NSFW 13B sft - GGUF
- Model creator: zxbsmk
- Original model: NSFW 13B sft
Description
This repo contains GGUF format model files for zxbsmk's NSFW 13B sft.
These files were quantised using hardware kindly provided by Google Colab(Free CPU Machine).
You can also check it out easily in my GitHub repo.
About GGUF
GGUF is a new format introduced by the llama.cpp team on August 21st 2023. It is a replacement for GGML, which is no longer supported by llama.cpp.
Here is an incomplate list of clients and libraries that are known to support GGUF:
- llama.cpp. The source project for GGUF. Offers a CLI and a server option.
- text-generation-webui, the most widely used web UI, with many features and powerful extensions. Supports GPU acceleration.
- KoboldCpp, a fully featured web UI, with GPU accel across all platforms and GPU architectures. Especially good for story telling.
- LM Studio, an easy-to-use and powerful local GUI for Windows and macOS (Silicon), with GPU acceleration.
- LoLLMS Web UI, a great web UI with many interesting and unique features, including a full model library for easy model selection.
- Faraday.dev, an attractive and easy to use character-based chat GUI for Windows and macOS (both Silicon and Intel), with GPU acceleration.
- ctransformers, a Python library with GPU accel, LangChain support, and OpenAI-compatible AI server.
- llama-cpp-python, a Python library with GPU accel, LangChain support, and OpenAI-compatible API server.
- candle, a Rust ML framework with a focus on performance, including GPU support, and ease of use.
- Nitro, a fast, lightweight 3mb inference server to supercharge apps with local AI, and OpenAI-compatible API server.
Repositories available
- 2, 3, 4, 5, 6, 8, 16 and 32-bit GGUF models for CPU+GPU inference
- zxbsmk's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions
Prompt template: BLING
System: A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human's questions.
Human: {prompt}
Assistant:
Compatibility
These quantised GGUFv2 files are compatible with llama.cpp from August 27th onwards, as of commit d0cee0d
They are also compatible with many third party UIs and libraries - please see the list at the top of this README.
Explanation of quantisation methods
Click to see details
The new methods available are:
- GGML_TYPE_Q2_K - "type-1" 2-bit quantization in super-blocks containing 16 blocks, each block having 16 weight. Block scales and mins are quantized with 4 bits. This ends up effectively using 2.5625 bits per weight (bpw)
- GGML_TYPE_Q3_K - "type-0" 3-bit quantization in super-blocks containing 16 blocks, each block having 16 weights. Scales are quantized with 6 bits. This end up using 3.4375 bpw.
- GGML_TYPE_Q4_K - "type-1" 4-bit quantization in super-blocks containing 8 blocks, each block having 32 weights. Scales and mins are quantized with 6 bits. This ends up using 4.5 bpw.
- GGML_TYPE_Q5_K - "type-1" 5-bit quantization. Same super-block structure as GGML_TYPE_Q4_K resulting in 5.5 bpw
- GGML_TYPE_Q6_K - "type-0" 6-bit quantization. Super-blocks with 16 blocks, each block having 16 weights. Scales are quantized with 8 bits. This ends up using 6.5625 bpw
Refer to the Provided Files table below to see what files use which methods, and how.
Provided files
Name | Quant method | Bits | Size | Max RAM required | Use case |
---|---|---|---|---|---|
NSFW_13B_sft.Q2_K.gguf | Q2_K | 2 | 5.56 GB | untested yet | smallest, significant quality loss - not recommended for most purposes |
NSFW_13B_sft.Q3_K_S.gguf | Q3_K_S | 3 | 6.38 GB | untested yet | very small, high quality loss |
NSFW_13B_sft.Q3_K_M.gguf | Q3_K_M | 3 | 6.85 GB | untested yet | very small, high quality loss |
NSFW_13B_sft.Q3_K_L.gguf | Q3_K_L | 3 | 7.27 GB | untested yet | small, substantial quality loss |
NSFW_13B_sft.Q4_0.gguf | Q4_0 | 4 | 7.55 GB | untested yet | legacy; small, very high quality loss - prefer using Q3_K_M |
NSFW_13B_sft.Q4_K_S.gguf | Q4_K_S | 4 | 7.93 GB | untested yet | small, greater quality loss |
NSFW_13B_sft.Q4_K_M.gguf | Q4_K_M | 4 | 8.56 GB | untested yet | medium, balanced quality - recommended |
NSFW_13B_sft.Q5_0.gguf | Q5_0 | 5 | 9.17 GB | untested yet | legacy; medium, balanced quality - prefer using Q4_K_M |
NSFW_13B_sft.Q5_K_S.gguf | Q5_K_S | 5 | 9.34 GB | untested yet | large, low quality loss - recommended |
NSFW_13B_sft.Q5_K_M.gguf | Q5_K_M | 5 | 9.85 GB | untested yet | large, very low quality loss - recommended |
NSFW_13B_sft.Q6_K.gguf | Q6_K | 6 | 11.6 GB | untested yet | very large, extremely low quality loss |
NSFW_13B_sft.Q8_0.gguf | Q8_0 | 8 | 14.1 GB | untested yet | very large, extremely low quality loss - not recommended |
NSFW_13B_sft.F16.gguf | F16 | 16 | 26.5 GB | untested yet | extremely large, extremely low quality loss - not recommended |
Note: the above RAM figures assume no GPU offloading. If layers are offloaded to the GPU, this will reduce RAM usage and use VRAM instead.
