 |
+
"transformer.encoder.layers.1.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 39 |
+
"transformer.encoder.layers.1.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 40 |
+
"transformer.encoder.layers.1.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 41 |
+
"transformer.encoder.layers.1.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 42 |
+
"transformer.encoder.layers.1.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 43 |
+
"transformer.encoder.layers.1.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 44 |
+
"transformer.encoder.layers.1.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 45 |
+
"transformer.encoder.layers.1.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 46 |
+
"transformer.encoder.layers.1.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 47 |
+
"transformer.encoder.layers.1.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 48 |
+
"transformer.encoder.layers.1.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 49 |
+
"transformer.encoder.layers.1.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 50 |
+
"transformer.encoder.layers.1.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 51 |
+
"transformer.encoder.layers.1.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 52 |
+
"transformer.encoder.layers.10.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 53 |
+
"transformer.encoder.layers.10.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 54 |
+
"transformer.encoder.layers.10.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 55 |
+
"transformer.encoder.layers.10.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 56 |
+
"transformer.encoder.layers.10.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 57 |
+
"transformer.encoder.layers.10.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 58 |
+
"transformer.encoder.layers.10.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 59 |
+
"transformer.encoder.layers.10.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 60 |
+
"transformer.encoder.layers.10.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 61 |
+
"transformer.encoder.layers.10.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 62 |
+
"transformer.encoder.layers.10.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 63 |
+
"transformer.encoder.layers.10.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 64 |
+
"transformer.encoder.layers.10.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 65 |
+
"transformer.encoder.layers.10.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 66 |
+
"transformer.encoder.layers.10.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 67 |
+
"transformer.encoder.layers.10.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 68 |
+
"transformer.encoder.layers.10.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 69 |
+
"transformer.encoder.layers.10.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 70 |
+
"transformer.encoder.layers.10.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 71 |
+
"transformer.encoder.layers.10.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 72 |
+
"transformer.encoder.layers.10.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 73 |
+
"transformer.encoder.layers.10.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 74 |
+
"transformer.encoder.layers.11.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 75 |
+
"transformer.encoder.layers.11.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 76 |
+
"transformer.encoder.layers.11.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 77 |
+
"transformer.encoder.layers.11.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 78 |
+
"transformer.encoder.layers.11.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 79 |
+
"transformer.encoder.layers.11.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 80 |
+
"transformer.encoder.layers.11.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 81 |
+
"transformer.encoder.layers.11.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 82 |
+
"transformer.encoder.layers.11.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 83 |
+
"transformer.encoder.layers.11.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 84 |
+
"transformer.encoder.layers.11.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 85 |
+
"transformer.encoder.layers.11.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 86 |
+
"transformer.encoder.layers.11.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 87 |
+
"transformer.encoder.layers.11.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 88 |
+
"transformer.encoder.layers.11.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 89 |
+
"transformer.encoder.layers.11.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 90 |
+
"transformer.encoder.layers.11.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 91 |
+
"transformer.encoder.layers.11.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 92 |
+
"transformer.encoder.layers.11.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 93 |
+
"transformer.encoder.layers.11.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 94 |
+
"transformer.encoder.layers.11.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 95 |
+
"transformer.encoder.layers.11.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 96 |
+
"transformer.encoder.layers.12.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 97 |
+
"transformer.encoder.layers.12.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 98 |
+
"transformer.encoder.layers.12.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 99 |
+
"transformer.encoder.layers.12.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 100 |
+
"transformer.encoder.layers.12.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 101 |
+
"transformer.encoder.layers.12.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 102 |
+
"transformer.encoder.layers.12.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 103 |
+
"transformer.encoder.layers.12.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 104 |
+
"transformer.encoder.layers.12.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 105 |
+
"transformer.encoder.layers.12.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 106 |
+
"transformer.encoder.layers.12.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 107 |
+
"transformer.encoder.layers.12.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 108 |
+
"transformer.encoder.layers.12.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 109 |
+
"transformer.encoder.layers.12.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 110 |
+
"transformer.encoder.layers.12.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 111 |
+
"transformer.encoder.layers.12.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 112 |
+
"transformer.encoder.layers.12.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 113 |
+
"transformer.encoder.layers.12.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 114 |
+
"transformer.encoder.layers.12.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 115 |
+
"transformer.encoder.layers.12.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 116 |
+
"transformer.encoder.layers.12.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 117 |
+
"transformer.encoder.layers.12.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 118 |
+
"transformer.encoder.layers.13.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 119 |
+
"transformer.encoder.layers.13.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 120 |
+
"transformer.encoder.layers.13.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 121 |
+
"transformer.encoder.layers.13.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 122 |
+
"transformer.encoder.layers.13.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 123 |
+
"transformer.encoder.layers.13.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 124 |
+
"transformer.encoder.layers.13.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 125 |
+
"transformer.encoder.layers.13.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 126 |
+
"transformer.encoder.layers.13.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 127 |
+
"transformer.encoder.layers.13.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 128 |
+
"transformer.encoder.layers.13.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 129 |
+
"transformer.encoder.layers.13.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 130 |
+
"transformer.encoder.layers.13.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 131 |
+
"transformer.encoder.layers.13.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 132 |
+
"transformer.encoder.layers.13.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 133 |
+
"transformer.encoder.layers.13.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 134 |
+
"transformer.encoder.layers.13.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 135 |
+
"transformer.encoder.layers.13.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 136 |
+
"transformer.encoder.layers.13.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 137 |
+
"transformer.encoder.layers.13.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 138 |
+
"transformer.encoder.layers.13.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 139 |
+
"transformer.encoder.layers.13.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 140 |
+
"transformer.encoder.layers.14.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 141 |
+
"transformer.encoder.layers.14.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 142 |
+
"transformer.encoder.layers.14.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 143 |
+
"transformer.encoder.layers.14.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 144 |
+
"transformer.encoder.layers.14.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 145 |
+
"transformer.encoder.layers.14.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 146 |
+
"transformer.encoder.layers.14.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 147 |
+
"transformer.encoder.layers.14.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 148 |
+
"transformer.encoder.layers.14.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 149 |
+
"transformer.encoder.layers.14.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 150 |
+
"transformer.encoder.layers.14.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 151 |
+
"transformer.encoder.layers.14.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 152 |
+
"transformer.encoder.layers.14.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 153 |
+
"transformer.encoder.layers.14.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 154 |
+
"transformer.encoder.layers.14.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 155 |
+
"transformer.encoder.layers.14.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 156 |
+
"transformer.encoder.layers.14.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 157 |
+
"transformer.encoder.layers.14.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 158 |
+
"transformer.encoder.layers.14.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 159 |
+
"transformer.encoder.layers.14.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 160 |
+
"transformer.encoder.layers.14.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 161 |
+
"transformer.encoder.layers.14.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 162 |
+
"transformer.encoder.layers.15.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 163 |
+
"transformer.encoder.layers.15.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 164 |
+
"transformer.encoder.layers.15.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 165 |
+
"transformer.encoder.layers.15.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 166 |
+
"transformer.encoder.layers.15.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 167 |
+
"transformer.encoder.layers.15.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 168 |
+
"transformer.encoder.layers.15.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 169 |
+
"transformer.encoder.layers.15.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 170 |
+
"transformer.encoder.layers.15.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 171 |
+
"transformer.encoder.layers.15.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 172 |
+
"transformer.encoder.layers.15.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 173 |
+
"transformer.encoder.layers.15.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 174 |
+
"transformer.encoder.layers.15.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 175 |
+
"transformer.encoder.layers.15.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 176 |
+
"transformer.encoder.layers.15.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 177 |
+
"transformer.encoder.layers.15.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 178 |
+
"transformer.encoder.layers.15.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 179 |
+
"transformer.encoder.layers.15.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 180 |
+
"transformer.encoder.layers.15.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 181 |
+
"transformer.encoder.layers.15.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 182 |
+
"transformer.encoder.layers.15.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 183 |
+
"transformer.encoder.layers.15.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 184 |
+
"transformer.encoder.layers.16.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 185 |
+
"transformer.encoder.layers.16.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 186 |
+
"transformer.encoder.layers.16.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 187 |
+
"transformer.encoder.layers.16.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 188 |
+
"transformer.encoder.layers.16.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 189 |
+
"transformer.encoder.layers.16.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 190 |
+
"transformer.encoder.layers.16.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 191 |
+
"transformer.encoder.layers.16.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 192 |
+
"transformer.encoder.layers.16.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 193 |
+
"transformer.encoder.layers.16.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 194 |
+
"transformer.encoder.layers.16.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 195 |
+
"transformer.encoder.layers.16.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 196 |
+
"transformer.encoder.layers.16.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 197 |
+
"transformer.encoder.layers.16.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 198 |
+
"transformer.encoder.layers.16.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 199 |
+
"transformer.encoder.layers.16.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 200 |
+
"transformer.encoder.layers.16.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 201 |
+
"transformer.encoder.layers.16.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 202 |
+
"transformer.encoder.layers.16.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 203 |
+
"transformer.encoder.layers.16.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 204 |
+
"transformer.encoder.layers.16.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 205 |
+
"transformer.encoder.layers.16.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 206 |
+
"transformer.encoder.layers.17.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 207 |
+
"transformer.encoder.layers.17.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 208 |
+
"transformer.encoder.layers.17.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 209 |
+
"transformer.encoder.layers.17.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 210 |
+
"transformer.encoder.layers.17.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 211 |
+
"transformer.encoder.layers.17.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 212 |
+
"transformer.encoder.layers.17.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 213 |
+
"transformer.encoder.layers.17.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 214 |
+
"transformer.encoder.layers.17.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 215 |
+
"transformer.encoder.layers.17.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 216 |
+
"transformer.encoder.layers.17.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 217 |
+
"transformer.encoder.layers.17.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 218 |
+
"transformer.encoder.layers.17.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 219 |
+
"transformer.encoder.layers.17.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 220 |
+
"transformer.encoder.layers.17.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 221 |
+
"transformer.encoder.layers.17.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 222 |
+
"transformer.encoder.layers.17.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 223 |
+
"transformer.encoder.layers.17.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 224 |
+
"transformer.encoder.layers.17.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 225 |
+
"transformer.encoder.layers.17.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 226 |
+
"transformer.encoder.layers.17.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 227 |
+
"transformer.encoder.layers.17.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 228 |
+
"transformer.encoder.layers.18.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 229 |
+
"transformer.encoder.layers.18.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 230 |
+
"transformer.encoder.layers.18.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 231 |
+
"transformer.encoder.layers.18.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 232 |
+
"transformer.encoder.layers.18.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 233 |
+
"transformer.encoder.layers.18.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 234 |
+
"transformer.encoder.layers.18.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 235 |
+
"transformer.encoder.layers.18.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 236 |
+
"transformer.encoder.layers.18.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 237 |
+
"transformer.encoder.layers.18.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 238 |
+
"transformer.encoder.layers.18.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 239 |
+
"transformer.encoder.layers.18.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 240 |
+
"transformer.encoder.layers.18.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 241 |
+
"transformer.encoder.layers.18.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 242 |
+
"transformer.encoder.layers.18.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 243 |
+
"transformer.encoder.layers.18.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 244 |
+
"transformer.encoder.layers.18.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 245 |
+
"transformer.encoder.layers.18.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 246 |
+
"transformer.encoder.layers.18.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 247 |
+
"transformer.encoder.layers.18.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 248 |
+
"transformer.encoder.layers.18.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 249 |
+
"transformer.encoder.layers.18.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 250 |
+
"transformer.encoder.layers.19.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 251 |
+
"transformer.encoder.layers.19.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 252 |
+
"transformer.encoder.layers.19.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 253 |
+
"transformer.encoder.layers.19.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 254 |
+
"transformer.encoder.layers.19.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 255 |
+
"transformer.encoder.layers.19.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 256 |
+
"transformer.encoder.layers.19.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 257 |
+
"transformer.encoder.layers.19.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 258 |
+
"transformer.encoder.layers.19.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 259 |
+
"transformer.encoder.layers.19.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 260 |
+
"transformer.encoder.layers.19.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 261 |
+
"transformer.encoder.layers.19.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 262 |
+
"transformer.encoder.layers.19.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 263 |
+
"transformer.encoder.layers.19.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 264 |
+
"transformer.encoder.layers.19.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 265 |
+
"transformer.encoder.layers.19.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 266 |
+
"transformer.encoder.layers.19.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 267 |
+
"transformer.encoder.layers.19.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 268 |
+
"transformer.encoder.layers.19.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 269 |
+
"transformer.encoder.layers.19.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 270 |
+
"transformer.encoder.layers.19.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 271 |
+
"transformer.encoder.layers.19.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 272 |
+
"transformer.encoder.layers.2.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 273 |
+
"transformer.encoder.layers.2.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 274 |
+
"transformer.encoder.layers.2.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 275 |
+
"transformer.encoder.layers.2.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 276 |
+
"transformer.encoder.layers.2.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 277 |
+
"transformer.encoder.layers.2.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 278 |
+
"transformer.encoder.layers.2.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 279 |
+
"transformer.encoder.layers.2.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 280 |
+
"transformer.encoder.layers.2.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 281 |
+
"transformer.encoder.layers.2.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 282 |
+
"transformer.encoder.layers.2.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 283 |
+
"transformer.encoder.layers.2.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 284 |
+
"transformer.encoder.layers.2.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 285 |
+
"transformer.encoder.layers.2.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 286 |
+
"transformer.encoder.layers.2.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 287 |
+
"transformer.encoder.layers.2.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 288 |
+
"transformer.encoder.layers.2.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 289 |
+
"transformer.encoder.layers.2.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 290 |
+
"transformer.encoder.layers.2.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 291 |
+
"transformer.encoder.layers.2.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 292 |
+
"transformer.encoder.layers.2.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 293 |
+
"transformer.encoder.layers.2.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 294 |
+
"transformer.encoder.layers.20.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 295 |
+
"transformer.encoder.layers.20.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 296 |
+
"transformer.encoder.layers.20.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 297 |
+
"transformer.encoder.layers.20.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 298 |
+
"transformer.encoder.layers.20.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 299 |
+
"transformer.encoder.layers.20.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 300 |
+
"transformer.encoder.layers.20.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 301 |
+
"transformer.encoder.layers.20.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 302 |
+
"transformer.encoder.layers.20.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 303 |
+
"transformer.encoder.layers.20.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 304 |
+
"transformer.encoder.layers.20.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 305 |
+
"transformer.encoder.layers.20.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 306 |
+
"transformer.encoder.layers.20.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 307 |
+
"transformer.encoder.layers.20.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 308 |
+
"transformer.encoder.layers.20.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 309 |
+
"transformer.encoder.layers.20.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 310 |
+
"transformer.encoder.layers.20.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 311 |
+
"transformer.encoder.layers.20.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 312 |
+
"transformer.encoder.layers.20.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 313 |
+
"transformer.encoder.layers.20.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 314 |
+
"transformer.encoder.layers.20.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 315 |
+
"transformer.encoder.layers.20.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 316 |
+
"transformer.encoder.layers.21.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 317 |
+
"transformer.encoder.layers.21.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 318 |
+
"transformer.encoder.layers.21.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 319 |
+
"transformer.encoder.layers.21.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 320 |
+
"transformer.encoder.layers.21.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 321 |
+
"transformer.encoder.layers.21.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 322 |
+
"transformer.encoder.layers.21.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 323 |
+
"transformer.encoder.layers.21.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 324 |
+
"transformer.encoder.layers.21.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 325 |
+
"transformer.encoder.layers.21.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 326 |
+
"transformer.encoder.layers.21.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 327 |
+
"transformer.encoder.layers.21.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 328 |
+
"transformer.encoder.layers.21.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 329 |
+
"transformer.encoder.layers.21.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 330 |
+
"transformer.encoder.layers.21.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 331 |
+
"transformer.encoder.layers.21.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 332 |
+
"transformer.encoder.layers.21.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 333 |
+
"transformer.encoder.layers.21.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 334 |
+
"transformer.encoder.layers.21.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 335 |
+
"transformer.encoder.layers.21.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 336 |
+
"transformer.encoder.layers.21.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 337 |
+
"transformer.encoder.layers.21.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 338 |
+
"transformer.encoder.layers.22.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 339 |
+
"transformer.encoder.layers.22.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 340 |
