Update README.md
Browse files
README.md
CHANGED
@@ -1,3 +1,102 @@
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: apache-2.0
|
3 |
---
|
4 |
+
|
5 |
+
|
6 |
+
## Description
|
7 |
+
This repo contains GGUF format model files for NeuralDarewin-7B.
|
8 |
+
|
9 |
+
## Files Provided
|
10 |
+
| Name | Quant | Bits | File Size | Remark |
|
11 |
+
| ---------------------------- | ------- | ---- | --------- | -------------------------------- |
|
12 |
+
| neuraldarewin-7b.IQ3_XXS.gguf | IQ3_XXS | 3 | 3.02 GB | 3.06 bpw quantization |
|
13 |
+
| neuraldarewin-7b.IQ3_S.gguf | IQ3_S | 3 | 3.18 GB | 3.44 bpw quantization |
|
14 |
+
| neuraldarewin-7b.IQ3_M.gguf | IQ3_M | 3 | 3.28 GB | 3.66 bpw quantization mix |
|
15 |
+
| neuraldarewin-7b.Q4_0.gguf | Q4_0 | 4 | 4.11 GB | 3.56G, +0.2166 ppl |
|
16 |
+
| neuraldarewin-7b.IQ4_NL.gguf | IQ4_NL | 4 | 4.16 GB | 4.25 bpw non-linear quantization |
|
17 |
+
| neuraldarewin-7b.Q4_K_M.gguf | Q4_K_M | 4 | 4.37 GB | 3.80G, +0.0532 ppl |
|
18 |
+
| neuraldarewin-7b.Q5_K_M.gguf | Q5_K_M | 5 | 5.13 GB | 4.45G, +0.0122 ppl |
|
19 |
+
| neuraldarewin-7b.Q6_K.gguf | Q6_K | 6 | 5.94 GB | 5.15G, +0.0008 ppl |
|
20 |
+
| neuraldarewin-7b.Q8_0.gguf | Q8_0 | 8 | 7.70 GB | 6.70G, +0.0004 ppl |
|
21 |
+
|
22 |
+
## Parameters
|
23 |
+
| path | type | architecture | rope_theta | sliding_win | max_pos_embed |
|
24 |
+
| ---------------------------- | ------- | ------------------ | ---------- | ----------- | ------------- |
|
25 |
+
| mlabonne/Darewin-7B | mistral | MistralForCausalLM | 10000.0 | 4096 | 32768 |
|
26 |
+
|
27 |
+
## Benchmarks
|
28 |
+
![](https://i.ibb.co/gjKpkcj/Neural-Darewin-7-B-GGUF.png)
|
29 |
+
|
30 |
+
|
31 |
+
# Original Model Card
|
32 |
+
|
33 |
+
Darewin-7B is a merge of the following models using [LazyMergekit](https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb?usp=sharing):
|
34 |
+
* [Intel/neural-chat-7b-v3-3](https://huggingface.co/Intel/neural-chat-7b-v3-3)
|
35 |
+
* [openaccess-ai-collective/DPOpenHermes-7B-v2](https://huggingface.co/openaccess-ai-collective/DPOpenHermes-7B-v2)
|
36 |
+
* [fblgit/una-cybertron-7b-v2-bf16](https://huggingface.co/fblgit/una-cybertron-7b-v2-bf16)
|
37 |
+
* [openchat/openchat-3.5-0106](https://huggingface.co/openchat/openchat-3.5-0106)
|
38 |
+
* [OpenPipe/mistral-ft-optimized-1227](https://huggingface.co/OpenPipe/mistral-ft-optimized-1227)
|
39 |
+
* [mlabonne/NeuralHermes-2.5-Mistral-7B](https://huggingface.co/mlabonne/NeuralHermes-2.5-Mistral-7B)
|
40 |
+
|
41 |
+
## 🧩 Configuration
|
42 |
+
|
43 |
+
```yaml
|
44 |
+
models:
|
45 |
+
- model: mistralai/Mistral-7B-v0.1
|
46 |
+
# No parameters necessary for base model
|
47 |
+
- model: Intel/neural-chat-7b-v3-3
|
48 |
+
parameters:
|
49 |
+
density: 0.6
|
50 |
+
weight: 0.2
|
51 |
+
- model: openaccess-ai-collective/DPOpenHermes-7B-v2
|
52 |
+
parameters:
|
53 |
+
density: 0.6
|
54 |
+
weight: 0.1
|
55 |
+
- model: fblgit/una-cybertron-7b-v2-bf16
|
56 |
+
parameters:
|
57 |
+
density: 0.6
|
58 |
+
weight: 0.2
|
59 |
+
- model: openchat/openchat-3.5-0106
|
60 |
+
parameters:
|
61 |
+
density: 0.6
|
62 |
+
weight: 0.15
|
63 |
+
- model: OpenPipe/mistral-ft-optimized-1227
|
64 |
+
parameters:
|
65 |
+
density: 0.6
|
66 |
+
weight: 0.25
|
67 |
+
- model: mlabonne/NeuralHermes-2.5-Mistral-7B
|
68 |
+
parameters:
|
69 |
+
density: 0.6
|
70 |
+
weight: 0.1
|
71 |
+
merge_method: dare_ties
|
72 |
+
base_model: mistralai/Mistral-7B-v0.1
|
73 |
+
parameters:
|
74 |
+
int8_mask: true
|
75 |
+
dtype: bfloat16
|
76 |
+
|
77 |
+
```
|
78 |
+
|
79 |
+
## 💻 Usage
|
80 |
+
|
81 |
+
```python
|
82 |
+
!pip install -qU transformers accelerate
|
83 |
+
|
84 |
+
from transformers import AutoTokenizer
|
85 |
+
import transformers
|
86 |
+
import torch
|
87 |
+
|
88 |
+
model = "mlabonne/NeuralDarewin-7B"
|
89 |
+
messages = [{"role": "user", "content": "What is a large language model?"}]
|
90 |
+
|
91 |
+
tokenizer = AutoTokenizer.from_pretrained(model)
|
92 |
+
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
|
93 |
+
pipeline = transformers.pipeline(
|
94 |
+
"text-generation",
|
95 |
+
model=model,
|
96 |
+
torch_dtype=torch.float16,
|
97 |
+
device_map="auto",
|
98 |
+
)
|
99 |
+
|
100 |
+
outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
|
101 |
+
print(outputs[0]["generated_text"])
|
102 |
+
```
|