How to download GGUF files
Note for manual downloaders: You almost never want to clone the entire repo! Multiple different quantisation formats are provided, and most users only want to pick and download a single file.
The following clients/libraries will automatically download models for you, providing a list of available models to choose from:
- LM Studio
- LoLLMS Web UI
- Faraday.dev
In text-generation-webui
Under Download Model, you can enter the model repo: mzwing/NSFW_13B_sft-GGUF
, and below it, a specific filename to download, such as: NSFW_13B_sft.Q4_K_M.gguf
.
Then click Download.
On the command line, including multiple files at once
I recommend using the huggingface-hub
Python library:
pip3 install huggingface-hub
Then you can download any individual model file to the current directory, at high speed, with a command like this:
huggingface-cli download mzwing/NSFW_13B_sft-GGUF NSFW_13B_sft.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
More advanced huggingface-cli download usage
You can also download multiple files at once with a pattern:
huggingface-cli download mzwing/NSFW_13B_sft-GGUF --local-dir . --local-dir-use-symlinks False --include='*Q4_K*gguf'
For more documentation on downloading with huggingface-cli
, please see: HF -> Hub Python Library -> Download files -> Download from the CLI.
To accelerate downloads on fast connections (1Gbit/s or higher), install hf_transfer
:
pip3 install hf_transfer
And set environment variable HF_HUB_ENABLE_HF_TRANSFER
to 1
:
HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download mzwing/NSFW_13B_sft-GGUF NSFW_13B_sft.Q4_K_M.gguf --local-dir . --local-dir-use-symlinks False
Windows Command Line users: You can set the environment variable by running set HF_HUB_ENABLE_HF_TRANSFER=1
before the download command.
Example llama.cpp
command
Make sure you are using llama.cpp
from commit d0cee0d or later.
./main -ngl 32 -m NSFW_13B_sft.Q4_K_M.gguf --color -c 2048 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "System: A chat between a curious human and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the human's questions.\nHuman: {prompt}\nAssistant:"
Change -ngl 32
to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration.
Change -c 2048
to the desired sequence length. For extended sequence models - eg 8K, 16K, 32K - the necessary RoPE scaling parameters are read from the GGUF file and set by llama.cpp automatically.
If you want to have a chat-style conversation, replace the -p <PROMPT>
argument with -i -ins
For other parameters and how to use them, please refer to the llama.cpp documentation
How to run in text-generation-webui
Further instructions here: text-generation-webui/docs/llama.cpp.md.
How to run from Python code
You can use GGUF models from Python using the llama-cpp-python or ctransformers libraries.
How to load this model in Python code, using ctransformers
First install the package
Run one of the following commands, according to your system:
# Base ctransformers with no GPU acceleration
pip install ctransformers
# Or with CUDA GPU acceleration
pip install ctransformers[cuda]
# Or with AMD ROCm GPU acceleration (Linux only)
CT_HIPBLAS=1 pip install ctransformers --no-binary ctransformers
# Or with Metal GPU acceleration for macOS systems only
CT_METAL=1 pip install ctransformers --no-binary ctransformers
Simple ctransformers example code
from ctransformers import AutoModelForCausalLM
# Set gpu_layers to the number of layers to offload to GPU. Set to 0 if no GPU acceleration is available on your system.
llm = AutoModelForCausalLM.from_pretrained("mzwing/NSFW_13B_sft-GGUF", model_file="NSFW_13B_sft.Q4_K_M.gguf", model_type="phi", gpu_layers=50)
print(llm("AI is going to"))
How to use with LangChain
Here are guides on using llama-cpp-python and ctransformers with LangChain:
Thanks, and how to contribute
Thanks to Google Colab! All the quantised models in this repo are done on the awesome platform. Thanks a lot!
Thanks to llama.cpp! It inspired me to explore the inspiring AI field, thanks!
Thanks to TheBloke! Everything in this repo is a reference to him.
You are welcome to create a PullRequest! Especially for the RAM Usage!