+
"transformer.encoder.layers.22.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 341 |
+
"transformer.encoder.layers.22.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 342 |
+
"transformer.encoder.layers.22.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 343 |
+
"transformer.encoder.layers.22.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 344 |
+
"transformer.encoder.layers.22.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 345 |
+
"transformer.encoder.layers.22.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 346 |
+
"transformer.encoder.layers.22.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 347 |
+
"transformer.encoder.layers.22.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 348 |
+
"transformer.encoder.layers.22.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 349 |
+
"transformer.encoder.layers.22.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 350 |
+
"transformer.encoder.layers.22.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 351 |
+
"transformer.encoder.layers.22.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 352 |
+
"transformer.encoder.layers.22.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 353 |
+
"transformer.encoder.layers.22.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 354 |
+
"transformer.encoder.layers.22.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 355 |
+
"transformer.encoder.layers.22.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 356 |
+
"transformer.encoder.layers.22.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 357 |
+
"transformer.encoder.layers.22.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 358 |
+
"transformer.encoder.layers.22.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 359 |
+
"transformer.encoder.layers.22.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 360 |
+
"transformer.encoder.layers.23.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 361 |
+
"transformer.encoder.layers.23.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 362 |
+
"transformer.encoder.layers.23.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 363 |
+
"transformer.encoder.layers.23.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 364 |
+
"transformer.encoder.layers.23.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 365 |
+
"transformer.encoder.layers.23.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 366 |
+
"transformer.encoder.layers.23.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 367 |
+
"transformer.encoder.layers.23.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 368 |
+
"transformer.encoder.layers.23.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 369 |
+
"transformer.encoder.layers.23.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 370 |
+
"transformer.encoder.layers.23.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 371 |
+
"transformer.encoder.layers.23.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 372 |
+
"transformer.encoder.layers.23.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 373 |
+
"transformer.encoder.layers.23.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 374 |
+
"transformer.encoder.layers.23.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 375 |
+
"transformer.encoder.layers.23.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 376 |
+
"transformer.encoder.layers.23.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 377 |
+
"transformer.encoder.layers.23.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 378 |
+
"transformer.encoder.layers.23.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 379 |
+
"transformer.encoder.layers.23.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 380 |
+
"transformer.encoder.layers.23.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 381 |
+
"transformer.encoder.layers.23.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 382 |
+
"transformer.encoder.layers.24.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 383 |
+
"transformer.encoder.layers.24.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 384 |
+
"transformer.encoder.layers.24.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 385 |
+
"transformer.encoder.layers.24.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 386 |
+
"transformer.encoder.layers.24.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 387 |
+
"transformer.encoder.layers.24.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 388 |
+
"transformer.encoder.layers.24.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 389 |
+
"transformer.encoder.layers.24.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 390 |
+
"transformer.encoder.layers.24.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 391 |
+
"transformer.encoder.layers.24.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 392 |
+
"transformer.encoder.layers.24.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 393 |
+
"transformer.encoder.layers.24.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 394 |
+
"transformer.encoder.layers.24.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 395 |
+
"transformer.encoder.layers.24.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 396 |
+
"transformer.encoder.layers.24.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 397 |
+
"transformer.encoder.layers.24.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 398 |
+
"transformer.encoder.layers.24.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 399 |
+
"transformer.encoder.layers.24.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 400 |
+
"transformer.encoder.layers.24.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 401 |
+
"transformer.encoder.layers.24.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 402 |
+
"transformer.encoder.layers.24.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 403 |
+
"transformer.encoder.layers.24.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 404 |
+
"transformer.encoder.layers.25.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 405 |
+
"transformer.encoder.layers.25.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 406 |
+
"transformer.encoder.layers.25.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 407 |
+
"transformer.encoder.layers.25.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 408 |
+
"transformer.encoder.layers.25.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 409 |
+
"transformer.encoder.layers.25.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 410 |
+
"transformer.encoder.layers.25.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 411 |
+
"transformer.encoder.layers.25.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 412 |
+
"transformer.encoder.layers.25.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 413 |
+
"transformer.encoder.layers.25.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 414 |
+
"transformer.encoder.layers.25.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 415 |
+
"transformer.encoder.layers.25.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 416 |
+
"transformer.encoder.layers.25.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 417 |
+
"transformer.encoder.layers.25.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 418 |
+
"transformer.encoder.layers.25.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 419 |
+
"transformer.encoder.layers.25.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 420 |
+
"transformer.encoder.layers.25.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 421 |
+
"transformer.encoder.layers.25.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 422 |
+
"transformer.encoder.layers.25.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 423 |
+
"transformer.encoder.layers.25.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 424 |
+
"transformer.encoder.layers.25.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 425 |
+
"transformer.encoder.layers.25.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 426 |
+
"transformer.encoder.layers.26.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 427 |
+
"transformer.encoder.layers.26.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 428 |
+
"transformer.encoder.layers.26.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 429 |
+
"transformer.encoder.layers.26.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 430 |
+
"transformer.encoder.layers.26.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 431 |
+
"transformer.encoder.layers.26.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 432 |
+
"transformer.encoder.layers.26.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 433 |
+
"transformer.encoder.layers.26.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 434 |
+
"transformer.encoder.layers.26.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 435 |
+
"transformer.encoder.layers.26.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 436 |
+
"transformer.encoder.layers.26.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 437 |
+
"transformer.encoder.layers.26.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 438 |
+
"transformer.encoder.layers.26.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 439 |
+
"transformer.encoder.layers.26.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 440 |
+
"transformer.encoder.layers.26.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 441 |
+
"transformer.encoder.layers.26.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 442 |
+
"transformer.encoder.layers.26.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 443 |
+
"transformer.encoder.layers.26.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 444 |
+
"transformer.encoder.layers.26.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 445 |
+
"transformer.encoder.layers.26.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 446 |
+
"transformer.encoder.layers.26.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 447 |
+
"transformer.encoder.layers.26.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 448 |
+
"transformer.encoder.layers.27.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 449 |
+
"transformer.encoder.layers.27.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 450 |
+
"transformer.encoder.layers.27.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 451 |
+
"transformer.encoder.layers.27.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 452 |
+
"transformer.encoder.layers.27.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 453 |
+
"transformer.encoder.layers.27.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 454 |
+
"transformer.encoder.layers.27.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 455 |
+
"transformer.encoder.layers.27.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 456 |
+
"transformer.encoder.layers.27.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 457 |
+
"transformer.encoder.layers.27.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 458 |
+
"transformer.encoder.layers.27.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 459 |
+
"transformer.encoder.layers.27.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 460 |
+
"transformer.encoder.layers.27.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 461 |
+
"transformer.encoder.layers.27.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 462 |
+
"transformer.encoder.layers.27.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 463 |
+
"transformer.encoder.layers.27.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 464 |
+
"transformer.encoder.layers.27.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 465 |
+
"transformer.encoder.layers.27.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 466 |
+
"transformer.encoder.layers.27.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 467 |
+
"transformer.encoder.layers.27.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 468 |
+
"transformer.encoder.layers.27.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 469 |
+
"transformer.encoder.layers.27.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 470 |
+
"transformer.encoder.layers.28.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 471 |
+
"transformer.encoder.layers.28.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 472 |
+
"transformer.encoder.layers.28.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 473 |
+
"transformer.encoder.layers.28.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 474 |
+
"transformer.encoder.layers.28.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 475 |
+
"transformer.encoder.layers.28.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 476 |
+
"transformer.encoder.layers.28.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 477 |
+
"transformer.encoder.layers.28.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 478 |
+
"transformer.encoder.layers.28.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 479 |
+
"transformer.encoder.layers.28.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 480 |
+
"transformer.encoder.layers.28.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 481 |
+
"transformer.encoder.layers.28.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 482 |
+
"transformer.encoder.layers.28.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 483 |
+
"transformer.encoder.layers.28.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 484 |
+
"transformer.encoder.layers.28.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 485 |
+
"transformer.encoder.layers.28.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 486 |
+
"transformer.encoder.layers.28.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 487 |
+
"transformer.encoder.layers.28.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 488 |
+
"transformer.encoder.layers.28.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 489 |
+
"transformer.encoder.layers.28.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 490 |
+
"transformer.encoder.layers.28.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 491 |
+
"transformer.encoder.layers.28.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 492 |
+
"transformer.encoder.layers.29.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 493 |
+
"transformer.encoder.layers.29.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 494 |
+
"transformer.encoder.layers.29.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 495 |
+
"transformer.encoder.layers.29.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 496 |
+
"transformer.encoder.layers.29.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 497 |
+
"transformer.encoder.layers.29.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 498 |
+
"transformer.encoder.layers.29.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 499 |
+
"transformer.encoder.layers.29.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 500 |
+
"transformer.encoder.layers.29.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 501 |
+
"transformer.encoder.layers.29.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 502 |
+
"transformer.encoder.layers.29.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 503 |
+
"transformer.encoder.layers.29.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 504 |
+
"transformer.encoder.layers.29.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 505 |
+
"transformer.encoder.layers.29.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 506 |
+
"transformer.encoder.layers.29.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 507 |
+
"transformer.encoder.layers.29.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 508 |
+
"transformer.encoder.layers.29.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 509 |
+
"transformer.encoder.layers.29.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 510 |
+
"transformer.encoder.layers.29.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 511 |
+
"transformer.encoder.layers.29.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 512 |
+
"transformer.encoder.layers.29.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 513 |
+
"transformer.encoder.layers.29.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 514 |
+
"transformer.encoder.layers.3.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 515 |
+
"transformer.encoder.layers.3.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 516 |
+
"transformer.encoder.layers.3.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 517 |
+
"transformer.encoder.layers.3.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 518 |
+
"transformer.encoder.layers.3.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 519 |
+
"transformer.encoder.layers.3.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 520 |
+
"transformer.encoder.layers.3.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 521 |
+
"transformer.encoder.layers.3.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 522 |
+
"transformer.encoder.layers.3.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 523 |
+
"transformer.encoder.layers.3.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 524 |
+
"transformer.encoder.layers.3.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 525 |
+
"transformer.encoder.layers.3.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 526 |
+
"transformer.encoder.layers.3.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 527 |
+
"transformer.encoder.layers.3.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 528 |
+
"transformer.encoder.layers.3.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 529 |
+
"transformer.encoder.layers.3.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 530 |
+
"transformer.encoder.layers.3.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 531 |
+
"transformer.encoder.layers.3.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 532 |
+
"transformer.encoder.layers.3.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 533 |
+
"transformer.encoder.layers.3.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 534 |
+
"transformer.encoder.layers.3.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 535 |
+
"transformer.encoder.layers.3.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 536 |
+
"transformer.encoder.layers.30.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 537 |
+
"transformer.encoder.layers.30.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 538 |
+
"transformer.encoder.layers.30.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 539 |
+
"transformer.encoder.layers.30.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 540 |
+
"transformer.encoder.layers.30.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 541 |
+
"transformer.encoder.layers.30.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 542 |
+
"transformer.encoder.layers.30.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 543 |
+
"transformer.encoder.layers.30.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 544 |
+
"transformer.encoder.layers.30.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 545 |
+
"transformer.encoder.layers.30.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 546 |
+
"transformer.encoder.layers.30.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 547 |
+
"transformer.encoder.layers.30.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 548 |
+
"transformer.encoder.layers.30.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 549 |
+
"transformer.encoder.layers.30.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 550 |
+
"transformer.encoder.layers.30.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 551 |
+
"transformer.encoder.layers.30.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 552 |
+
"transformer.encoder.layers.30.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 553 |
+
"transformer.encoder.layers.30.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 554 |
+
"transformer.encoder.layers.30.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 555 |
+
"transformer.encoder.layers.30.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 556 |
+
"transformer.encoder.layers.30.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 557 |
+
"transformer.encoder.layers.30.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 558 |
+
"transformer.encoder.layers.31.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 559 |
+
"transformer.encoder.layers.31.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 560 |
+
"transformer.encoder.layers.31.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 561 |
+
"transformer.encoder.layers.31.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 562 |
+
"transformer.encoder.layers.31.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 563 |
+
"transformer.encoder.layers.31.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 564 |
+
"transformer.encoder.layers.31.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 565 |
+
"transformer.encoder.layers.31.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 566 |
+
"transformer.encoder.layers.31.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 567 |
+
"transformer.encoder.layers.31.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 568 |
+
"transformer.encoder.layers.31.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 569 |
+
"transformer.encoder.layers.31.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 570 |
+
"transformer.encoder.layers.31.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 571 |
+
"transformer.encoder.layers.31.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 572 |
+
"transformer.encoder.layers.31.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 573 |
+
"transformer.encoder.layers.31.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 574 |
+
"transformer.encoder.layers.31.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 575 |
+
"transformer.encoder.layers.31.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 576 |
+
"transformer.encoder.layers.31.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 577 |
+
"transformer.encoder.layers.31.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 578 |
+
"transformer.encoder.layers.31.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 579 |
+
"transformer.encoder.layers.31.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 580 |
+
"transformer.encoder.layers.32.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 581 |
+
"transformer.encoder.layers.32.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 582 |
+
"transformer.encoder.layers.32.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 583 |
+
"transformer.encoder.layers.32.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 584 |
+
"transformer.encoder.layers.32.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 585 |
+
"transformer.encoder.layers.32.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 586 |
+
"transformer.encoder.layers.32.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 587 |
+
"transformer.encoder.layers.32.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 588 |
+
"transformer.encoder.layers.32.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 589 |
+
"transformer.encoder.layers.32.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 590 |
+
"transformer.encoder.layers.32.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 591 |
+
"transformer.encoder.layers.32.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 592 |
+
"transformer.encoder.layers.32.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 593 |
+
"transformer.encoder.layers.32.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 594 |
+
"transformer.encoder.layers.32.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 595 |
+
"transformer.encoder.layers.32.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 596 |
+
"transformer.encoder.layers.32.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 597 |
+
"transformer.encoder.layers.32.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 598 |
+
"transformer.encoder.layers.32.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 599 |
+
"transformer.encoder.layers.32.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 600 |
+
"transformer.encoder.layers.32.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 601 |
+
"transformer.encoder.layers.32.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 602 |
+
"transformer.encoder.layers.33.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 603 |
+
"transformer.encoder.layers.33.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 604 |
+
"transformer.encoder.layers.33.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 605 |
+
"transformer.encoder.layers.33.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 606 |
+
"transformer.encoder.layers.33.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 607 |
+
"transformer.encoder.layers.33.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 608 |
+
"transformer.encoder.layers.33.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 609 |
+
"transformer.encoder.layers.33.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 610 |
+
"transformer.encoder.layers.33.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 611 |
+
"transformer.encoder.layers.33.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 612 |
+
"transformer.encoder.layers.33.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 613 |
+
"transformer.encoder.layers.33.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 614 |
+
"transformer.encoder.layers.33.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 615 |
+
"transformer.encoder.layers.33.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 616 |
+
"transformer.encoder.layers.33.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 617 |
+
"transformer.encoder.layers.33.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 618 |
+
"transformer.encoder.layers.33.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 619 |
+
"transformer.encoder.layers.33.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 620 |
+
"transformer.encoder.layers.33.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 621 |
+
"transformer.encoder.layers.33.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 622 |
+
"transformer.encoder.layers.33.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 623 |
+
"transformer.encoder.layers.33.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 624 |
+
"transformer.encoder.layers.34.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 625 |
+
"transformer.encoder.layers.34.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 626 |
+
"transformer.encoder.layers.34.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 627 |
+
"transformer.encoder.layers.34.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 628 |
+
"transformer.encoder.layers.34.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 629 |
+
"transformer.encoder.layers.34.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 630 |
+
"transformer.encoder.layers.34.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 631 |
+
"transformer.encoder.layers.34.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 632 |
+
"transformer.encoder.layers.34.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 633 |
+
"transformer.encoder.layers.34.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 634 |
+
"transformer.encoder.layers.34.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 635 |
+
"transformer.encoder.layers.34.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 636 |
+
"transformer.encoder.layers.34.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 637 |
+
"transformer.encoder.layers.34.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 638 |
+
"transformer.encoder.layers.34.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 639 |
+
"transformer.encoder.layers.34.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 640 |
+
"transformer.encoder.layers.34.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 641 |
