4bit/stable-vicuna-13B-GPTQ

StableVicuna-13B-GPTQ

This repo contains 4bit GPTQ format quantised models of CarterAI's StableVicuna 13B.

It is the result of first merging the deltas from the above repository with the original Llama 13B weights, then quantising to 4bit using GPTQ-for-LLaMa.

Repositories available

• 4bit GPTQ models for GPU inference.
• 4bit and 5bit GGML models for CPU inference.
• Unquantised 16bit model in HF format.

PROMPT TEMPLATE

This model works best with the following prompt template:

### Human: your prompt here
### Assistant:

How to easily download and use this model in text-generation-webui

Load text-generation-webui as you normally do.

1. Click the Model tab.
2. Under Download custom model or LoRA, enter this repo name: TheBloke/stable-vicuna-13B-GPTQ.
3. Click Download.
4. Wait until it says it's finished downloading.
5. As this is a GPTQ model, fill in the GPTQ parameters on the right: Bits = 4, Groupsize = 128, model_type = Llama
6. Now click the Refresh icon next to Model in the top left.
7. In the Model drop-down: choose this model: stable-vicuna-13B-GPTQ.
8. Click Reload the Model in the top right.
9. Once it says it's loaded, click the Text Generation tab and enter a prompt!

Provided files

I have uploaded two versions of the GPTQ.

Compatible file - stable-vicuna-13B-GPTQ-4bit.compat.no-act-order.safetensors

In the main branch - the default one - you will find stable-vicuna-13B-GPTQ-4bit.compat.no-act-order.safetensors

This will work with all versions of GPTQ-for-LLaMa. It has maximum compatibility

It was created without the --act-order parameter. It may have slightly lower inference quality compared to the other file, but is guaranteed to work on all versions of GPTQ-for-LLaMa and text-generation-webui.

• stable-vicuna-13B-GPTQ-4bit.compat.no-act-order.safetensors
  • Works with all versions of GPTQ-for-LLaMa code, both Triton and CUDA branches
  • Works with text-generation-webui one-click-installers
  • Parameters: Groupsize = 128g. No act-order.
  • Command used to create the GPTQ:

    CUDA_VISIBLE_DEVICES=0 python3 llama.py stable-vicuna-13B-HF c4 --wbits 4 --true-sequential --groupsize 128 --save_safetensors stable-vicuna-13B-GPTQ-4bit.no-act-order.safetensors
    

Latest file - stable-vicuna-13B-GPTQ-4bit.latest.act-order.safetensors

Created for more recent versions of GPTQ-for-LLaMa, and uses the --act-order flag for maximum theoretical performance.

To access this file, please switch to the latest branch fo this repo and download from there.

• stable-vicuna-13B-GPTQ-4bit.latest.act-order.safetensors
  • Only works with recent GPTQ-for-LLaMa code
  • Does not work with text-generation-webui one-click-installers
  • Parameters: Groupsize = 128g. act-order.
  • Offers highest quality quantisation, but requires recent GPTQ-for-LLaMa code
  • Command used to create the GPTQ:

    CUDA_VISIBLE_DEVICES=0 python3 llama.py stable-vicuna-13B-HF c4 --wbits 4 --true-sequential --act-order --groupsize 128 --save_safetensors stable-vicuna-13B-GPTQ-4bit.act-order.safetensors
    

Manual instructions for text-generation-webui

File stable-vicuna-13B-GPTQ-4bit.compat.no-act-order.safetensors can be loaded the same as any other GPTQ file, without requiring any updates to oobaboogas text-generation-webui.

Instructions on using GPTQ 4bit files in text-generation-webui are here.

The other safetensors model file was created using --act-order to give the maximum possible quantisation quality, but this means it requires that the latest GPTQ-for-LLaMa is used inside the UI.

If you want to use the act-order safetensors files and need to update the Triton branch of GPTQ-for-LLaMa, here are the commands I used to clone the Triton branch of GPTQ-for-LLaMa, clone text-generation-webui, and install GPTQ into the UI:

# Clone text-generation-webui, if you don't already have it
git clone https://github.com/oobabooga/text-generation-webui
# Make a repositories directory
mkdir text-generation-webui/repositories
cd text-generation-webui/repositories
# Clone the latest GPTQ-for-LLaMa code inside text-generation-webui
git clone https://github.com/qwopqwop200/GPTQ-for-LLaMa

Then install this model into text-generation-webui/models and launch the UI as follows:

cd text-generation-webui
python server.py --model stable-vicuna-13B-GPTQ --wbits 4 --groupsize 128 --model_type Llama # add any other command line args you want

The above commands assume you have installed all dependencies for GPTQ-for-LLaMa and text-generation-webui. Please see their respective repositories for further information.