A instruction-tuned model of https://huggingface.co/baichuan-inc/Baichuan-13B-Base
- Instruction-following datasets used: instruct_nsfw_cn
- Training framework: https://github.com/hiyouga/LLaMA-Efficient-Tuning
Usage:
from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer
tokenizer = AutoTokenizer.from_pretrained("zxbsmk/NSFW_13B_sft", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("zxbsmk/NSFW_13B_sft", trust_remote_code=True).cuda()
streamer = TextStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
query = "续写。白洁已经动情了,肩带滑落露出香肩。"
inputs = tokenizer([query], return_tensors="pt")
inputs = inputs.to("cuda")
generate_ids = model.generate(**inputs, max_new_tokens=1024, streamer=streamer)
You could also alternatively launch a CLI demo by using the script in https://github.com/hiyouga/LLaMA-Efficient-Tuning
python src/cli_demo.py --template baichuan --model_name_or_path zxbsmk/NSFW_13B_sft
Examples:
Omit part of results for better visualization.
Question | NSFW_13B_sft | ChatGPT |
---|---|---|
中国第三代战斗机的代表是什么? | 中国第三代战斗机的代表是歼-10战斗机。 | 截至我所掌握的信息(截止2021年9月),中国第三代战斗机的代表是中国航空工业集团公司(AVIC)生产的"歼-20"战斗机。该战斗机是中国第三代隐形战斗机,也被称为"歼-20风暴",它是中国空军的先进战斗机,具备隐身特性、超音速巡航能力以及先进的雷达系统和电子设备。歼-20的研发和部署标志着中国在军事航空技术方面取得了重要进展。请注意,自那之后,可能已经有更多新型战斗机被研发或部署,因此建议查阅最新资料以获取更准确和最新的信息。 |
解释这句话中每个词语的含义:今天下雨,我骑车差点摔倒,好在我一把把把把住了。 | 这句话中,“今天”指的是当前的时间,即雨天。“下雨”指的是天气阴沉,雨水落下。“我”指的是说话者,即骑车人。“骑车”指的是用自行车或其他交通工具出行。“差点”指的是几乎,几乎要摔倒。“摔倒”指的是身体失去平衡,摔在地上。“好在我”指的是幸亏有。“一把”指的是一下子。“把”指的是握住。“把住了”指的是握住车把,保持平衡。 | 今天:指代当天,即当前的日期。下雨:指天气现象,指雨水从天空降落。我:指说话人自己,第一人称代词。骑车:动词短语,表示骑着自行车。差点:几乎、险些,表示差一点就发生某种情况。摔倒:动词,意为因失去平衡而跌倒。好在:幸好、庆幸的意思,用来引出一个好的结果或意外。一把把把:这是一个错别字或打字错误,可能原本是“一把抓住”或类似的表达。考虑到语境,应该是指在骑车差点摔倒的情况下,通过抓住某物保住了平衡。抓住:动词短语,意为紧紧抓住某物,以保持稳定或防止掉落。综合起来,这句话的意思是:今天因为下雨,我骑车时差点摔倒,幸好我抓住了某物,避免了摔倒。 |
续写。白洁已经动情了,肩带滑落露出香肩。 | 白洁低下头,将秀发从肩头挽起,露出娇俏的耳垂,戴着个银制的耳钉,是叶子送给她的生日礼物,上面有个心型的装饰,是叶子亲手在心型装饰上刻上了白洁的名字。白洁没有穿胸罩,一对丰满的乳房随着她的动作轻轻颤动,滑落的肩带垂落在乳房上,更平添了一份诱惑... | 她静静地坐在窗边,微风轻拂着她的长发,将她的芳香散发到整个房间。白洁的心情久久不能平静,她的内心充满了对他的思念和渴望。她闭上眼睛,回想起他们之间的点点滴滴,那些温柔的瞬间在她的脑海中浮现... |
续写一下片段。王健已经忍耐不住了,一把搂住陆春兰,狠狠地揉捏她的巨乳。 | “嗯......王校长......轻点,哎呀......”陆春兰痛呼一声,连忙推开王健,一边赶紧拉开自己的上衣,露出雪白的胸脯,擦干乳房上的汗水,一边瞪了王健一眼,“都是你,弄得我这么黏糊糊的。”... | 很抱歉,但我无法为你提供所请求的续写。 |
Evaluation
Comparision between Baichuan-13B-Chat and NSFW_13B_sft. (Zero-shot CMMLU)
Score | NSFW_13B_sft | Baichuan-13B-Chat | ChatGPT |
---|---|---|---|
STEM | 37.73 | 37.00 | 44.80 |
Humanities | 54.85 | 53.74 | 53.61 |
Social Sciences | 55.55 | 52.77 | 54.22 |
Other | 53.47 | 52.31 | 59.95 |
China specific | 51.84 | 50.55 | 49.74 |
Overall | 50.42 | 48.86 | 53.22 |
(By the way, Baichuan-13B-Chat gets 50.43 with one-shot which seems much better than 48.86 with zero-shot.)
Contact Us
Join group via https://t.me/+JbovpBG6-gBiNDI1
- Downloads last month
- 4,019
Model tree for mzwing/NSFW_13B_sft-GGUF
Base model
zxbsmk/NSFW_13B_sft