+
"transformer.encoder.layers.34.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 642 |
+
"transformer.encoder.layers.34.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 643 |
+
"transformer.encoder.layers.34.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 644 |
+
"transformer.encoder.layers.34.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 645 |
+
"transformer.encoder.layers.34.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 646 |
+
"transformer.encoder.layers.35.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 647 |
+
"transformer.encoder.layers.35.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 648 |
+
"transformer.encoder.layers.35.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 649 |
+
"transformer.encoder.layers.35.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 650 |
+
"transformer.encoder.layers.35.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 651 |
+
"transformer.encoder.layers.35.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 652 |
+
"transformer.encoder.layers.35.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 653 |
+
"transformer.encoder.layers.35.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 654 |
+
"transformer.encoder.layers.35.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 655 |
+
"transformer.encoder.layers.35.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 656 |
+
"transformer.encoder.layers.35.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 657 |
+
"transformer.encoder.layers.35.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 658 |
+
"transformer.encoder.layers.35.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 659 |
+
"transformer.encoder.layers.35.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 660 |
+
"transformer.encoder.layers.35.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 661 |
+
"transformer.encoder.layers.35.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 662 |
+
"transformer.encoder.layers.35.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 663 |
+
"transformer.encoder.layers.35.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 664 |
+
"transformer.encoder.layers.35.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 665 |
+
"transformer.encoder.layers.35.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 666 |
+
"transformer.encoder.layers.35.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 667 |
+
"transformer.encoder.layers.35.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 668 |
+
"transformer.encoder.layers.36.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 669 |
+
"transformer.encoder.layers.36.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 670 |
+
"transformer.encoder.layers.36.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 671 |
+
"transformer.encoder.layers.36.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 672 |
+
"transformer.encoder.layers.36.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 673 |
+
"transformer.encoder.layers.36.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 674 |
+
"transformer.encoder.layers.36.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 675 |
+
"transformer.encoder.layers.36.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 676 |
+
"transformer.encoder.layers.36.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 677 |
+
"transformer.encoder.layers.36.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 678 |
+
"transformer.encoder.layers.36.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 679 |
+
"transformer.encoder.layers.36.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 680 |
+
"transformer.encoder.layers.36.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 681 |
+
"transformer.encoder.layers.36.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 682 |
+
"transformer.encoder.layers.36.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 683 |
+
"transformer.encoder.layers.36.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 684 |
+
"transformer.encoder.layers.36.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 685 |
+
"transformer.encoder.layers.36.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 686 |
+
"transformer.encoder.layers.36.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 687 |
+
"transformer.encoder.layers.36.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 688 |
+
"transformer.encoder.layers.36.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 689 |
+
"transformer.encoder.layers.36.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 690 |
+
"transformer.encoder.layers.37.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 691 |
+
"transformer.encoder.layers.37.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 692 |
+
"transformer.encoder.layers.37.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 693 |
+
"transformer.encoder.layers.37.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 694 |
+
"transformer.encoder.layers.37.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 695 |
+
"transformer.encoder.layers.37.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 696 |
+
"transformer.encoder.layers.37.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 697 |
+
"transformer.encoder.layers.37.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 698 |
+
"transformer.encoder.layers.37.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 699 |
+
"transformer.encoder.layers.37.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 700 |
+
"transformer.encoder.layers.37.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 701 |
+
"transformer.encoder.layers.37.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 702 |
+
"transformer.encoder.layers.37.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 703 |
+
"transformer.encoder.layers.37.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 704 |
+
"transformer.encoder.layers.37.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 705 |
+
"transformer.encoder.layers.37.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 706 |
+
"transformer.encoder.layers.37.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 707 |
+
"transformer.encoder.layers.37.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 708 |
+
"transformer.encoder.layers.37.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 709 |
+
"transformer.encoder.layers.37.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 710 |
+
"transformer.encoder.layers.37.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 711 |
+
"transformer.encoder.layers.37.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 712 |
+
"transformer.encoder.layers.38.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 713 |
+
"transformer.encoder.layers.38.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 714 |
+
"transformer.encoder.layers.38.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 715 |
+
"transformer.encoder.layers.38.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 716 |
+
"transformer.encoder.layers.38.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 717 |
+
"transformer.encoder.layers.38.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 718 |
+
"transformer.encoder.layers.38.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 719 |
+
"transformer.encoder.layers.38.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 720 |
+
"transformer.encoder.layers.38.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 721 |
+
"transformer.encoder.layers.38.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 722 |
+
"transformer.encoder.layers.38.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 723 |
+
"transformer.encoder.layers.38.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 724 |
+
"transformer.encoder.layers.38.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 725 |
+
"transformer.encoder.layers.38.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 726 |
+
"transformer.encoder.layers.38.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 727 |
+
"transformer.encoder.layers.38.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 728 |
+
"transformer.encoder.layers.38.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 729 |
+
"transformer.encoder.layers.38.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 730 |
+
"transformer.encoder.layers.38.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 731 |
+
"transformer.encoder.layers.38.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 732 |
+
"transformer.encoder.layers.38.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 733 |
+
"transformer.encoder.layers.38.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 734 |
+
"transformer.encoder.layers.39.input_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 735 |
+
"transformer.encoder.layers.39.mlp.dense_4h_to_h.bias": "model-00002-of-00003.safetensors",
|
| 736 |
+
"transformer.encoder.layers.39.mlp.dense_4h_to_h.g_idx": "model-00002-of-00003.safetensors",
|
| 737 |
+
"transformer.encoder.layers.39.mlp.dense_4h_to_h.qweight": "model-00002-of-00003.safetensors",
|
| 738 |
+
"transformer.encoder.layers.39.mlp.dense_4h_to_h.qzeros": "model-00002-of-00003.safetensors",
|
| 739 |
+
"transformer.encoder.layers.39.mlp.dense_4h_to_h.scales": "model-00002-of-00003.safetensors",
|
| 740 |
+
"transformer.encoder.layers.39.mlp.dense_h_to_4h.bias": "model-00002-of-00003.safetensors",
|
| 741 |
+
"transformer.encoder.layers.39.mlp.dense_h_to_4h.g_idx": "model-00002-of-00003.safetensors",
|
| 742 |
+
"transformer.encoder.layers.39.mlp.dense_h_to_4h.qweight": "model-00002-of-00003.safetensors",
|
| 743 |
+
"transformer.encoder.layers.39.mlp.dense_h_to_4h.qzeros": "model-00002-of-00003.safetensors",
|
| 744 |
+
"transformer.encoder.layers.39.mlp.dense_h_to_4h.scales": "model-00002-of-00003.safetensors",
|
| 745 |
+
"transformer.encoder.layers.39.post_attention_layernorm.weight": "model-00002-of-00003.safetensors",
|
| 746 |
+
"transformer.encoder.layers.39.self_attention.dense.bias": "model-00002-of-00003.safetensors",
|
| 747 |
+
"transformer.encoder.layers.39.self_attention.dense.g_idx": "model-00002-of-00003.safetensors",
|
| 748 |
+
"transformer.encoder.layers.39.self_attention.dense.qweight": "model-00002-of-00003.safetensors",
|
| 749 |
+
"transformer.encoder.layers.39.self_attention.dense.qzeros": "model-00002-of-00003.safetensors",
|
| 750 |
+
"transformer.encoder.layers.39.self_attention.dense.scales": "model-00002-of-00003.safetensors",
|
| 751 |
+
"transformer.encoder.layers.39.self_attention.query_key_value.bias": "model-00002-of-00003.safetensors",
|
| 752 |
+
"transformer.encoder.layers.39.self_attention.query_key_value.g_idx": "model-00002-of-00003.safetensors",
|
| 753 |
+
"transformer.encoder.layers.39.self_attention.query_key_value.qweight": "model-00002-of-00003.safetensors",
|
| 754 |
+
"transformer.encoder.layers.39.self_attention.query_key_value.qzeros": "model-00002-of-00003.safetensors",
|
| 755 |
+
"transformer.encoder.layers.39.self_attention.query_key_value.scales": "model-00002-of-00003.safetensors",
|
| 756 |
+
"transformer.encoder.layers.4.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 757 |
+
"transformer.encoder.layers.4.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 758 |
+
"transformer.encoder.layers.4.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 759 |
+
"transformer.encoder.layers.4.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 760 |
+
"transformer.encoder.layers.4.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 761 |
+
"transformer.encoder.layers.4.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 762 |
+
"transformer.encoder.layers.4.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 763 |
+
"transformer.encoder.layers.4.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 764 |
+
"transformer.encoder.layers.4.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 765 |
+
"transformer.encoder.layers.4.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 766 |
+
"transformer.encoder.layers.4.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 767 |
+
"transformer.encoder.layers.4.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 768 |
+
"transformer.encoder.layers.4.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 769 |
+
"transformer.encoder.layers.4.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 770 |
+
"transformer.encoder.layers.4.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 771 |
+
"transformer.encoder.layers.4.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 772 |
+
"transformer.encoder.layers.4.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 773 |
+
"transformer.encoder.layers.4.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 774 |
+
"transformer.encoder.layers.4.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 775 |
+
"transformer.encoder.layers.4.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 776 |
+
"transformer.encoder.layers.4.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 777 |
+
"transformer.encoder.layers.4.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 778 |
+
"transformer.encoder.layers.5.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 779 |
+
"transformer.encoder.layers.5.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 780 |
+
"transformer.encoder.layers.5.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 781 |
+
"transformer.encoder.layers.5.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 782 |
+
"transformer.encoder.layers.5.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 783 |
+
"transformer.encoder.layers.5.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 784 |
+
"transformer.encoder.layers.5.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 785 |
+
"transformer.encoder.layers.5.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 786 |
+
"transformer.encoder.layers.5.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 787 |
+
"transformer.encoder.layers.5.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 788 |
+
"transformer.encoder.layers.5.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 789 |
+
"transformer.encoder.layers.5.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 790 |
+
"transformer.encoder.layers.5.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 791 |
+
"transformer.encoder.layers.5.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 792 |
+
"transformer.encoder.layers.5.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 793 |
+
"transformer.encoder.layers.5.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 794 |
+
"transformer.encoder.layers.5.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 795 |
+
"transformer.encoder.layers.5.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 796 |
+
"transformer.encoder.layers.5.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 797 |
+
"transformer.encoder.layers.5.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 798 |
+
"transformer.encoder.layers.5.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 799 |
+
"transformer.encoder.layers.5.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 800 |
+
"transformer.encoder.layers.6.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 801 |
+
"transformer.encoder.layers.6.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 802 |
+
"transformer.encoder.layers.6.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 803 |
+
"transformer.encoder.layers.6.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 804 |
+
"transformer.encoder.layers.6.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 805 |
+
"transformer.encoder.layers.6.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 806 |
+
"transformer.encoder.layers.6.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 807 |
+
"transformer.encoder.layers.6.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 808 |
+
"transformer.encoder.layers.6.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 809 |
+
"transformer.encoder.layers.6.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 810 |
+
"transformer.encoder.layers.6.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 811 |
+
"transformer.encoder.layers.6.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 812 |
+
"transformer.encoder.layers.6.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 813 |
+
"transformer.encoder.layers.6.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 814 |
+
"transformer.encoder.layers.6.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 815 |
+
"transformer.encoder.layers.6.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 816 |
+
"transformer.encoder.layers.6.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 817 |
+
"transformer.encoder.layers.6.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 818 |
+
"transformer.encoder.layers.6.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 819 |
+
"transformer.encoder.layers.6.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 820 |
+
"transformer.encoder.layers.6.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 821 |
+
"transformer.encoder.layers.6.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 822 |
+
"transformer.encoder.layers.7.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 823 |
+
"transformer.encoder.layers.7.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 824 |
+
"transformer.encoder.layers.7.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 825 |
+
"transformer.encoder.layers.7.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 826 |
+
"transformer.encoder.layers.7.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 827 |
+
"transformer.encoder.layers.7.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 828 |
+
"transformer.encoder.layers.7.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 829 |
+
"transformer.encoder.layers.7.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 830 |
+
"transformer.encoder.layers.7.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 831 |
+
"transformer.encoder.layers.7.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 832 |
+
"transformer.encoder.layers.7.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 833 |
+
"transformer.encoder.layers.7.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 834 |
+
"transformer.encoder.layers.7.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 835 |
+
"transformer.encoder.layers.7.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 836 |
+
"transformer.encoder.layers.7.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 837 |
+
"transformer.encoder.layers.7.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 838 |
+
"transformer.encoder.layers.7.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 839 |
+
"transformer.encoder.layers.7.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 840 |
+
"transformer.encoder.layers.7.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 841 |
+
"transformer.encoder.layers.7.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 842 |
+
"transformer.encoder.layers.7.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 843 |
+
"transformer.encoder.layers.7.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 844 |
+
"transformer.encoder.layers.8.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 845 |
+
"transformer.encoder.layers.8.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 846 |
+
"transformer.encoder.layers.8.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 847 |
+
"transformer.encoder.layers.8.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 848 |
+
"transformer.encoder.layers.8.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 849 |
+
"transformer.encoder.layers.8.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 850 |
+
"transformer.encoder.layers.8.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 851 |
+
"transformer.encoder.layers.8.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 852 |
+
"transformer.encoder.layers.8.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 853 |
+
"transformer.encoder.layers.8.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 854 |
+
"transformer.encoder.layers.8.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 855 |
+
"transformer.encoder.layers.8.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 856 |
+
"transformer.encoder.layers.8.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 857 |
+
"transformer.encoder.layers.8.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 858 |
+
"transformer.encoder.layers.8.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 859 |
+
"transformer.encoder.layers.8.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 860 |
+
"transformer.encoder.layers.8.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 861 |
+
"transformer.encoder.layers.8.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 862 |
+
"transformer.encoder.layers.8.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 863 |
+
"transformer.encoder.layers.8.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 864 |
+
"transformer.encoder.layers.8.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 865 |
+
"transformer.encoder.layers.8.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 866 |
+
"transformer.encoder.layers.9.input_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 867 |
+
"transformer.encoder.layers.9.mlp.dense_4h_to_h.bias": "model-00001-of-00003.safetensors",
|
| 868 |
+
"transformer.encoder.layers.9.mlp.dense_4h_to_h.g_idx": "model-00001-of-00003.safetensors",
|
| 869 |
+
"transformer.encoder.layers.9.mlp.dense_4h_to_h.qweight": "model-00001-of-00003.safetensors",
|
| 870 |
+
"transformer.encoder.layers.9.mlp.dense_4h_to_h.qzeros": "model-00001-of-00003.safetensors",
|
| 871 |
+
"transformer.encoder.layers.9.mlp.dense_4h_to_h.scales": "model-00001-of-00003.safetensors",
|
| 872 |
+
"transformer.encoder.layers.9.mlp.dense_h_to_4h.bias": "model-00001-of-00003.safetensors",
|
| 873 |
+
"transformer.encoder.layers.9.mlp.dense_h_to_4h.g_idx": "model-00001-of-00003.safetensors",
|
| 874 |
+
"transformer.encoder.layers.9.mlp.dense_h_to_4h.qweight": "model-00001-of-00003.safetensors",
|
| 875 |
+
"transformer.encoder.layers.9.mlp.dense_h_to_4h.qzeros": "model-00001-of-00003.safetensors",
|
| 876 |
+
"transformer.encoder.layers.9.mlp.dense_h_to_4h.scales": "model-00001-of-00003.safetensors",
|
| 877 |
+
"transformer.encoder.layers.9.post_attention_layernorm.weight": "model-00001-of-00003.safetensors",
|
| 878 |
+
"transformer.encoder.layers.9.self_attention.dense.bias": "model-00001-of-00003.safetensors",
|
| 879 |
+
"transformer.encoder.layers.9.self_attention.dense.g_idx": "model-00001-of-00003.safetensors",
|
| 880 |
+
"transformer.encoder.layers.9.self_attention.dense.qweight": "model-00001-of-00003.safetensors",
|
| 881 |
+
"transformer.encoder.layers.9.self_attention.dense.qzeros": "model-00001-of-00003.safetensors",
|
| 882 |
+
"transformer.encoder.layers.9.self_attention.dense.scales": "model-00001-of-00003.safetensors",
|
| 883 |
+
"transformer.encoder.layers.9.self_attention.query_key_value.bias": "model-00001-of-00003.safetensors",
|
| 884 |
+
"transformer.encoder.layers.9.self_attention.query_key_value.g_idx": "model-00001-of-00003.safetensors",
|
| 885 |
+
"transformer.encoder.layers.9.self_attention.query_key_value.qweight": "model-00001-of-00003.safetensors",
|
| 886 |
+
"transformer.encoder.layers.9.self_attention.query_key_value.qzeros": "model-00001-of-00003.safetensors",
|
| 887 |
+
"transformer.encoder.layers.9.self_attention.query_key_value.scales": "model-00001-of-00003.safetensors",
|
| 888 |
+
"transformer.output_layer.weight": "model-00003-of-00003.safetensors",
|
| 889 |
+
"transformer.rotary_pos_emb.inv_freq": "model-00001-of-00003.safetensors"
|
| 890 |
+
}
|
| 891 |
+
}
|
modeling_chatglm.py
ADDED
|
@@ -0,0 +1,1218 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
""" PyTorch ChatGLM model. """
|
| 2 |
+
import json
|
| 3 |
+
import math
|
| 4 |
+
import copy
|
| 5 |
+
import warnings
|
| 6 |
+
import re
|
| 7 |
+
import sys
|
| 8 |
+
|
| 9 |
+
import torch
|
| 10 |
+
import torch.utils.checkpoint
|
| 11 |
+
import torch.nn.functional as F
|
| 12 |
+
from torch import nn
|
| 13 |
+
from torch.nn import CrossEntropyLoss, LayerNorm, MSELoss, BCEWithLogitsLoss
|
| 14 |
+
from torch.nn.utils import skip_init
|
| 15 |
+
from typing import Optional, Tuple, Union, List, Callable, Dict, Any
|
| 16 |
+
from copy import deepcopy
|
| 17 |
+
|
| 18 |
+
from transformers.modeling_outputs import (
|
| 19 |
+
BaseModelOutputWithPast,
|
| 20 |
+
CausalLMOutputWithPast,
|
| 21 |
+
SequenceClassifierOutputWithPast,
|
| 22 |
+
)
|
| 23 |
+
from transformers.modeling_utils import PreTrainedModel
|
| 24 |
+
from transformers.utils import logging
|
| 25 |
+
from transformers.generation.logits_process import LogitsProcessor
|
| 26 |
+
from transformers.generation.utils import LogitsProcessorList, StoppingCriteriaList, GenerationConfig, ModelOutput
|
| 27 |
+
|
| 28 |
+
from .configuration_chatglm import ChatGLMConfig
|
| 29 |
+
|
| 30 |
+
# flags required to enable jit fusion kernels
|
| 31 |
+
|
| 32 |
+
if sys.platform != 'darwin':
|
| 33 |
+
torch._C._jit_set_profiling_mode(False)
|
| 34 |
+
torch._C._jit_set_profiling_executor(False)
|
| 35 |
+
torch._C._jit_override_can_fuse_on_cpu(True)
|
| 36 |
+
torch._C._jit_override_can_fuse_on_gpu(True)
|
| 37 |
+
|
| 38 |
+
logger = logging.get_logger(__name__)
|
| 39 |
+
|
| 40 |
+
_CHECKPOINT_FOR_DOC = "THUDM/ChatGLM"
|
| 41 |
+
_CONFIG_FOR_DOC = "ChatGLMConfig"
|
| 42 |
+
|
| 43 |
+
def default_init(cls, *args, **kwargs):
|
| 44 |
+
return cls(*args, **kwargs)
|
| 45 |
+
|
| 46 |
+
|
| 47 |
+
class InvalidScoreLogitsProcessor(LogitsProcessor):
|
| 48 |
+
def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor) -> torch.FloatTensor:
|
| 49 |
+
if torch.isnan(scores).any() or torch.isinf(scores).any():
|
| 50 |
+
scores.zero_()
|
| 51 |
+
scores[..., 198] = 5e4
|
| 52 |
+
return scores
|
| 53 |
+
|
| 54 |
+
|
| 55 |
+
def split_tensor_along_last_dim(
|
| 56 |
+
tensor: torch.Tensor,
|
| 57 |
+
num_partitions: int,
|
| 58 |
+
contiguous_split_chunks: bool = False,
|
| 59 |
+
) -> List[torch.Tensor]:
|
| 60 |
+
"""Split a tensor along its last dimension.
|
| 61 |
+
|
| 62 |
+
Arguments:
|
| 63 |
+
tensor: input tensor.
|
| 64 |
+
num_partitions: number of partitions to split the tensor
|
| 65 |
+
contiguous_split_chunks: If True, make each chunk contiguous
|
| 66 |
+
in memory.
|
| 67 |
+
|
| 68 |
+
Returns:
|
| 69 |
+
A list of Tensors
|
| 70 |
+
"""
|
| 71 |
+
# Get the size and dimension.
|
| 72 |
+
last_dim = tensor.dim() - 1
|
| 73 |
+
last_dim_size = tensor.size()[last_dim] // num_partitions
|
| 74 |
+
# Split.
|
| 75 |
+
tensor_list = torch.split(tensor, last_dim_size, dim=last_dim)
|
| 76 |
+
# Note: torch.split does not create contiguous tensors by default.
|
| 77 |
+
if contiguous_split_chunks:
|
| 78 |
+
return tuple(chunk.contiguous() for chunk in tensor_list)
|
| 79 |
+
|
| 80 |
+
return tensor_list
|
| 81 |
+
|
| 82 |
+
|
| 83 |
+
class RotaryEmbedding(nn.Module):
|
| 84 |
+
def __init__(self, dim, rope_ratio=1, original_impl=False, device=None, dtype=None):
|
| 85 |
+
super().__init__()
|
| 86 |
+
inv_freq = 1.0 / (10000 ** (torch.arange(0, dim, 2, device=device).to(dtype=dtype) / dim))
|
| 87 |
+
self.register_buffer("inv_freq", inv_freq)
|
| 88 |
+
self.dim = dim
|
| 89 |
+
self.original_impl = original_impl
|
| 90 |
+
self.rope_ratio = rope_ratio
|
| 91 |
+
|
| 92 |
+
def forward_impl(
|
| 93 |
+
self, seq_len: int, n_elem: int, dtype: torch.dtype, device: torch.device, base: int = 10000
|
| 94 |
+
):
|
| 95 |
+
"""Enhanced Transformer with Rotary Position Embedding.
|
| 96 |
+
|
| 97 |
+
Derived from: https://github.com/labmlai/annotated_deep_learning_paper_implementations/blob/master/labml_nn/
|
| 98 |
+
transformers/rope/__init__.py. MIT License:
|
| 99 |
+
https://github.com/labmlai/annotated_deep_learning_paper_implementations/blob/master/license.
|
| 100 |
+
"""
|
| 101 |
+
# $\Theta = {\theta_i = 10000^{\frac{2(i-1)}{d}}, i \in [1, 2, ..., \frac{d}{2}]}$
|
| 102 |
+
base = base * self.rope_ratio
|
| 103 |
+
theta = 1.0 / (base ** (torch.arange(0, n_elem, 2, dtype=torch.float, device=device) / n_elem))
|
| 104 |
+
|
| 105 |
+
# Create position indexes `[0, 1, ..., seq_len - 1]`
|
| 106 |
+
seq_idx = torch.arange(seq_len, dtype=torch.float, device=device)
|
| 107 |
+
|
| 108 |
+
# Calculate the product of position index and $\theta_i$
|
| 109 |
+
idx_theta = torch.outer(seq_idx, theta).float()
|
| 110 |
+
|
| 111 |
+
cache = torch.stack([torch.cos(idx_theta), torch.sin(idx_theta)], dim=-1)
|
| 112 |
+
|
| 113 |
+
# this is to mimic the behaviour of complex32, else we will get different results
|
| 114 |
+
if dtype in (torch.float16, torch.bfloat16, torch.int8):
|
| 115 |
+
cache = cache.bfloat16() if dtype == torch.bfloat16 else cache.half()
|
| 116 |
+
return cache
|
| 117 |
+
|
| 118 |
+
def forward(self, max_seq_len, offset=0):
|
| 119 |
+
return self.forward_impl(
|
| 120 |
+
max_seq_len, self.dim, dtype=self.inv_freq.dtype, device=self.inv_freq.device
|
| 121 |
+
)
|
| 122 |
+
|
| 123 |
+
|
| 124 |
+
@torch.jit.script
|
| 125 |
+
def apply_rotary_pos_emb(x: torch.Tensor, rope_cache: torch.Tensor) -> torch.Tensor:
|
| 126 |
+
# x: [b, np, sq, hn]
|
| 127 |
+
b, np, sq, hn = x.size(0), x.size(1), x.size(2), x.size(3)
|
| 128 |
+
rot_dim = rope_cache.shape[-2] * 2
|
| 129 |
+
x, x_pass = x[..., :rot_dim], x[..., rot_dim:]
|
| 130 |
+
# truncate to support variable sizes
|
| 131 |
+
rope_cache = rope_cache[:, :sq]
|
| 132 |
+
xshaped = x.reshape(b, np, sq, rot_dim // 2, 2)
|
| 133 |
+
rope_cache = rope_cache.view(-1, 1, sq, xshaped.size(3), 2)
|
| 134 |
+
x_out2 = torch.stack(
|
| 135 |
+
[
|
| 136 |
+
xshaped[..., 0] * rope_cache[..., 0] - xshaped[..., 1] * rope_cache[..., 1],
|
| 137 |
+
xshaped[..., 1] * rope_cache[..., 0] + xshaped[..., 0] * rope_cache[..., 1],
|
| 138 |
+
],
|
| 139 |
+
-1,
|
| 140 |
+
)
|
| 141 |
+
x_out2 = x_out2.flatten(3)
|
| 142 |
+
return torch.cat((x_out2, x_pass), dim=-1)
|
| 143 |
+
|
| 144 |
+
|
| 145 |
+
class RMSNorm(torch.nn.Module):
|
| 146 |
+
def __init__(self, normalized_shape, eps=1e-5, device=None, dtype=None, **kwargs):
|
| 147 |
+
super().__init__()
|
| 148 |
+
self.weight = torch.nn.Parameter(torch.empty(normalized_shape, device=device, dtype=dtype))
|
| 149 |
+
self.eps = eps
|
| 150 |
+
|
| 151 |
+
def forward(self, hidden_states: torch.Tensor):
|
| 152 |
+
input_dtype = hidden_states.dtype
|
| 153 |
+
variance = hidden_states.to(torch.float32).pow(2).mean(-1, keepdim=True)
|
| 154 |
+
hidden_states = hidden_states * torch.rsqrt(variance + self.eps)
|
| 155 |
+
|
| 156 |
+
return (self.weight * hidden_states).to(input_dtype)
|
| 157 |
+
|
| 158 |
+
|
| 159 |
+
class CoreAttention(torch.nn.Module):
|
| 160 |
+
def __init__(self, config: ChatGLMConfig, layer_number):
|
| 161 |
+
super(CoreAttention, self).__init__()
|
| 162 |
+
|
| 163 |
+
self.apply_query_key_layer_scaling = config.apply_query_key_layer_scaling
|
| 164 |
+
self.attention_softmax_in_fp32 = config.attention_softmax_in_fp32
|
| 165 |
+
if self.apply_query_key_layer_scaling:
|
| 166 |
+
self.attention_softmax_in_fp32 = True
|
| 167 |
+
self.layer_number = max(1, layer_number)
|
| 168 |
+
|
| 169 |
+
projection_size = config.kv_channels * config.num_attention_heads
|
| 170 |
+
|
| 171 |
+
# Per attention head and per partition values.
|
| 172 |
+
self.hidden_size_per_partition = projection_size
|
| 173 |
+
self.hidden_size_per_attention_head = projection_size // config.num_attention_heads
|
| 174 |
+
self.num_attention_heads_per_partition = config.num_attention_heads
|
| 175 |
+
|
| 176 |
+
coeff = None
|
| 177 |
+
self.norm_factor = math.sqrt(self.hidden_size_per_attention_head)
|
| 178 |
+
if self.apply_query_key_layer_scaling:
|
| 179 |
+
coeff = self.layer_number
|
| 180 |
+
self.norm_factor *= coeff
|
| 181 |
+
self.coeff = coeff
|
| 182 |
+
|
| 183 |
+
self.attention_dropout = torch.nn.Dropout(config.attention_dropout)
|
| 184 |
+
|
| 185 |
+
def forward(self, query_layer, key_layer, value_layer, attention_mask):
|
| 186 |
+
pytorch_major_version = int(torch.__version__.split('.')[0])
|
| 187 |
+
if pytorch_major_version >= 2:
|
| 188 |
+
if attention_mask is None and query_layer.shape[2] == key_layer.shape[2]:
|
| 189 |
+
context_layer = torch.nn.functional.scaled_dot_product_attention(query_layer, key_layer, value_layer,
|
| 190 |
+
is_causal=True)
|
| 191 |
+
else:
|
| 192 |
+
if attention_mask is not None:
|
| 193 |
+
attention_mask = ~attention_mask
|
| 194 |
+
context_layer = torch.nn.functional.scaled_dot_product_attention(query_layer, key_layer, value_layer,
|
| 195 |
+
attention_mask)
|
| 196 |
+
context_layer = context_layer.transpose(1, 2).contiguous()
|
| 197 |
+
new_context_layer_shape = context_layer.size()[:-2] + (self.hidden_size_per_partition,)
|
| 198 |
+
context_layer = context_layer.reshape(*new_context_layer_shape)
|
| 199 |
+
else:
|
| 200 |
+
# Raw attention scores
|
| 201 |
+
|
| 202 |
+
# [b, np, sq, sk]
|
| 203 |
+
output_size = (query_layer.size(0), query_layer.size(1), query_layer.size(2), key_layer.size(2))
|
| 204 |
+
|
| 205 |
+
# [b, np, sq, hn] -> [b * np, sq, hn]
|
| 206 |
+
query_layer = query_layer.view(output_size[0] * output_size[1], output_size[2], -1)
|
| 207 |
+
# [b, np, sk, hn] -> [b * np, sk, hn]
|
| 208 |
+
key_layer = key_layer.view(output_size[0] * output_size[1], output_size[3], -1)
|
| 209 |
+
|
| 210 |
+
# preallocting input tensor: [b * np, sq, sk]
|
| 211 |
+
matmul_input_buffer = torch.empty(
|
| 212 |
+
output_size[0] * output_size[1], output_size[2], output_size[3], dtype=query_layer.dtype,
|
| 213 |
+
device=query_layer.device
|
| 214 |
+
)
|
| 215 |
+
|
| 216 |
+
# Raw attention scores. [b * np, sq, sk]
|
| 217 |
+
matmul_result = torch.baddbmm(
|
| 218 |
+
matmul_input_buffer,
|
| 219 |
+
query_layer, # [b * np, sq, hn]
|
| 220 |
+
key_layer.transpose(1, 2), # [b * np, hn, sk]
|
| 221 |
+
beta=0.0,
|
| 222 |
+
alpha=(1.0 / self.norm_factor),
|
| 223 |
+
)
|
| 224 |
+
|
| 225 |
+
# change view to [b, np, sq, sk]
|
| 226 |
+
attention_scores = matmul_result.view(*output_size)
|
| 227 |
+
|
| 228 |
+
# ===========================
|
| 229 |
+
# Attention probs and dropout
|
| 230 |
+
# ===========================
|
| 231 |
+
|
| 232 |
+
# attention scores and attention mask [b, np, sq, sk]
|
| 233 |
+
if self.attention_softmax_in_fp32:
|
| 234 |
+
attention_scores = attention_scores.float()
|
| 235 |
+
if self.coeff is not None:
|
| 236 |
+
attention_scores = attention_scores * self.coeff
|
| 237 |
+
if attention_mask is None and attention_scores.shape[2] == attention_scores.shape[3]:
|
| 238 |
+
attention_mask = torch.ones(output_size[0], 1, output_size[2], output_size[3],
|
| 239 |
+
device=attention_scores.device, dtype=torch.bool)
|
| 240 |
+
attention_mask.tril_()
|
| 241 |
+
attention_mask = ~attention_mask
|
| 242 |
+
if attention_mask is not None:
|
| 243 |
+
attention_scores = attention_scores.masked_fill(attention_mask, float("-inf"))
|
| 244 |
+
attention_probs = F.softmax(attention_scores, dim=-1)
|
| 245 |
+
attention_probs = attention_probs.type_as(value_layer)
|
| 246 |
+
|
| 247 |
+
# This is actually dropping out entire tokens to attend to, which might
|
| 248 |
+
# seem a bit unusual, but is taken from the original Transformer paper.
|
| 249 |
+
attention_probs = self.attention_dropout(attention_probs)
|
| 250 |
+
# =========================
|
| 251 |
+
# Context layer. [sq, b, hp]
|
| 252 |
+
# =========================
|
| 253 |
+
|
| 254 |
+
# value_layer -> context layer.
|
| 255 |
+
# [sk, b, np, hn] --> [b, np, sq, hn]
|
| 256 |
+
|
| 257 |
+
# context layer shape: [b, np, sq, hn]
|
| 258 |
+
output_size = (value_layer.size(1), value_layer.size(2), query_layer.size(0), value_layer.size(3))
|
| 259 |
+
# change view [b * np, sk, hn]
|
| 260 |
+
value_layer = value_layer.view(output_size[0] * output_size[1], value_layer.size(2), -1)
|
| 261 |
+
# change view [b * np, sq, sk]
|
| 262 |
+
attention_probs = attention_probs.view(output_size[0] * output_size[1], output_size[2], -1)
|
| 263 |
+
# matmul: [b * np, sq, hn]
|
| 264 |
+
context_layer = torch.bmm(attention_probs, value_layer)
|
| 265 |
+
# change view [b, np, sq, hn]
|
| 266 |
+
context_layer = context_layer.view(*output_size)
|
| 267 |
+
# [b, np, sq, hn] --> [b, sq, np, hn]
|
| 268 |
+
context_layer = context_layer.transpose(1, 2).contiguous()
|
| 269 |
+
# [b, sq, np, hn] --> [b, sq, hp]
|
| 270 |
+
new_context_layer_shape = context_layer.size()[:-2] + (self.hidden_size_per_partition,)
|
| 271 |
+
context_layer = context_layer.reshape(*new_context_layer_shape)
|
| 272 |
+
|
| 273 |
+
return context_layer
|
| 274 |
+
|
| 275 |
+
|
| 276 |
+
class SelfAttention(torch.nn.Module):
|
| 277 |
+
"""Parallel self-attention layer abstract class.
|
| 278 |
+
|
| 279 |
+
Self-attention layer takes input with size [s, b, h]
|
| 280 |
+
and returns output of the same size.
|
| 281 |
+
"""
|
| 282 |
+
|
| 283 |
+
def __init__(self, config: ChatGLMConfig, layer_number, device=None):
|
| 284 |
+
super(SelfAttention, self).__init__()
|
| 285 |
+
self.layer_number = max(1, layer_number)
|
| 286 |
+
|
| 287 |
+
self.projection_size = config.kv_channels * config.num_attention_heads
|
| 288 |
+
|
| 289 |
+
# Per attention head and per partition values.
|
| 290 |
+
self.hidden_size_per_attention_head = self.projection_size // config.num_attention_heads
|
| 291 |
+
self.num_attention_heads_per_partition = config.num_attention_heads
|
| 292 |
+
|
| 293 |
+
self.multi_query_attention = config.multi_query_attention
|
| 294 |
+
self.qkv_hidden_size = 3 * self.projection_size
|
| 295 |
+
if self.multi_query_attention:
|
| 296 |
+
self.num_multi_query_groups_per_partition = config.multi_query_group_num
|
| 297 |
+
self.qkv_hidden_size = (
|
| 298 |
+
self.projection_size + 2 * self.hidden_size_per_attention_head * config.multi_query_group_num
|
| 299 |
+
)
|
| 300 |
+
self.query_key_value = nn.Linear(config.hidden_size, self.qkv_hidden_size,
|
| 301 |
+
bias=config.add_bias_linear or config.add_qkv_bias,
|
| 302 |
+
device=device, **_config_to_kwargs(config)
|
| 303 |
+
)
|
| 304 |
+
|
| 305 |
+
self.core_attention = CoreAttention(config, self.layer_number)
|
| 306 |
+
|
| 307 |
+
# Output.
|
| 308 |
+
self.dense = nn.Linear(self.projection_size, config.hidden_size, bias=config.add_bias_linear,
|
| 309 |
+
device=device, **_config_to_kwargs(config)
|
| 310 |
+
)
|
| 311 |
+
|
| 312 |
+
def _allocate_memory(self, inference_max_sequence_len, batch_size, device=None, dtype=None):
|
| 313 |
+
if self.multi_query_attention:
|
| 314 |
+
num_attention_heads = self.num_multi_query_groups_per_partition
|
| 315 |
+
else:
|
| 316 |
+
num_attention_heads = self.num_attention_heads_per_partition
|
| 317 |
+
return torch.empty(
|
| 318 |
+
inference_max_sequence_len,
|
| 319 |
+
batch_size,
|
| 320 |
+
num_attention_heads,
|
| 321 |
+
self.hidden_size_per_attention_head,
|
| 322 |
+
dtype=dtype,
|
| 323 |
+
device=device,
|
| 324 |
+
)
|
| 325 |
+
|
| 326 |
+
def forward(
|
| 327 |
+
self, hidden_states, attention_mask, rotary_pos_emb, kv_cache=None, use_cache=True
|
| 328 |
+
):
|
| 329 |
+
# hidden_states: [b, sq, h]
|
| 330 |
+
|
| 331 |
+
# =================================================
|
| 332 |
+
# Pre-allocate memory for key-values for inference.
|
| 333 |
+
# =================================================
|
| 334 |
+
# =====================
|
| 335 |
+
# Query, Key, and Value
|
| 336 |
+
# =====================
|
| 337 |
+
|
| 338 |
+
# Attention heads [b, sq, h] --> [b, sq, (np * 3 * hn)]
|
| 339 |
+
mixed_x_layer = self.query_key_value(hidden_states)
|
| 340 |
+
|
| 341 |
+
if self.multi_query_attention:
|
| 342 |
+
(query_layer, key_layer, value_layer) = mixed_x_layer.split(
|
| 343 |
+
[
|
| 344 |
+
self.num_attention_heads_per_partition * self.hidden_size_per_attention_head,
|
| 345 |
+
self.num_multi_query_groups_per_partition * self.hidden_size_per_attention_head,
|
| 346 |
+
self.num_multi_query_groups_per_partition * self.hidden_size_per_attention_head,
|
| 347 |
+
],
|
| 348 |
+
dim=-1,
|
| 349 |
+
)
|
| 350 |
+
query_layer = query_layer.view(
|
| 351 |
+
query_layer.size()[:-1] + (self.num_attention_heads_per_partition, self.hidden_size_per_attention_head)
|
| 352 |
+
)
|
| 353 |
+
key_layer = key_layer.view(
|
| 354 |
+
key_layer.size()[:-1] + (self.num_multi_query_groups_per_partition, self.hidden_size_per_attention_head)
|
| 355 |
+
)
|
| 356 |
+
value_layer = value_layer.view(
|
| 357 |
+
value_layer.size()[:-1]
|
| 358 |
+
+ (self.num_multi_query_groups_per_partition, self.hidden_size_per_attention_head)
|
| 359 |
+
)
|
| 360 |
+
else:
|
| 361 |
+
new_tensor_shape = mixed_x_layer.size()[:-1] + \
|
| 362 |
+
(self.num_attention_heads_per_partition,
|
| 363 |
+
3 * self.hidden_size_per_attention_head)
|
| 364 |
+
mixed_x_layer = mixed_x_layer.view(*new_tensor_shape)
|
| 365 |
+
|
| 366 |
+
# [b, sq, np, 3 * hn] --> 3 [b, sq, np, hn]
|
| 367 |
+
(query_layer, key_layer, value_layer) = split_tensor_along_last_dim(mixed_x_layer, 3)
|
| 368 |
+
|
| 369 |
+
# [b, sq, np, hn] -> [b, np, sq, hn]
|
| 370 |
+
query_layer, key_layer, value_layer = [k.transpose(1, 2) for k in [query_layer, key_layer, value_layer]]
|
| 371 |
+
|
| 372 |
+
# apply relative positional encoding (rotary embedding)
|
| 373 |
+
if rotary_pos_emb is not None:
|
| 374 |
+
query_layer = apply_rotary_pos_emb(query_layer, rotary_pos_emb)
|
| 375 |
+
key_layer = apply_rotary_pos_emb(key_layer, rotary_pos_emb)
|
| 376 |
+
|
| 377 |
+
# adjust key and value for inference
|
| 378 |
+
if kv_cache is not None:
|
| 379 |
+
cache_k, cache_v = kv_cache
|
| 380 |
+
key_layer = torch.cat((cache_k, key_layer), dim=2)
|
| 381 |
+
value_layer = torch.cat((cache_v, value_layer), dim=2)
|
| 382 |
+
if use_cache:
|
| 383 |
+
if kv_cache is None:
|
| 384 |
+
kv_cache = torch.cat((key_layer.unsqueeze(0).unsqueeze(0), value_layer.unsqueeze(0).unsqueeze(0)), dim=1)
|
| 385 |
+
else:
|
| 386 |
+
kv_cache = (key_layer, value_layer)
|
| 387 |
+
else:
|
| 388 |
+
kv_cache = None
|
| 389 |
+
|
| 390 |
+
if self.multi_query_attention:
|
| 391 |
+
key_layer = key_layer.unsqueeze(2)
|
| 392 |
+
key_layer = key_layer.expand(
|
| 393 |
+
-1, -1, self.num_attention_heads_per_partition // self.num_multi_query_groups_per_partition, -1, -1
|
| 394 |
+
)
|
| 395 |
+
key_layer = key_layer.contiguous().view(
|
| 396 |
+
key_layer.size()[:1] + (self.num_attention_heads_per_partition,) + key_layer.size()[3:]
|
| 397 |
+
)
|
| 398 |
+
value_layer = value_layer.unsqueeze(2)
|
| 399 |
+
value_layer = value_layer.expand(
|
| 400 |
+
-1, -1, self.num_attention_heads_per_partition // self.num_multi_query_groups_per_partition, -1, -1
|
| 401 |
+
)
|
| 402 |
+
value_layer = value_layer.contiguous().view(
|
| 403 |
+
value_layer.size()[:1] + (self.num_attention_heads_per_partition,) + value_layer.size()[3:]
|
| 404 |
+
)
|
| 405 |
+
|
| 406 |
+
# ==================================
|
| 407 |
+
# core attention computation
|
| 408 |
+
# ==================================
|
| 409 |
+
|
| 410 |
+
context_layer = self.core_attention(query_layer, key_layer, value_layer, attention_mask)
|
| 411 |
+
|
| 412 |
+
# =================
|
| 413 |
+
# Output. [sq, b, h]
|
| 414 |
+
# =================
|
| 415 |
+
|
| 416 |
+
output = self.dense(context_layer)
|
| 417 |
+
|
| 418 |
+
return output, kv_cache
|
| 419 |
+
|
| 420 |
+
|
| 421 |
+
def _config_to_kwargs(args):
|
| 422 |
+
common_kwargs = {
|
| 423 |
+
"dtype": args.torch_dtype,
|
| 424 |
+
}
|
| 425 |
+
return common_kwargs
|
| 426 |
+
|
| 427 |
+
|
| 428 |
+
class MLP(torch.nn.Module):
|
| 429 |
+
"""MLP.
|
| 430 |
+
|
| 431 |
+
MLP will take the input with h hidden state, project it to 4*h
|
| 432 |
+
hidden dimension, perform nonlinear transformation, and project the
|
| 433 |
+
state back into h hidden dimension.
|
| 434 |
+
"""
|
| 435 |
+
|
| 436 |
+
def __init__(self, config: ChatGLMConfig, device=None):
|
| 437 |
+
super(MLP, self).__init__()
|
| 438 |
+
|
| 439 |
+
self.add_bias = config.add_bias_linear
|
| 440 |
+
|
| 441 |
+
# Project to 4h. If using swiglu double the output width, see https://arxiv.org/pdf/2002.05202.pdf
|
| 442 |
+
self.dense_h_to_4h = nn.Linear(
|
| 443 |
+
config.hidden_size,
|
| 444 |
+
config.ffn_hidden_size * 2,
|
| 445 |
+
bias=self.add_bias,
|
| 446 |
+
device=device,
|
| 447 |
+
**_config_to_kwargs(config)
|
| 448 |
+
)
|
| 449 |
+
|
| 450 |
+
def swiglu(x):
|
| 451 |
+
x = torch.chunk(x, 2, dim=-1)
|
| 452 |
+
return F.silu(x[0]) * x[1]
|
| 453 |
+
|
| 454 |
+
self.activation_func = swiglu
|
| 455 |
+
|
| 456 |
+
# Project back to h.
|
| 457 |
+
self.dense_4h_to_h = nn.Linear(
|
| 458 |
+
config.ffn_hidden_size,
|
| 459 |
+
config.hidden_size,
|
| 460 |
+
bias=self.add_bias,
|
| 461 |
+
device=device,
|
| 462 |
+
**_config_to_kwargs(config)
|
| 463 |
+
)
|
| 464 |
+
|
| 465 |
+
def forward(self, hidden_states):
|
| 466 |
+
# [s, b, 4hp]
|
| 467 |
+
intermediate_parallel = self.dense_h_to_4h(hidden_states)
|
| 468 |
+
intermediate_parallel = self.activation_func(intermediate_parallel)
|
| 469 |
+
# [s, b, h]
|
| 470 |
+
output = self.dense_4h_to_h(intermediate_parallel)
|
| 471 |
+
return output
|
| 472 |
+
|
| 473 |
+
|
| 474 |
+
class GLMBlock(torch.nn.Module):
|
| 475 |
+
"""A single transformer layer.
|
| 476 |
+
|
| 477 |
+
Transformer layer takes input with size [s, b, h] and returns an
|
| 478 |
+
output of the same size.
|
| 479 |
+
"""
|
| 480 |
+
|
| 481 |
+
def __init__(self, config: ChatGLMConfig, layer_number, device=None):
|
| 482 |
+
super(GLMBlock, self).__init__()
|
| 483 |
+
self.layer_number = layer_number
|
| 484 |
+
|
| 485 |
+
self.apply_residual_connection_post_layernorm = config.apply_residual_connection_post_layernorm
|
| 486 |
+
|
| 487 |
+
self.fp32_residual_connection = config.fp32_residual_connection
|
| 488 |
+
|
| 489 |
+
LayerNormFunc = RMSNorm if config.rmsnorm else LayerNorm
|
| 490 |
+
# Layernorm on the input data.
|
| 491 |
+
self.input_layernorm = LayerNormFunc(config.hidden_size, eps=config.layernorm_epsilon, device=device,
|
| 492 |
+
dtype=config.torch_dtype)
|
| 493 |
+
|
| 494 |
+
# Self attention.
|
| 495 |
+
self.self_attention = SelfAttention(config, layer_number, device=device)
|
| 496 |
+
self.hidden_dropout = config.hidden_dropout
|
| 497 |
+
|
| 498 |
+
# Layernorm on the attention output
|
| 499 |
+
self.post_attention_layernorm = LayerNormFunc(config.hidden_size, eps=config.layernorm_epsilon, device=device,
|
| 500 |
+
dtype=config.torch_dtype)
|
| 501 |
+
|
| 502 |
+
# MLP
|
| 503 |
+
self.mlp = MLP(config, device=device)
|
| 504 |
+
|
| 505 |
+
def forward(
|
| 506 |
+
self, hidden_states, attention_mask, rotary_pos_emb, kv_cache=None, use_cache=True,
|
| 507 |
+
):
|
| 508 |
+
# hidden_states: [s, b, h]
|
| 509 |
+
|
| 510 |
+
# Layer norm at the beginning of the transformer layer.
|
| 511 |
+
layernorm_output = self.input_layernorm(hidden_states)
|
| 512 |
+
# Self attention.
|
| 513 |
+
attention_output, kv_cache = self.self_attention(
|
| 514 |
+
layernorm_output,
|
| 515 |
+
attention_mask,
|
| 516 |
+
rotary_pos_emb,
|
| 517 |
+
kv_cache=kv_cache,
|
| 518 |
+
use_cache=use_cache
|
| 519 |
+
)
|
| 520 |
+
|
| 521 |
+
# Residual connection.
|
| 522 |
+
if self.apply_residual_connection_post_layernorm:
|
| 523 |
+
residual = layernorm_output
|
| 524 |
+
else:
|
| 525 |
+
residual = hidden_states
|
| 526 |
+
|
| 527 |
+
layernorm_input = torch.nn.functional.dropout(attention_output, p=self.hidden_dropout, training=self.training)
|
| 528 |
+
layernorm_input = residual + layernorm_input
|
| 529 |
+
|
| 530 |
+
# Layer norm post the self attention.
|
| 531 |
+
layernorm_output = self.post_attention_layernorm(layernorm_input)
|
| 532 |
+
|
| 533 |
+
# MLP.
|
| 534 |
+
mlp_output = self.mlp(layernorm_output)
|
| 535 |
+
|
| 536 |
+
# Second residual connection.
|
| 537 |
+
if self.apply_residual_connection_post_layernorm:
|
| 538 |
+
residual = layernorm_output
|
| 539 |
+
else:
|
| 540 |
+
residual = layernorm_input
|
| 541 |
+
|
| 542 |
+
output = torch.nn.functional.dropout(mlp_output, p=self.hidden_dropout, training=self.training)
|
| 543 |
+
output = residual + output
|
| 544 |
+
|
| 545 |
+
return output, kv_cache
|
| 546 |
+
|
| 547 |
+
|
| 548 |
+
class GLMTransformer(torch.nn.Module):
|
| 549 |
+
"""Transformer class."""
|
| 550 |
+
|
| 551 |
+
def __init__(self, config: ChatGLMConfig, device=None):
|
| 552 |
+
super(GLMTransformer, self).__init__()
|
| 553 |
+
|
| 554 |
+
self.fp32_residual_connection = config.fp32_residual_connection
|
| 555 |
+
self.post_layer_norm = config.post_layer_norm
|
| 556 |
+
|
| 557 |
+
# Number of layers.
|
| 558 |
+
self.num_layers = config.num_layers
|
| 559 |
+
|
| 560 |
+
# Transformer layers.
|
| 561 |
+
def build_layer(layer_number):
|
| 562 |
+
return GLMBlock(config, layer_number, device=device)
|
| 563 |
+
|
| 564 |
+
self.layers = torch.nn.ModuleList([build_layer(i + 1) for i in range(self.num_layers)])
|
| 565 |
+
|
| 566 |
+
if self.post_layer_norm:
|
| 567 |
+
LayerNormFunc = RMSNorm if config.rmsnorm else LayerNorm
|
| 568 |
+
# Final layer norm before output.
|
| 569 |
+
self.final_layernorm = LayerNormFunc(config.hidden_size, eps=config.layernorm_epsilon, device=device,
|
| 570 |
+
dtype=config.torch_dtype)
|
| 571 |
+
|
| 572 |
+
self.gradient_checkpointing = False
|
| 573 |
+
|
| 574 |
+
def _get_layer(self, layer_number):
|
| 575 |
+
return self.layers[layer_number]
|
| 576 |
+
|
| 577 |
+
def forward(
|
| 578 |
+
self, hidden_states, attention_mask, rotary_pos_emb, kv_caches=None,
|
| 579 |
+
use_cache: Optional[bool] = True,
|
| 580 |
+
output_hidden_states: Optional[bool] = False,
|
| 581 |
+
):
|
| 582 |
+
if not kv_caches:
|
| 583 |
+
kv_caches = [None for _ in range(self.num_layers)]
|
| 584 |
+
presents = () if use_cache else None
|
| 585 |
+
if self.gradient_checkpointing and self.training:
|
| 586 |
+
if use_cache:
|
| 587 |
+
logger.warning_once(
|
| 588 |
+
"`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`..."
|
| 589 |
+
)
|
| 590 |
+
use_cache = False
|
| 591 |
+
|
| 592 |
+
all_self_attentions = None
|
| 593 |
+
all_hidden_states = () if output_hidden_states else None
|
| 594 |
+
for index in range(self.num_layers):
|
| 595 |
+
if output_hidden_states:
|
| 596 |
+
all_hidden_states = all_hidden_states + (hidden_states,)
|
| 597 |
+
|
| 598 |
+
layer = self._get_layer(index)
|
| 599 |
+
if self.gradient_checkpointing and self.training:
|
| 600 |
+
layer_ret = torch.utils.checkpoint.checkpoint(
|
| 601 |
+
layer,
|
| 602 |
+
hidden_states,
|
| 603 |
+
attention_mask,
|
| 604 |
+
rotary_pos_emb,
|
| 605 |
+
kv_caches[index],
|
| 606 |
+
use_cache,
|
| 607 |
+
use_reentrant=False
|
| 608 |
+
)
|
| 609 |
+
else:
|
| 610 |
+
layer_ret = layer(
|
| 611 |
+
hidden_states,
|
| 612 |
+
attention_mask,
|
| 613 |
+
rotary_pos_emb,
|
| 614 |
+
kv_cache=kv_caches[index],
|
| 615 |
+
use_cache=use_cache
|
| 616 |
+
)
|
| 617 |
+
hidden_states, kv_cache = layer_ret
|
| 618 |
+
if use_cache:
|
| 619 |
+
# token by token decoding, use tuple format
|
| 620 |
+
if kv_caches[0] is not None:
|
| 621 |
+
presents = presents + (kv_cache,)
|
| 622 |
+
# prefilling in decoding, use tensor format to save cuda memory
|
| 623 |
+
else:
|
| 624 |
+
if len(presents) == 0:
|
| 625 |
+
presents = kv_cache
|
| 626 |
+
else:
|
| 627 |
+
presents = torch.cat((presents, kv_cache), dim=0)
|
| 628 |
+
|
| 629 |
+
if output_hidden_states:
|
| 630 |
+
all_hidden_states = all_hidden_states + (hidden_states,)
|
| 631 |
+
|
| 632 |
+
# Final layer norm.
|
| 633 |
+
if self.post_layer_norm:
|
| 634 |
+
hidden_states = self.final_layernorm(hidden_states)
|
| 635 |
+
|
| 636 |
+
return hidden_states, presents, all_hidden_states, all_self_attentions
|
| 637 |
+
|
| 638 |
+
|
| 639 |
+
class ChatGLMPreTrainedModel(PreTrainedModel):
|
| 640 |
+
"""
|
| 641 |
+
An abstract class to handle weights initialization and
|
| 642 |
+
a simple interface for downloading and loading pretrained models.
|
| 643 |
+
"""
|
| 644 |
+
|
| 645 |
+
is_parallelizable = False
|
| 646 |
+
supports_gradient_checkpointing = True
|
| 647 |
+
config_class = ChatGLMConfig
|
| 648 |
+
base_model_prefix = "transformer"
|
| 649 |
+
_no_split_modules = ["GLMBlock"]
|
| 650 |
+
|
| 651 |
+
def _init_weights(self, module: nn.Module):
|
| 652 |
+
"""Initialize the weights."""
|
| 653 |
+
return
|
| 654 |
+
|
| 655 |
+
def get_masks(self, input_ids, past_key_values, padding_mask=None):
|
| 656 |
+
batch_size, seq_length = input_ids.shape
|
| 657 |
+
full_attention_mask = torch.ones(batch_size, seq_length, seq_length, device=input_ids.device)
|
| 658 |
+
full_attention_mask.tril_()
|
| 659 |
+
past_length = 0
|
| 660 |
+
if past_key_values:
|
| 661 |
+
past_length = past_key_values[0][0].shape[2]
|
| 662 |
+
if past_length:
|
| 663 |
+
full_attention_mask = torch.cat((torch.ones(batch_size, seq_length, past_length,
|
| 664 |
+
device=input_ids.device), full_attention_mask), dim=-1)
|
| 665 |
+
if padding_mask is not None:
|
| 666 |
+
full_attention_mask = full_attention_mask * padding_mask.unsqueeze(1)
|
| 667 |
+
if not past_length and padding_mask is not None:
|
| 668 |
+
full_attention_mask -= padding_mask.unsqueeze(-1) - 1
|
| 669 |
+
full_attention_mask = (full_attention_mask < 0.5).bool()
|
| 670 |
+
full_attention_mask.unsqueeze_(1)
|
| 671 |
+
return full_attention_mask
|
| 672 |
+
|
| 673 |
+
def get_position_ids(self, input_ids, device):
|
| 674 |
+
batch_size, seq_length = input_ids.shape
|
| 675 |
+
position_ids = torch.arange(seq_length, dtype=torch.long, device=device).unsqueeze(0).repeat(batch_size, 1)
|
| 676 |
+
return position_ids
|
| 677 |
+
|
| 678 |
+
def gradient_checkpointing_enable(self, gradient_checkpointing_kwargs=None):
|
| 679 |
+
if not self.supports_gradient_checkpointing:
|
| 680 |
+
raise ValueError(f"{self.__class__.__name__} does not support gradient checkpointing.")
|
| 681 |
+
|
| 682 |
+
|
| 683 |
+
class Embedding(torch.nn.Module):
|
| 684 |
+
"""Language model embeddings."""
|
| 685 |
+
|
| 686 |
+
def __init__(self, config: ChatGLMConfig, device=None):
|
| 687 |
+
super(Embedding, self).__init__()
|
| 688 |
+
|
| 689 |
+
self.hidden_size = config.hidden_size
|
| 690 |
+
# Word embeddings (parallel).
|
| 691 |
+
self.word_embeddings = nn.Embedding(
|
| 692 |
+
config.padded_vocab_size,
|
| 693 |
+
self.hidden_size,
|
| 694 |
+
dtype=config.torch_dtype,
|
| 695 |
+
device=device
|
| 696 |
+
)
|
| 697 |
+
self.fp32_residual_connection = config.fp32_residual_connection
|
| 698 |
+
|
| 699 |
+
def forward(self, input_ids):
|
| 700 |
+
# Embeddings.
|
| 701 |
+
words_embeddings = self.word_embeddings(input_ids)
|
| 702 |
+
embeddings = words_embeddings
|
| 703 |
+
# If the input flag for fp32 residual connection is set, convert for float.
|
| 704 |
+
if self.fp32_residual_connection:
|
| 705 |
+
embeddings = embeddings.float()
|
| 706 |
+
return embeddings
|
| 707 |
+
|
| 708 |
+
|
| 709 |
+
class ChatGLMModel(ChatGLMPreTrainedModel):
|
| 710 |
+
def __init__(self, config: ChatGLMConfig, device=None, empty_init=True):
|
| 711 |
+
super().__init__(config)
|
| 712 |
+
if empty_init:
|
| 713 |
+
init_method = skip_init
|
| 714 |
+
else:
|
| 715 |
+
init_method = default_init
|
| 716 |
+
init_kwargs = {}
|
| 717 |
+
if device is not None:
|
| 718 |
+
init_kwargs["device"] = device
|
| 719 |
+
self.embedding = init_method(Embedding, config, **init_kwargs)
|
| 720 |
+
self.num_layers = config.num_layers
|
| 721 |
+
self.multi_query_group_num = config.multi_query_group_num
|
| 722 |
+
self.kv_channels = config.kv_channels
|
| 723 |
+
|
| 724 |
+
# Rotary positional embeddings
|
| 725 |
+
self.seq_length = config.seq_length
|
| 726 |
+
rotary_dim = (
|
| 727 |
+
config.hidden_size // config.num_attention_heads if config.kv_channels is None else config.kv_channels
|
| 728 |
+
)
|
| 729 |
+
|
| 730 |
+
self.rotary_pos_emb = RotaryEmbedding(rotary_dim // 2, rope_ratio=config.rope_ratio, original_impl=config.original_rope,
|
| 731 |
+
device=device, dtype=config.torch_dtype)
|
| 732 |
+
self.encoder = init_method(GLMTransformer, config, **init_kwargs)
|
| 733 |
+
self.output_layer = init_method(nn.Linear, config.hidden_size, config.padded_vocab_size, bias=False,
|
| 734 |
+
dtype=config.torch_dtype, **init_kwargs)
|
| 735 |
+
|
| 736 |
+
def get_input_embeddings(self):
|
| 737 |
+
return self.embedding.word_embeddings
|
| 738 |
+
|
| 739 |
+
def set_input_embeddings(self, value):
|
| 740 |
+
self.embedding.word_embeddings = value
|
| 741 |
+
|
| 742 |
+
def forward(
|
| 743 |
+
self,
|
| 744 |
+
input_ids,
|
| 745 |
+
position_ids: Optional[torch.Tensor] = None,
|
| 746 |
+
attention_mask: Optional[torch.BoolTensor] = None,
|
| 747 |
+
full_attention_mask: Optional[torch.BoolTensor] = None,
|
| 748 |
+
past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
|
| 749 |
+
inputs_embeds: Optional[torch.Tensor] = None,
|
| 750 |
+
use_cache: Optional[bool] = None,
|
| 751 |
+
output_hidden_states: Optional[bool] = None,
|
| 752 |
+
return_dict: Optional[bool] = None,
|
| 753 |
+
):
|
| 754 |
+
output_hidden_states = (
|
| 755 |
+
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
|
| 756 |
+
)
|
| 757 |
+
use_cache = use_cache if use_cache is not None else self.config.use_cache
|
| 758 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
| 759 |
+
|
| 760 |
+
batch_size, seq_length = input_ids.shape
|
| 761 |
+
|
| 762 |
+
if inputs_embeds is None:
|
| 763 |
+
inputs_embeds = self.embedding(input_ids)
|
| 764 |
+
|
| 765 |
+
if full_attention_mask is None:
|
| 766 |
+
if (attention_mask is not None and not attention_mask.all()) or (past_key_values and seq_length != 1):
|
| 767 |
+
full_attention_mask = self.get_masks(input_ids, past_key_values, padding_mask=attention_mask)
|
| 768 |
+
|
| 769 |
+
# Rotary positional embeddings
|
| 770 |
+
rotary_pos_emb = self.rotary_pos_emb(self.seq_length)
|
| 771 |
+
if position_ids is not None:
|
| 772 |
+
rotary_pos_emb = rotary_pos_emb[position_ids]
|
| 773 |
+
else:
|
| 774 |
+
rotary_pos_emb = rotary_pos_emb[None, :seq_length]
|
| 775 |
+
|
| 776 |
+
# Run encoder.
|
| 777 |
+
hidden_states, presents, all_hidden_states, all_self_attentions = self.encoder(
|
| 778 |
+
inputs_embeds, full_attention_mask, rotary_pos_emb=rotary_pos_emb,
|
| 779 |
+
kv_caches=past_key_values, use_cache=use_cache, output_hidden_states=output_hidden_states
|
| 780 |
+
)
|
| 781 |
+
if presents is not None and type(presents) is torch.Tensor:
|
| 782 |
+
presents = presents.split(1, dim=0)
|
| 783 |
+
presents = list(presents)
|
| 784 |
+
presents = [list(x.squeeze(0).split(1, dim=0)) for x in presents]
|
| 785 |
+
presents = [tuple([x.squeeze(0) for x in y]) for y in presents]
|
| 786 |
+
presents = tuple(presents)
|
| 787 |
+
|
| 788 |
+
if not return_dict:
|
| 789 |
+
return tuple(v for v in [hidden_states, presents, all_hidden_states, all_self_attentions] if v is not None)
|
| 790 |
+
|
| 791 |
+
return BaseModelOutputWithPast(
|
| 792 |
+
last_hidden_state=hidden_states,
|
| 793 |
+
past_key_values=presents,
|
| 794 |
+
hidden_states=all_hidden_states,
|
| 795 |
+
attentions=all_self_attentions,
|
| 796 |
+
)
|
| 797 |
+
|
| 798 |
+
|
| 799 |
+
class ChatGLMForConditionalGeneration(ChatGLMPreTrainedModel):
|
| 800 |
+
def __init__(self, config: ChatGLMConfig, empty_init=True, device=None):
|
| 801 |
+
super().__init__(config)
|
| 802 |
+
|
| 803 |
+
self.max_sequence_length = config.max_length
|
| 804 |
+
self.transformer = ChatGLMModel(config, empty_init=empty_init, device=device)
|
| 805 |
+
self.config = config
|
| 806 |
+
|
| 807 |
+
def _update_model_kwargs_for_generation(
|
| 808 |
+
self,
|
| 809 |
+
outputs: ModelOutput,
|
| 810 |
+
model_kwargs: Dict[str, Any],
|
| 811 |
+
is_encoder_decoder: bool = False,
|
| 812 |
+
standardize_cache_format: bool = False,
|
| 813 |
+
) -> Dict[str, Any]:
|
| 814 |
+
# update past_key_values
|
| 815 |
+
model_kwargs["past_key_values"] = self._extract_past_from_model_output(
|
| 816 |
+
outputs, standardize_cache_format=standardize_cache_format
|
| 817 |
+
)
|
| 818 |
+
|
| 819 |
+
# update attention mask
|
| 820 |
+
if "attention_mask" in model_kwargs:
|
| 821 |
+
attention_mask = model_kwargs["attention_mask"]
|
| 822 |
+
model_kwargs["attention_mask"] = torch.cat(
|
| 823 |
+
[attention_mask, attention_mask.new_ones((attention_mask.shape[0], 1))], dim=-1
|
| 824 |
+
)
|
| 825 |
+
|
| 826 |
+
# update position ids
|
| 827 |
+
if "position_ids" in model_kwargs:
|
| 828 |
+
position_ids = model_kwargs["position_ids"]
|
| 829 |
+
new_position_id = position_ids[..., -1:].clone()
|
| 830 |
+
new_position_id += 1
|
| 831 |
+
model_kwargs["position_ids"] = torch.cat(
|
| 832 |
+
[position_ids, new_position_id], dim=-1
|
| 833 |
+
)
|
| 834 |
+
|
| 835 |
+
model_kwargs["is_first_forward"] = False
|
| 836 |
+
return model_kwargs
|
| 837 |
+
|
| 838 |
+
def prepare_inputs_for_generation(
|
| 839 |
+
self,
|
| 840 |
+
input_ids: torch.LongTensor,
|
| 841 |
+
past_key_values: Optional[torch.Tensor] = None,
|
| 842 |
+
attention_mask: Optional[torch.Tensor] = None,
|
| 843 |
+
position_ids: Optional[torch.Tensor] = None,
|
| 844 |
+
use_cache: Optional[bool] = None,
|
| 845 |
+
is_first_forward: bool = True,
|
| 846 |
+
**kwargs
|
| 847 |
+
) -> dict:
|
| 848 |
+
# only last token for input_ids if past is not None
|
| 849 |
+
if position_ids is None:
|
| 850 |
+
position_ids = self.get_position_ids(input_ids, device=input_ids.device)
|
| 851 |
+
if not is_first_forward:
|
| 852 |
+
if past_key_values is not None:
|
| 853 |
+
position_ids = position_ids[..., -1:]
|
| 854 |
+
input_ids = input_ids[:, -1:]
|
| 855 |
+
return {
|
| 856 |
+
"input_ids": input_ids,
|
| 857 |
+
"past_key_values": past_key_values,
|
| 858 |
+
"position_ids": position_ids,
|
| 859 |
+
"attention_mask": attention_mask,
|
| 860 |
+
"return_last_logit": True,
|
| 861 |
+
"use_cache": use_cache
|
| 862 |
+
}
|
| 863 |
+
|
| 864 |
+
def forward(
|
| 865 |
+
self,
|
| 866 |
+
input_ids: Optional[torch.Tensor] = None,
|
| 867 |
+
position_ids: Optional[torch.Tensor] = None,
|
| 868 |
+
attention_mask: Optional[torch.Tensor] = None,
|
| 869 |
+
past_key_values: Optional[Tuple[torch.FloatTensor]] = None,
|
| 870 |
+
inputs_embeds: Optional[torch.Tensor] = None,
|
| 871 |
+
labels: Optional[torch.Tensor] = None,
|
| 872 |
+
use_cache: Optional[bool] = None,
|
| 873 |
+
output_attentions: Optional[bool] = None,
|
| 874 |
+
output_hidden_states: Optional[bool] = None,
|
| 875 |
+
return_dict: Optional[bool] = None,
|
| 876 |
+
return_last_logit: Optional[bool] = False,
|
| 877 |
+
):
|
| 878 |
+
use_cache = use_cache if use_cache is not None else self.config.use_cache
|
| 879 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
| 880 |
+
|
| 881 |
+
transformer_outputs = self.transformer(
|
| 882 |
+
input_ids=input_ids,
|
| 883 |
+
position_ids=position_ids,
|
| 884 |
+
attention_mask=attention_mask,
|
| 885 |
+
past_key_values=past_key_values,
|
| 886 |
+
inputs_embeds=inputs_embeds,
|
| 887 |
+
use_cache=use_cache,
|
| 888 |
+
output_hidden_states=output_hidden_states,
|
| 889 |
+
return_dict=return_dict,
|
| 890 |
+
)
|
| 891 |
+
|
| 892 |
+
hidden_states = transformer_outputs[0]
|
| 893 |
+
if return_last_logit:
|
| 894 |
+
hidden_states = hidden_states[:, -1:]
|
| 895 |
+
lm_logits = self.transformer.output_layer(hidden_states)
|
| 896 |
+
|
| 897 |
+
loss = None
|
| 898 |
+
if labels is not None:
|
| 899 |
+
lm_logits = lm_logits.to(torch.float32)
|
| 900 |
+
|
| 901 |
+
# Shift so that tokens < n predict n
|
| 902 |
+
shift_logits = lm_logits[..., :-1, :].contiguous()
|
| 903 |
+
shift_labels = labels[..., 1:].contiguous()
|
| 904 |
+
# Flatten the tokens
|
| 905 |
+
loss_fct = CrossEntropyLoss(ignore_index=-100)
|
| 906 |
+
loss = loss_fct(shift_logits.view(-1, shift_logits.size(-1)), shift_labels.view(-1))
|
| 907 |
+
|
| 908 |
+
lm_logits = lm_logits.to(hidden_states.dtype)
|
| 909 |
+
loss = loss.to(hidden_states.dtype)
|
| 910 |
+
|
| 911 |
+
if not return_dict:
|
| 912 |
+
output = (lm_logits,) + transformer_outputs[1:]
|
| 913 |
+
return ((loss,) + output) if loss is not None else output
|
| 914 |
+
|
| 915 |
+
return CausalLMOutputWithPast(
|
| 916 |
+
loss=loss,
|
| 917 |
+
logits=lm_logits,
|
| 918 |
+
past_key_values=transformer_outputs.past_key_values,
|
| 919 |
+
hidden_states=transformer_outputs.hidden_states,
|
| 920 |
+
attentions=transformer_outputs.attentions,
|
| 921 |
+
)
|
| 922 |
+
|
| 923 |
+
@staticmethod
|
| 924 |
+
def _reorder_cache(
|
| 925 |
+
past: Tuple[Tuple[torch.Tensor, torch.Tensor], ...], beam_idx: torch.LongTensor
|
| 926 |
+
) -> Tuple[Tuple[torch.Tensor, torch.Tensor], ...]:
|
| 927 |
+
"""
|
| 928 |
+
This function is used to re-order the `past_key_values` cache if [`~PreTrainedModel.beam_search`] or
|
| 929 |
+
[`~PreTrainedModel.beam_sample`] is called. This is required to match `past_key_values` with the correct
|
| 930 |
+
beam_idx at every generation step.
|
| 931 |
+
|
| 932 |
+
Output shares the same memory storage as `past`.
|
| 933 |
+
"""
|
| 934 |
+
return tuple(
|
| 935 |
+
(
|
| 936 |
+
layer_past[0].index_select(0, beam_idx.to(layer_past[0].device)),
|
| 937 |
+
layer_past[1].index_select(0, beam_idx.to(layer_past[1].device)),
|
| 938 |
+
)
|
| 939 |
+
for layer_past in past
|
| 940 |
+
)
|
| 941 |
+
|
| 942 |
+
def process_response(self, output, history):
|
| 943 |
+
content = ""
|
| 944 |
+
history = deepcopy(history)
|
| 945 |
+
for response in output.split("<|assistant|>"):
|
| 946 |
+
if "\n" in response:
|
| 947 |
+
metadata, content = response.split("\n", maxsplit=1)
|
| 948 |
+
else:
|
| 949 |
+
metadata, content = "", response
|
| 950 |
+
if not metadata.strip():
|
| 951 |
+
content = content.strip()
|
| 952 |
+
history.append({"role": "assistant", "metadata": metadata, "content": content})
|
| 953 |
+
content = content.replace("[[训练时间]]", "2023年")
|
| 954 |
+
else:
|
| 955 |
+
history.append({"role": "assistant", "metadata": metadata, "content": content})
|
| 956 |
+
if history[0]["role"] == "system" and "tools" in history[0]:
|
| 957 |
+
parameters = json.loads(content)
|
| 958 |
+
content = {"name": metadata.strip(), "parameters": parameters}
|
| 959 |
+
else:
|
| 960 |
+
content = {"name": metadata.strip(), "content": content}
|
| 961 |
+
return content, history
|
| 962 |
+
|
| 963 |
+
@torch.inference_mode()
|
| 964 |
+
def chat(self, tokenizer, query: str, history: List[Dict] = None, role: str = "user",
|
| 965 |
+
max_length: int = 8192, num_beams=1, do_sample=True, top_p=0.8, temperature=0.8, logits_processor=None,
|
| 966 |
+
**kwargs):
|
| 967 |
+
if history is None:
|
| 968 |
+
history = []
|
| 969 |
+
if logits_processor is None:
|
| 970 |
+
logits_processor = LogitsProcessorList()
|
| 971 |
+
logits_processor.append(InvalidScoreLogitsProcessor())
|
| 972 |
+
gen_kwargs = {"max_length": max_length, "num_beams": num_beams, "do_sample": do_sample, "top_p": top_p,
|
| 973 |
+
"temperature": temperature, "logits_processor": logits_processor, **kwargs}
|
| 974 |
+
history.append({"role": role, "content": query})
|
| 975 |
+
inputs = tokenizer.apply_chat_template(history, add_generation_prompt=True, tokenize=True,
|
| 976 |
+
return_tensors="pt", return_dict=True)
|
| 977 |
+
inputs = inputs.to(self.device)
|
| 978 |
+
eos_token_id = [tokenizer.eos_token_id, tokenizer.convert_tokens_to_ids("<|user|>"),
|
| 979 |
+
tokenizer.convert_tokens_to_ids("<|observation|>")]
|
| 980 |
+
outputs = self.generate(**inputs, **gen_kwargs, eos_token_id=eos_token_id)
|
| 981 |
+
outputs = outputs.tolist()[0][len(inputs["input_ids"][0]):-1]
|
| 982 |
+
response = tokenizer.decode(outputs)
|
| 983 |
+
response, history = self.process_response(response, history)
|
| 984 |
+
return response, history
|
| 985 |
+
|
| 986 |
+
@torch.inference_mode()
|
| 987 |
+
def stream_chat(self, tokenizer, query: str, history: List[Dict] = None, role: str = "user",
|
| 988 |
+
past_key_values=None, max_length: int = 8192, do_sample=True, top_p=0.8, temperature=0.8,
|
| 989 |
+
logits_processor=None, return_past_key_values=False, **kwargs):
|
| 990 |
+
if history is None:
|
| 991 |
+
history = []
|
| 992 |
+
if logits_processor is None:
|
| 993 |
+
logits_processor = LogitsProcessorList()
|
| 994 |
+
logits_processor.append(InvalidScoreLogitsProcessor())
|
| 995 |
+
eos_token_id = [tokenizer.eos_token_id, tokenizer.convert_tokens_to_ids("<|user|>"),
|
| 996 |
+
tokenizer.convert_tokens_to_ids("<|observation|>")]
|
| 997 |
+
gen_kwargs = {"max_length": max_length, "do_sample": do_sample, "top_p": top_p,
|
| 998 |
+
"temperature": temperature, "logits_processor": logits_processor, **kwargs}
|
| 999 |
+
if past_key_values is None:
|
| 1000 |
+
inputs = tokenizer.apply_chat_template(history + [{"role": role, "content": query}],
|
| 1001 |
+
add_generation_prompt=True, tokenize=True, return_tensors="pt",
|
| 1002 |
+
return_dict=True)
|
| 1003 |
+
else:
|
| 1004 |
+
inputs = tokenizer.apply_chat_template([{"role": role, "content": query}], add_special_tokens=False,
|
| 1005 |
+
add_generation_prompt=True, tokenize=True, return_tensors="pt",
|
| 1006 |
+
return_dict=True)
|
| 1007 |
+
inputs = inputs.to(self.device)
|
| 1008 |
+
if past_key_values is not None:
|
| 1009 |
+
past_length = past_key_values[0][0].shape[2]
|
| 1010 |
+
inputs.position_ids += past_length
|
| 1011 |
+
attention_mask = inputs.attention_mask
|
| 1012 |
+
attention_mask = torch.cat((attention_mask.new_ones(1, past_length), attention_mask), dim=1)
|
| 1013 |
+
inputs['attention_mask'] = attention_mask
|
| 1014 |
+
history.append({"role": role, "content": query})
|
| 1015 |
+
for outputs in self.stream_generate(**inputs, past_key_values=past_key_values,
|
| 1016 |
+
eos_token_id=eos_token_id, return_past_key_values=return_past_key_values,
|
| 1017 |
+
**gen_kwargs):
|
| 1018 |
+
if return_past_key_values:
|
| 1019 |
+
outputs, past_key_values = outputs
|
| 1020 |
+
outputs = outputs.tolist()[0][len(inputs["input_ids"][0]):-1]
|
| 1021 |
+
response = tokenizer.decode(outputs)
|
| 1022 |
+
if response and response[-1] != "�":
|
| 1023 |
+
response, new_history = self.process_response(response, history)
|
| 1024 |
+
if return_past_key_values:
|
| 1025 |
+
yield response, new_history, past_key_values
|
| 1026 |
+
else:
|
| 1027 |
+
yield response, new_history
|
| 1028 |
+
|
| 1029 |
+
@torch.inference_mode()
|
| 1030 |
+
def stream_generate(
|
| 1031 |
+
self,
|
| 1032 |
+
input_ids,
|
| 1033 |
+
generation_config: Optional[GenerationConfig] = None,
|
| 1034 |
+
logits_processor: Optional[LogitsProcessorList] = None,
|
| 1035 |
+
stopping_criteria: Optional[StoppingCriteriaList] = None,
|
| 1036 |
+
prefix_allowed_tokens_fn: Optional[Callable[[int, torch.Tensor], List[int]]] = None,
|
| 1037 |
+
return_past_key_values=False,
|
| 1038 |
+
**kwargs,
|
| 1039 |
+
):
|
| 1040 |
+
batch_size, input_ids_seq_length = input_ids.shape[0], input_ids.shape[-1]
|
| 1041 |
+
|
| 1042 |
+
if generation_config is None:
|
| 1043 |
+
generation_config = self.generation_config
|
| 1044 |
+
generation_config = copy.deepcopy(generation_config)
|
| 1045 |
+
model_kwargs = generation_config.update(**kwargs)
|
| 1046 |
+
model_kwargs["use_cache"] = generation_config.use_cache
|
| 1047 |
+
bos_token_id, eos_token_id = generation_config.bos_token_id, generation_config.eos_token_id
|
| 1048 |
+
|
| 1049 |
+
if isinstance(eos_token_id, int):
|
| 1050 |
+
eos_token_id = [eos_token_id]
|
| 1051 |
+
eos_token_id_tensor = torch.tensor(eos_token_id).to(input_ids.device) if eos_token_id is not None else None
|
| 1052 |
+
|
| 1053 |
+
has_default_max_length = kwargs.get("max_length") is None and generation_config.max_length is not None
|
| 1054 |
+
if has_default_max_length and generation_config.max_new_tokens is None:
|
| 1055 |
+
warnings.warn(
|
| 1056 |
+
f"Using `max_length`'s default ({generation_config.max_length}) to control the generation length. "
|
| 1057 |
+
"This behaviour is deprecated and will be removed from the config in v5 of Transformers -- we"
|
| 1058 |
+
" recommend using `max_new_tokens` to control the maximum length of the generation.",
|
| 1059 |
+
UserWarning,
|
| 1060 |
+
)
|
| 1061 |
+
elif generation_config.max_new_tokens is not None:
|
| 1062 |
+
generation_config.max_length = generation_config.max_new_tokens + input_ids_seq_length
|
| 1063 |
+
if not has_default_max_length:
|
| 1064 |
+
logger.warn(
|
| 1065 |
+
f"Both `max_new_tokens` (={generation_config.max_new_tokens}) and `max_length`(="
|
| 1066 |
+
f"{generation_config.max_length}) seem to have been set. `max_new_tokens` will take precedence. "
|
| 1067 |
+
"Please refer to the documentation for more information. "
|
| 1068 |
+
"(https://huggingface.co/docs/transformers/main/en/main_classes/text_generation)",
|
| 1069 |
+
UserWarning,
|
| 1070 |
+
)
|
| 1071 |
+
|
| 1072 |
+
if input_ids_seq_length >= generation_config.max_length:
|
| 1073 |
+
input_ids_string = "decoder_input_ids" if self.config.is_encoder_decoder else "input_ids"
|
| 1074 |
+
logger.warning(
|
| 1075 |
+
f"Input length of {input_ids_string} is {input_ids_seq_length}, but `max_length` is set to"
|
| 1076 |
+
f" {generation_config.max_length}. This can lead to unexpected behavior. You should consider"
|
| 1077 |
+
" increasing `max_new_tokens`."
|
| 1078 |
+
)
|
| 1079 |
+
|
| 1080 |
+
# 2. Set generation parameters if not already defined
|
| 1081 |
+
logits_processor = logits_processor if logits_processor is not None else LogitsProcessorList()
|
| 1082 |
+
stopping_criteria = stopping_criteria if stopping_criteria is not None else StoppingCriteriaList()
|
| 1083 |
+
|
| 1084 |
+
logits_processor = self._get_logits_processor(
|
| 1085 |
+
generation_config=generation_config,
|
| 1086 |
+
input_ids_seq_length=input_ids_seq_length,
|
| 1087 |
+
encoder_input_ids=input_ids,
|
| 1088 |
+
prefix_allowed_tokens_fn=prefix_allowed_tokens_fn,
|
| 1089 |
+
logits_processor=logits_processor,
|
| 1090 |
+
)
|
| 1091 |
+
|
| 1092 |
+
stopping_criteria = self._get_stopping_criteria(
|
| 1093 |
+
generation_config=generation_config, stopping_criteria=stopping_criteria
|
| 1094 |
+
)
|
| 1095 |
+
logits_warper = self._get_logits_warper(generation_config)
|
| 1096 |
+
|
| 1097 |
+
unfinished_sequences = input_ids.new(input_ids.shape[0]).fill_(1)
|
| 1098 |
+
scores = None
|
| 1099 |
+
while True:
|
| 1100 |
+
model_inputs = self.prepare_inputs_for_generation(input_ids, **model_kwargs)
|
| 1101 |
+
# forward pass to get next token
|
| 1102 |
+
outputs = self(
|
| 1103 |
+
**model_inputs,
|
| 1104 |
+
return_dict=True,
|
| 1105 |
+
output_attentions=False,
|
| 1106 |
+
output_hidden_states=False,
|
| 1107 |
+
)
|
| 1108 |
+
|
| 1109 |
+
next_token_logits = outputs.logits[:, -1, :]
|
| 1110 |
+
|
| 1111 |
+
# pre-process distribution
|
| 1112 |
+
next_token_scores = logits_processor(input_ids, next_token_logits)
|
| 1113 |
+
next_token_scores = logits_warper(input_ids, next_token_scores)
|
| 1114 |
+
|
| 1115 |
+
# sample
|
| 1116 |
+
probs = nn.functional.softmax(next_token_scores, dim=-1)
|
| 1117 |
+
if generation_config.do_sample:
|
| 1118 |
+
next_tokens = torch.multinomial(probs, num_samples=1).squeeze(1)
|
| 1119 |
+
else:
|
| 1120 |
+
next_tokens = torch.argmax(probs, dim=-1)
|
| 1121 |
+
# update generated ids, model inputs, and length for next step
|
| 1122 |
+
input_ids = torch.cat([input_ids, next_tokens[:, None]], dim=-1)
|
| 1123 |
+
model_kwargs = self._update_model_kwargs_for_generation(
|
| 1124 |
+
outputs, model_kwargs, is_encoder_decoder=self.config.is_encoder_decoder
|
| 1125 |
+
)
|
| 1126 |
+
unfinished_sequences = unfinished_sequences.mul(
|
| 1127 |
+
next_tokens.tile(eos_token_id_tensor.shape[0], 1).ne(eos_token_id_tensor.unsqueeze(1)).prod(dim=0)
|
| 1128 |
+
)
|
| 1129 |
+
if return_past_key_values:
|
| 1130 |
+
yield input_ids, outputs.past_key_values
|
| 1131 |
+
else:
|
| 1132 |
+
yield input_ids
|
| 1133 |
+
# stop when each sentence is finished, or if we exceed the maximum length
|
| 1134 |
+
if unfinished_sequences.max() == 0 or stopping_criteria(input_ids, scores):
|
| 1135 |
+
break
|
| 1136 |
+
|
| 1137 |
+
|
| 1138 |
+
class ChatGLMForSequenceClassification(ChatGLMPreTrainedModel):
|
| 1139 |
+
def __init__(self, config: ChatGLMConfig, empty_init=True, device=None):
|
| 1140 |
+
super().__init__(config)
|
| 1141 |
+
|
| 1142 |
+
self.num_labels = config.num_labels
|
| 1143 |
+
self.transformer = ChatGLMModel(config, empty_init=empty_init, device=device)
|
| 1144 |
+
|
| 1145 |
+
self.classifier_head = nn.Linear(config.hidden_size, config.num_labels, bias=True, dtype=torch.half)
|
| 1146 |
+
if config.classifier_dropout is not None:
|
| 1147 |
+
self.dropout = nn.Dropout(config.classifier_dropout)
|
| 1148 |
+
else:
|
| 1149 |
+
self.dropout = None
|
| 1150 |
+
self.config = config
|
| 1151 |
+
|
| 1152 |
+
def forward(
|
| 1153 |
+
self,
|
| 1154 |
+
input_ids: Optional[torch.LongTensor] = None,
|
| 1155 |
+
position_ids: Optional[torch.LongTensor] = None,
|
| 1156 |
+
attention_mask: Optional[torch.Tensor] = None,
|
| 1157 |
+
full_attention_mask: Optional[torch.Tensor] = None,
|
| 1158 |
+
past_key_values: Optional[Tuple[Tuple[torch.Tensor, torch.Tensor], ...]] = None,
|
| 1159 |
+
inputs_embeds: Optional[torch.LongTensor] = None,
|
| 1160 |
+
labels: Optional[torch.LongTensor] = None,
|
| 1161 |
+
use_cache: Optional[bool] = None,
|
| 1162 |
+
output_hidden_states: Optional[bool] = None,
|
| 1163 |
+
return_dict: Optional[bool] = None,
|
| 1164 |
+
) -> Union[Tuple[torch.Tensor, ...], SequenceClassifierOutputWithPast]:
|
| 1165 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
| 1166 |
+
|
| 1167 |
+
transformer_outputs = self.transformer(
|
| 1168 |
+
input_ids=input_ids,
|
| 1169 |
+
position_ids=position_ids,
|
| 1170 |
+
attention_mask=attention_mask,
|
| 1171 |
+
full_attention_mask=full_attention_mask,
|
| 1172 |
+
past_key_values=past_key_values,
|
| 1173 |
+
inputs_embeds=inputs_embeds,
|
| 1174 |
+
use_cache=use_cache,
|
| 1175 |
+
output_hidden_states=output_hidden_states,
|
| 1176 |
+
return_dict=return_dict,
|
| 1177 |
+
)
|
| 1178 |
+
|
| 1179 |
+
hidden_states = transformer_outputs[0]
|
| 1180 |
+
pooled_hidden_states = hidden_states[-1]
|
| 1181 |
+
if self.dropout is not None:
|
| 1182 |
+
pooled_hidden_states = self.dropout(pooled_hidden_states)
|
| 1183 |
+
logits = self.classifier_head(pooled_hidden_states)
|
| 1184 |
+
|
| 1185 |
+
loss = None
|
| 1186 |
+
if labels is not None:
|
| 1187 |
+
if self.config.problem_type is None:
|
| 1188 |
+
if self.num_labels == 1:
|
| 1189 |
+
self.config.problem_type = "regression"
|
| 1190 |
+
elif self.num_labels > 1 and (labels.dtype == torch.long or labels.dtype == torch.int):
|
| 1191 |
+
self.config.problem_type = "single_label_classification"
|
| 1192 |
+
else:
|
| 1193 |
+
self.config.problem_type = "multi_label_classification"
|
| 1194 |
+
|
| 1195 |
+
if self.config.problem_type == "regression":
|
| 1196 |
+
loss_fct = MSELoss()
|
| 1197 |
+
if self.num_labels == 1:
|
| 1198 |
+
loss = loss_fct(logits.squeeze().float(), labels.squeeze())
|
| 1199 |
+
else:
|
| 1200 |
+
loss = loss_fct(logits.float(), labels)
|
| 1201 |
+
elif self.config.problem_type == "single_label_classification":
|
| 1202 |
+
loss_fct = CrossEntropyLoss()
|
| 1203 |
+
loss = loss_fct(logits.view(-1, self.num_labels).float(), labels.view(-1))
|
| 1204 |
+
elif self.config.problem_type == "multi_label_classification":
|
| 1205 |
+
loss_fct = BCEWithLogitsLoss()
|
| 1206 |
+
loss = loss_fct(logits.float(), labels.view(-1, self.num_labels))
|
| 1207 |
+
|
| 1208 |
+
if not return_dict:
|
| 1209 |
+
output = (logits,) + transformer_outputs[1:]
|
| 1210 |
+
return ((loss,) + output) if loss is not None else output
|
| 1211 |
+
|
| 1212 |
+
return SequenceClassifierOutputWithPast(
|
| 1213 |
+
loss=loss,
|
| 1214 |
+
logits=logits,
|
| 1215 |
+
past_key_values=transformer_outputs.past_key_values,
|
| 1216 |
+
hidden_states=transformer_outputs.hidden_states,
|
| 1217 |
+
attentions=transformer_outputs.attentions,
|
| 1218 |
+
)
|
special_tokens_map.json
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"additional_special_tokens": [
|
| 3 |
+
"<|endoftext|>",
|
| 4 |
+
"[MASK]",
|
| 5 |
+
"[gMASK]",
|
| 6 |
+
"[sMASK]",
|
| 7 |
+
"<sop>",
|
| 8 |
+
"<eop>",
|
| 9 |
+
"<|system|>",
|
| 10 |
+
"<|user|>",
|
| 11 |
+
"<|assistant|>",
|
| 12 |
+
"<|observation|>",
|
| 13 |
+
"<|begin_of_image|>",
|
| 14 |
+
"<|end_of_image|>",
|
| 15 |
+
"<|begin_of_video|>",
|
| 16 |
+
"<|end_of_video|>"
|
| 17 |
+
],
|
| 18 |
+
"eos_token": {
|
| 19 |
+
"content": "<|endoftext|>",
|
| 20 |
+
"lstrip": false,
|
| 21 |
+
"normalized": false,
|
| 22 |
+
"rstrip": false,
|
| 23 |
+
"single_word": false
|
| 24 |
+
},
|
| 25 |
+
"pad_token": {
|
| 26 |
+
"content": "<|endoftext|>",
|
| 27 |
+
"lstrip": false,
|
| 28 |
+
"normalized": false,
|
| 29 |
+
"rstrip": false,
|
| 30 |
+
"single_word": false
|
| 31 |
+
}
|
| 32 |
+
}
|
tokenization_chatglm.py
ADDED
|
@@ -0,0 +1,323 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import regex as re
|
| 2 |
+
import base64
|
| 3 |
+
import os
|
| 4 |
+
import json
|
| 5 |
+
import tiktoken
|
| 6 |
+
from torch import TensorType
|
| 7 |
+
from typing import List, Optional, Union, Dict, Any
|
| 8 |
+
from transformers import PreTrainedTokenizer
|
| 9 |
+
from transformers.utils import logging, PaddingStrategy
|
| 10 |
+
from transformers.tokenization_utils_base import EncodedInput, BatchEncoding
|
| 11 |
+
|
| 12 |
+
|
| 13 |
+
class ChatGLM4Tokenizer(PreTrainedTokenizer):
|
| 14 |
+
vocab_files_names = {"vocab_file": "tokenizer.model"}
|
| 15 |
+
model_input_names = ["input_ids", "attention_mask", "position_ids"]
|
| 16 |
+
|
| 17 |
+
def __init__(
|
| 18 |
+
self,
|
| 19 |
+
vocab_file,
|
| 20 |
+
padding_side="left",
|
| 21 |
+
clean_up_tokenization_spaces=False,
|
| 22 |
+
encode_special_tokens=False,
|
| 23 |
+
**kwargs
|
| 24 |
+
):
|
| 25 |
+
self.name = "GLM4Tokenizer"
|
| 26 |
+
self.vocab_file = vocab_file
|
| 27 |
+
pat_str = "(?i:'s|'t|'re|'ve|'m|'ll|'d)|[^\\r\\n\\p{L}\\p{N}]?\\p{L}+|\\p{N}{1,3}| ?[^\\s\\p{L}\\p{N}]+[\\r\\n]*|\\s*[\\r\\n]+|\\s+(?!\\S)|\\s+"
|
| 28 |
+
self.pat_str = re.compile(pat_str)
|
| 29 |
+
self.encode_special_tokens = encode_special_tokens
|
| 30 |
+
|
| 31 |
+
mergeable_ranks = {}
|
| 32 |
+
with open(vocab_file) as f:
|
| 33 |
+
for line in f:
|
| 34 |
+
token, rank = line.strip().split()
|
| 35 |
+
rank = int(rank)
|
| 36 |
+
token = base64.b64decode(token)
|
| 37 |
+
mergeable_ranks[token] = rank
|
| 38 |
+
|
| 39 |
+
self.mergeable_ranks = mergeable_ranks
|
| 40 |
+
|
| 41 |
+
self.tokenizer = tiktoken.Encoding(
|
| 42 |
+
name="my_tokenizer",
|
| 43 |
+
pat_str=pat_str,
|
| 44 |
+
mergeable_ranks=mergeable_ranks,
|
| 45 |
+
special_tokens={}
|
| 46 |
+
)
|
| 47 |
+
self.decoder = {rank: token for token, rank in mergeable_ranks.items()}
|
| 48 |
+
self.n_words = len(self.decoder)
|
| 49 |
+
|
| 50 |
+
super().__init__(
|
| 51 |
+
padding_side=padding_side,
|
| 52 |
+
clean_up_tokenization_spaces=clean_up_tokenization_spaces,
|
| 53 |
+
**kwargs
|
| 54 |
+
)
|
| 55 |
+
|
| 56 |
+
@property
|
| 57 |
+
def vocab_size(self):
|
| 58 |
+
return self.n_words
|
| 59 |
+
|
| 60 |
+
def get_vocab(self):
|
| 61 |
+
""" Returns vocab as a dict """
|
| 62 |
+
vocab = {self._convert_id_to_token(i): i for i in range(self.vocab_size)}
|
| 63 |
+
vocab.update(self.added_tokens_encoder)
|
| 64 |
+
return vocab
|
| 65 |
+
|
| 66 |
+
def convert_tokens_to_string(self, tokens: List[Union[bytes, str]]) -> str:
|
| 67 |
+
"""
|
| 68 |
+
Converts a sequence of tokens in a single string.
|
| 69 |
+
"""
|
| 70 |
+
text = ""
|
| 71 |
+
temp = b""
|
| 72 |
+
for t in tokens:
|
| 73 |
+
if isinstance(t, str):
|
| 74 |
+
if temp:
|
| 75 |
+
text += temp.decode("utf-8", errors="replace")
|
| 76 |
+
temp = b""
|
| 77 |
+
text += t
|
| 78 |
+
elif isinstance(t, bytes):
|
| 79 |
+
temp += t
|
| 80 |
+
else:
|
| 81 |
+
raise TypeError("token should only be of type types or str")
|
| 82 |
+
if temp:
|
| 83 |
+
text += temp.decode("utf-8", errors="replace")
|
| 84 |
+
return text
|
| 85 |
+
|
| 86 |
+
def _tokenize(self, text, **kwargs):
|
| 87 |
+
tokens = []
|
| 88 |
+
ids = self.tokenizer.encode(text)
|
| 89 |
+
for t in ids:
|
| 90 |
+
tokens.append(self.decoder[t])
|
| 91 |
+
return tokens
|
| 92 |
+
|
| 93 |
+
def _convert_token_to_id(self, token):
|
| 94 |
+
""" Converts a token (str) in an id using the vocab. """
|
| 95 |
+
return self.mergeable_ranks[token]
|
| 96 |
+
|
| 97 |
+
def _convert_id_to_token(self, index):
|
| 98 |
+
"""Converts an index (integer) in a token (str) using the vocab."""
|
| 99 |
+
return self.decoder.get(index, "")
|
| 100 |
+
|
| 101 |
+
def save_vocabulary(self, save_directory, filename_prefix=None):
|
| 102 |
+
"""
|
| 103 |
+
Save the vocabulary and special tokens file to a directory.
|
| 104 |
+
|
| 105 |
+
Args:
|
| 106 |
+
save_directory (`str`):
|
| 107 |
+
The directory in which to save the vocabulary.
|
| 108 |
+
filename_prefix (`str`, *optional*):
|
| 109 |
+
An optional prefix to add to the named of the saved files.
|
| 110 |
+
|
| 111 |
+
Returns:
|
| 112 |
+
`Tuple(str)`: Paths to the files saved.
|
| 113 |
+
"""
|
| 114 |
+
if os.path.isdir(save_directory):
|
| 115 |
+
vocab_file = os.path.join(
|
| 116 |
+
save_directory, self.vocab_files_names["vocab_file"]
|
| 117 |
+
)
|
| 118 |
+
else:
|
| 119 |
+
vocab_file = save_directory
|
| 120 |
+
|
| 121 |
+
with open(self.vocab_file, 'rb') as fin:
|
| 122 |
+
proto_str = fin.read()
|
| 123 |
+
|
| 124 |
+
with open(vocab_file, "wb") as writer:
|
| 125 |
+
writer.write(proto_str)
|
| 126 |
+
|
| 127 |
+
return (vocab_file,)
|
| 128 |
+
|
| 129 |
+
def get_prefix_tokens(self):
|
| 130 |
+
prefix_tokens = [self.convert_tokens_to_ids("[gMASK]"), self.convert_tokens_to_ids("<sop>")]
|
| 131 |
+
return prefix_tokens
|
| 132 |
+
|
| 133 |
+
def build_single_message(self, role, metadata, message, tokenize=True):
|
| 134 |
+
assert role in ["system", "user", "assistant", "observation"], role
|
| 135 |
+
if tokenize:
|
| 136 |
+
role_tokens = [self.convert_tokens_to_ids(f"<|{role}|>")] + self.tokenizer.encode(f"{metadata}\n",
|
| 137 |
+
disallowed_special=())
|
| 138 |
+
message_tokens = self.tokenizer.encode(message, disallowed_special=())
|
| 139 |
+
tokens = role_tokens + message_tokens
|
| 140 |
+
return tokens
|
| 141 |
+
else:
|
| 142 |
+
return str(f"<|{role}|>{metadata}\n{message}")
|
| 143 |
+
|
| 144 |
+
def apply_chat_template(
|
| 145 |
+
self,
|
| 146 |
+
conversation: Union[List[Dict[str, str]], List[List[Dict[str, str]]], "Conversation"],
|
| 147 |
+
add_generation_prompt: bool = False,
|
| 148 |
+
tokenize: bool = True,
|
| 149 |
+
padding: bool = False,
|
| 150 |
+
truncation: bool = False,
|
| 151 |
+
max_length: Optional[int] = None,
|
| 152 |
+
return_tensors: Optional[Union[str, TensorType]] = None,
|
| 153 |
+
return_dict: bool = False,
|
| 154 |
+
tokenizer_kwargs: Optional[Dict[str, Any]] = None,
|
| 155 |
+
add_special_tokens: bool = True,
|
| 156 |
+
**kwargs,
|
| 157 |
+
) -> Union[str, List[int], List[str], List[List[int]], BatchEncoding]:
|
| 158 |
+
|
| 159 |
+
if return_dict and not tokenize:
|
| 160 |
+
raise ValueError(
|
| 161 |
+
"`return_dict=True` is incompatible with `tokenize=False`, because there is no dict "
|
| 162 |
+
"of tokenizer outputs to return."
|
| 163 |
+
)
|
| 164 |
+
|
| 165 |
+
def handle_single_conversation(conversation):
|
| 166 |
+
input_ids = self.get_prefix_tokens() if add_special_tokens else []
|
| 167 |
+
input_message = "[gMASK]<sop>" if add_special_tokens else ""
|
| 168 |
+
for item in conversation:
|
| 169 |
+
if item.get("tools"):
|
| 170 |
+
tools = item["tools"]
|
| 171 |
+
content = "你是一个名为 GLM-4 的人工智能助手。你是基于智谱AI训练的语言模型 GLM-4 模型开发的，你的任务是针对用户的问题和要求提供适当的答复和支持。"
|
| 172 |
+
for tool in tools:
|
| 173 |
+
if tool["type"] == "function":
|
| 174 |
+
function = tool["function"]
|
| 175 |
+
content += f"\n\n## {function['name']}\n\n{json.dumps(function, ensure_ascii=False, indent=4)}"
|
| 176 |
+
content += "\n在调用上述函数时，请使用 Json 格式表示调用的参数。"
|
| 177 |
+
elif tool["type"] == "python":
|
| 178 |
+
content += "\n\n## python\n\n当你向 `python` 发送包含 Python 代码的消息时，该代码将会在一个有状态的 Jupyter notebook 环境中执行。\n`python` 返回代码执行的输出，或在执行 60 秒后返回超时。\n`/mnt/data` 将会持久化存储你的文件。在此会话中，`python` 无法访问互联网。不要使用 `python` 进行任何网络请求或者在线 API 调用，这些在线内容的访问将不会成功。"
|
| 179 |
+
elif tool["type"] == "simple_browser":
|
| 180 |
+
content += "\n\n## simple_browser\n\n你可以使用 `simple_browser` 工具。该工具支持以下函数：\n`search(query: str, recency_days: int)`：使用搜索引擎进行查询并显示结果，可以使用 `recency_days` 参数控制搜索内容的时效性。\n`mclick(ids: list[int])`：获取一系列指定 id 的页面内容。每次调用时，须选择3-10个页面。选择多个角度的页面，同时尽可能选择可信任的信息来源。考虑到部分页面是无法加载的，你也可以多打开一些可能有用的页面而不用担心内容过多。\n`open_url(url: str)`：打开指定的 URL。\n\n使用 `【{引用 id}†{引用文本}】` 来引用内容。\n\n操作步骤：1. 使用 `search` 来获得信息列表; 2. 使用 `mclick` 来获取指定 ID 页面的内容; 3. 根据获得的内容进行回复。在回复中应当引用信息来源。\n 如果用户提供了 URL，也可以用 `open_url` 直接打开页面。\n如果初次搜索结果没有找到合适的信息，也可以再次使用 `search` 进行搜索。"
|
| 181 |
+
elif tool["type"] == "cogview":
|
| 182 |
+
content += "\n\n## cogview\n\n如果用户的请求中包含了对图像的描述，你可以使用 `cogview` 来生成图像并展示给用户。你需要向 `cogview` 发送图像描述，规则：\n- 发送给 `cogview` 的消息必须使用英语。用户的中文描述必须完全翻译为英语。\n- 应当尽可能详细地描述图像生成的需求，需求描述约 100 英文单词。\n- 保持用户原始描述的意图。不要虚构内容或者没见过的人物。\n- 如无特殊说明，所在地为中国，持有中国立场并遵循中国社会主义价值观。"
|
| 183 |
+
else:
|
| 184 |
+
raise NotImplementedError(f"Unknown tool type {tool['type']}")
|
| 185 |
+
input = self.build_single_message("system", "", content, tokenize=tokenize)
|
| 186 |
+
if tokenize:
|
| 187 |
+
input_ids.extend(input)
|
| 188 |
+
else:
|
| 189 |
+
input_message += input
|
| 190 |
+
if item["content"]:
|
| 191 |
+
input = self.build_single_message(
|
| 192 |
+
item["role"],
|
| 193 |
+
item.get("metadata", ""),
|
| 194 |
+
item["content"],
|
| 195 |
+
tokenize=tokenize
|
| 196 |
+
)
|
| 197 |
+
if tokenize:
|
| 198 |
+
input_ids.extend(input)
|
| 199 |
+
else:
|
| 200 |
+
input_message += input
|
| 201 |
+
if add_generation_prompt:
|
| 202 |
+
if tokenize:
|
| 203 |
+
input_ids.extend([self.convert_tokens_to_ids("<|assistant|>")])
|
| 204 |
+
else:
|
| 205 |
+
input_message += "<|assistant|>"
|
| 206 |
+
|
| 207 |
+
return input_ids if tokenize else input_message
|
| 208 |
+
|
| 209 |
+
# Main logic to handle different conversation formats
|
| 210 |
+
if isinstance(conversation, list) and all(isinstance(i, dict) for i in conversation):
|
| 211 |
+
result = handle_single_conversation(conversation)
|
| 212 |
+
elif isinstance(conversation, list) and all(isinstance(i, list) for i in conversation):
|
| 213 |
+
result = [handle_single_conversation(c) for c in conversation]
|
| 214 |
+
elif hasattr(conversation, "messages"):
|
| 215 |
+
result = handle_single_conversation(conversation.messages)
|
| 216 |
+
else:
|
| 217 |
+
raise ValueError("Invalid conversation format")
|
| 218 |
+
|
| 219 |
+
if tokenize:
|
| 220 |
+
output = self.batch_encode_plus(
|
| 221 |
+
[result] if isinstance(result[0], int) else result,
|
| 222 |
+
padding=padding,
|
| 223 |
+
truncation=truncation,
|
| 224 |
+
max_length=max_length,
|
| 225 |
+
return_tensors=return_tensors,
|
| 226 |
+
is_split_into_words=True,
|
| 227 |
+
add_special_tokens=False
|
| 228 |
+
)
|
| 229 |
+
if return_dict:
|
| 230 |
+
return output
|
| 231 |
+
else:
|
| 232 |
+
return output["input_ids"]
|
| 233 |
+
else:
|
| 234 |
+
return result
|
| 235 |
+
|
| 236 |
+
|
| 237 |
+
def build_inputs_with_special_tokens(
|
| 238 |
+
self, token_ids_0: List[int], token_ids_1: Optional[List[int]] = None
|
| 239 |
+
) -> List[int]:
|
| 240 |
+
"""
|
| 241 |
+
Build model inputs from a sequence or a pair of sequence for sequence classification tasks by concatenating and
|
| 242 |
+
adding special tokens. A BERT sequence has the following format:
|
| 243 |
+
|
| 244 |
+
- single sequence: `[CLS] X [SEP]`
|
| 245 |
+
- pair of sequences: `[CLS] A [SEP] B [SEP]`
|
| 246 |
+
|
| 247 |
+
Args:
|
| 248 |
+
token_ids_0 (`List[int]`):
|
| 249 |
+
List of IDs to which the special tokens will be added.
|
| 250 |
+
token_ids_1 (`List[int]`, *optional*):
|
| 251 |
+
Optional second list of IDs for sequence pairs.
|
| 252 |
+
|
| 253 |
+
Returns:
|
| 254 |
+
`List[int]`: List of [input IDs](../glossary#input-ids) with the appropriate special tokens.
|
| 255 |
+
"""
|
| 256 |
+
prefix_tokens = self.get_prefix_tokens()
|
| 257 |
+
token_ids_0 = prefix_tokens + token_ids_0
|
| 258 |
+
if token_ids_1 is not None:
|
| 259 |
+
token_ids_0 = token_ids_0 + token_ids_1 + [self.convert_tokens_to_ids("<eos>")]
|
| 260 |
+
return token_ids_0
|
| 261 |
+
|
| 262 |
+
def _pad(
|
| 263 |
+
self,
|
| 264 |
+
encoded_inputs: Union[Dict[str, EncodedInput], BatchEncoding],
|
| 265 |
+
max_length: Optional[int] = None,
|
| 266 |
+
padding_strategy: PaddingStrategy = PaddingStrategy.DO_NOT_PAD,
|
| 267 |
+
pad_to_multiple_of: Optional[int] = None,
|
| 268 |
+
return_attention_mask: Optional[bool] = None,
|
| 269 |
+
) -> dict:
|
| 270 |
+
"""
|
| 271 |
+
Pad encoded inputs (on left/right and up to predefined length or max length in the batch)
|
| 272 |
+
|
| 273 |
+
Args:
|
| 274 |
+
encoded_inputs:
|
| 275 |
+
Dictionary of tokenized inputs (`List[int]`) or batch of tokenized inputs (`List[List[int]]`).
|
| 276 |
+
max_length: maximum length of the returned list and optionally padding length (see below).
|
| 277 |
+
Will truncate by taking into account the special tokens.
|
| 278 |
+
padding_strategy: PaddingStrategy to use for padding.
|
| 279 |
+
|
| 280 |
+
- PaddingStrategy.LONGEST Pad to the longest sequence in the batch
|
| 281 |
+
- PaddingStrategy.MAX_LENGTH: Pad to the max length (default)
|
| 282 |
+
- PaddingStrategy.DO_NOT_PAD: Do not pad
|
| 283 |
+
The tokenizer padding sides are defined in self.padding_side:
|
| 284 |
+
|
| 285 |
+
- 'left': pads on the left of the sequences
|
| 286 |
+
- 'right': pads on the right of the sequences
|
| 287 |
+
pad_to_multiple_of: (optional) Integer if set will pad the sequence to a multiple of the provided value.
|
| 288 |
+
This is especially useful to enable the use of Tensor Core on NVIDIA hardware with compute capability
|
| 289 |
+
`>= 7.5` (Volta).
|
| 290 |
+
return_attention_mask:
|
| 291 |
+
(optional) Set to False to avoid returning attention mask (default: set to model specifics)
|
| 292 |
+
"""
|
| 293 |
+
# Load from model defaults
|
| 294 |
+
assert self.padding_side == "left"
|
| 295 |
+
|
| 296 |
+
required_input = encoded_inputs[self.model_input_names[0]]
|
| 297 |
+
seq_length = len(required_input)
|
| 298 |
+
|
| 299 |
+
if padding_strategy == PaddingStrategy.LONGEST:
|
| 300 |
+
max_length = len(required_input)
|
| 301 |
+
|
| 302 |
+
if max_length is not None and pad_to_multiple_of is not None and (max_length % pad_to_multiple_of != 0):
|
| 303 |
+
max_length = ((max_length // pad_to_multiple_of) + 1) * pad_to_multiple_of
|
| 304 |
+
|
| 305 |
+
needs_to_be_padded = padding_strategy != PaddingStrategy.DO_NOT_PAD and len(required_input) != max_length
|
| 306 |
+
|
| 307 |
+
# Initialize attention mask if not present.
|
| 308 |
+
if "attention_mask" not in encoded_inputs:
|
| 309 |
+
encoded_inputs["attention_mask"] = [1] * seq_length
|
| 310 |
+
|
| 311 |
+
if "position_ids" not in encoded_inputs:
|
| 312 |
+
encoded_inputs["position_ids"] = list(range(seq_length))
|
| 313 |
+
|
| 314 |
+
if needs_to_be_padded:
|
| 315 |
+
difference = max_length - len(required_input)
|
| 316 |
+
|
| 317 |
+
if "attention_mask" in encoded_inputs:
|
| 318 |
+
encoded_inputs["attention_mask"] = [0] * difference + encoded_inputs["attention_mask"]
|
| 319 |
+
if "position_ids" in encoded_inputs:
|
| 320 |
+
encoded_inputs["position_ids"] = [0] * difference + encoded_inputs["position_ids"]
|
| 321 |
+
encoded_inputs[self.model_input_names[0]] = [self.pad_token_id] * difference + required_input
|
| 322 |
+
|
| 323 |
+
return encoded_inputs
|
tokenizer.model
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5a493598071550244b2ee7f26118f3edec2150b9dfa967929a99052ac83fe716
|
| 3 |
+
size 2623634
|
tokenizer_config.json
ADDED
|
@@ -0,0 +1,146 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"added_tokens_decoder": {
|
| 3 |
+
"151329": {
|
| 4 |
+
"content": "<|endoftext|>",
|
| 5 |
+
"lstrip": false,
|
| 6 |
+
"normalized": false,
|
| 7 |
+
"rstrip": false,
|
| 8 |
+
"single_word": false,
|
| 9 |
+
"special": true
|
| 10 |
+
},
|
| 11 |
+
"151330": {
|
| 12 |
+
"content": "[MASK]",
|
| 13 |
+
"lstrip": false,
|
| 14 |
+
"normalized": false,
|
| 15 |
+
"rstrip": false,
|
| 16 |
+
"single_word": false,
|
| 17 |
+
"special": true
|
| 18 |
+
},
|
| 19 |
+
"151331": {
|
| 20 |
+
"content": "[gMASK]",
|
| 21 |
+
"lstrip": false,
|
| 22 |
+
"normalized": false,
|
| 23 |
+
"rstrip": false,
|
| 24 |
+
"single_word": false,
|
| 25 |
+
"special": true
|
| 26 |
+
},
|
| 27 |
+
"151332": {
|
| 28 |
+
"content": "[sMASK]",
|
| 29 |
+
"lstrip": false,
|
| 30 |
+
"normalized": false,
|
| 31 |
+
"rstrip": false,
|
| 32 |
+
"single_word": false,
|
| 33 |
+
"special": true
|
| 34 |
+
},
|
| 35 |
+
"151333": {
|
| 36 |
+
"content": "<sop>",
|
| 37 |
+
"lstrip": false,
|
| 38 |
+
"normalized": false,
|
| 39 |
+
"rstrip": false,
|
| 40 |
+
"single_word": false,
|
| 41 |
+
"special": true
|
| 42 |
+
},
|
| 43 |
+
"151334": {
|
| 44 |
+
"content": "<eop>",
|
| 45 |
+
"lstrip": false,
|
| 46 |
+
"normalized": false,
|
| 47 |
+
"rstrip": false,
|
| 48 |
+
"single_word": false,
|
| 49 |
+
"special": true
|
| 50 |
+
},
|
| 51 |
+
"151335": {
|
| 52 |
+
"content": "<|system|>",
|
| 53 |
+
"lstrip": false,
|
| 54 |
+
"normalized": false,
|
| 55 |
+
"rstrip": false,
|
| 56 |
+
"single_word": false,
|
| 57 |
+
"special": true
|
| 58 |
+
},
|
| 59 |
+
"151336": {
|
| 60 |
+
"content": "<|user|>",
|
| 61 |
+
"lstrip": false,
|
| 62 |
+
"normalized": false,
|
| 63 |
+
"rstrip": false,
|
| 64 |
+
"single_word": false,
|
| 65 |
+
"special": true
|
| 66 |
+
},
|
| 67 |
+
"151337": {
|
| 68 |
+
"content": "<|assistant|>",
|
| 69 |
+
"lstrip": false,
|
| 70 |
+
"normalized": false,
|
| 71 |
+
"rstrip": false,
|
| 72 |
+
"single_word": false,
|
| 73 |
+
"special": true
|
| 74 |
+
},
|
| 75 |
+
"151338": {
|
| 76 |
+
"content": "<|observation|>",
|
| 77 |
+
"lstrip": false,
|
| 78 |
+
"normalized": false,
|
| 79 |
+
"rstrip": false,
|
| 80 |
+
"single_word": false,
|
| 81 |
+
"special": true
|
| 82 |
+
},
|
| 83 |
+
"151339": {
|
| 84 |
+
"content": "<|begin_of_image|>",
|
| 85 |
+
"lstrip": false,
|
| 86 |
+
"normalized": false,
|
| 87 |
+
"rstrip": false,
|
| 88 |
+
"single_word": false,
|
| 89 |
+
"special": true
|
| 90 |
+
},
|
| 91 |
+
"151340": {
|
| 92 |
+
"content": "<|end_of_image|>",
|
| 93 |
+
"lstrip": false,
|
| 94 |
+
"normalized": false,
|
| 95 |
+
"rstrip": false,
|
| 96 |
+
"single_word": false,
|
| 97 |
+
"special": true
|
| 98 |
+
},
|
| 99 |
+
"151341": {
|
| 100 |
+
"content": "<|begin_of_video|>",
|
| 101 |
+
"lstrip": false,
|
| 102 |
+
"normalized": false,
|
| 103 |
+
"rstrip": false,
|
| 104 |
+
"single_word": false,
|
| 105 |
+
"special": true
|
| 106 |
+
},
|
| 107 |
+
"151342": {
|
| 108 |
+
"content": "<|end_of_video|>",
|
| 109 |
+
"lstrip": false,
|
| 110 |
+
"normalized": false,
|
| 111 |
+
"rstrip": false,
|
| 112 |
+
"single_word": false,
|
| 113 |
+
"special": true
|
| 114 |
+
}
|
| 115 |
+
},
|
| 116 |
+
"additional_special_tokens": [
|
| 117 |
+
"<|endoftext|>",
|
| 118 |
+
"[MASK]",
|
| 119 |
+
"[gMASK]",
|
| 120 |
+
"[sMASK]",
|
| 121 |
+
"<sop>",
|
| 122 |
+
"<eop>",
|
| 123 |
+
"<|system|>",
|
| 124 |
+
"<|user|>",
|
| 125 |
+
"<|assistant|>",
|
| 126 |
+
"<|observation|>",
|
| 127 |
+
"<|begin_of_image|>",
|
| 128 |
+
"<|end_of_image|>",
|
| 129 |
+
"<|begin_of_video|>",
|
| 130 |
+
"<|end_of_video|>"
|
| 131 |
+
],
|
| 132 |
+
"auto_map": {
|
| 133 |
+
"AutoTokenizer": [
|
| 134 |
+
"tokenization_chatglm.ChatGLM4Tokenizer",
|
| 135 |
+
null
|
| 136 |
+
]
|
| 137 |
+
},
|
| 138 |
+
"clean_up_tokenization_spaces": false,
|
| 139 |
+
"do_lower_case": false,
|
| 140 |
+
"eos_token": "<|endoftext|>",
|
| 141 |
+
"model_max_length": 128000,
|
| 142 |
+
"pad_token": "<|endoftext|>",
|
| 143 |
+
"padding_side": "left",
|
| 144 |
+
"remove_space": false,
|
| 145 |
+
"tokenizer_class": "ChatGLM4Tokenizer"
|
| 146 |
+
}
|