If you can't update GPTQ-for-LLaMa or don't want to, you can use stable-vicuna-13B-GPTQ-4bit.no-act-order.safetensors as mentioned above, which should work without any upgrades to text-generation-webui.

Original StableVicuna-13B model card

Model Description

StableVicuna-13B is a Vicuna-13B v0 model fine-tuned using reinforcement learning from human feedback (RLHF) via Proximal Policy Optimization (PPO) on various conversational and instructional datasets.

Model Details

• Trained by: Duy Phung of CarperAI
• Model type: StableVicuna-13B is an auto-regressive language model based on the LLaMA transformer architecture.
• Language(s): English
• Library: trlX
• License for delta weights: CC-BY-NC-SA-4.0
  • Note: License for the base LLaMA model's weights is Meta's non-commercial bespoke license.
• Contact: For questions and comments about the model, visit the CarperAI and StableFoundation Discord servers.

Hyperparameter	Value
$n_{\text{parameters}}n\_\backslash text\{parameters\}$	13B
$d_{\text{model}}d\_\backslash text\{model\}$	5120
$n_{\text{layers}}n\_\backslash text\{layers\}$	40
$n_{\text{heads}}n\_\backslash text\{heads\}$	40

Training

Training Dataset

StableVicuna-13B is fine-tuned on a mix of three datasets. OpenAssistant Conversations Dataset (OASST1), a human-generated, human-annotated assistant-style conversation corpus consisting of 161,443 messages distributed across 66,497 conversation trees, in 35 different languages; GPT4All Prompt Generations, a dataset of 400k prompts and responses generated by GPT-4; and Alpaca, a dataset of 52,000 instructions and demonstrations generated by OpenAI's text-davinci-003 engine.

The reward model used during RLHF was also trained on OpenAssistant Conversations Dataset (OASST1) along with two other datasets: Anthropic HH-RLHF, a dataset of preferences about AI assistant helpfulness and harmlessness; and Stanford Human Preferences Dataset a dataset of 385K collective human preferences over responses to questions/instructions in 18 different subject areas, from cooking to legal advice.

Training Procedure

CarperAI/stable-vicuna-13b-delta was trained using PPO as implemented in trlX with the following configuration:

Hyperparameter	Value
num_rollouts	128
chunk_size	16
ppo_epochs	4
init_kl_coef	0.1
target	6
horizon	10000
gamma	1
lam	0.95
cliprange	0.2
cliprange_value	0.2
vf_coef	1.0
scale_reward	None
cliprange_reward	10
generation_kwargs
max_length	512
min_length	48
top_k	0.0
top_p	1.0
do_sample	True
temperature	1.0

Use and Limitations

Intended Use

This model is intended to be used for text generation with a focus on conversational tasks. Users may further fine-tune the model on their own data to improve the model's performance on their specific tasks in accordance with the non-commercial license.

Limitations and bias

The base LLaMA model is trained on various data, some of which may contain offensive, harmful, and biased content that can lead to toxic behavior. See Section 5.1 of the LLaMA paper. We have not performed any studies to determine how fine-tuning on the aforementioned datasets affect the model's behavior and toxicity. Do not treat chat responses from this model as a substitute for human judgment or as a source of truth. Please use responsibly.

Acknowledgements

This work would not have been possible without the support of Stability AI.

Citations

@article{touvron2023llama,
  title={LLaMA: Open and Efficient Foundation Language Models},
  author={Touvron, Hugo and Lavril, Thibaut and Izacard, Gautier and Martinet, Xavier and Lachaux, Marie-Anne and Lacroix, Timoth{\'e}e and Rozi{\`e}re, Baptiste and Goyal, Naman and Hambro, Eric and Azhar, Faisal and Rodriguez, Aurelien and Joulin, Armand and Grave, Edouard and Lample, Guillaume},
  journal={arXiv preprint arXiv:2302.13971},
  year={2023}
}

@misc{vicuna2023,
    title = {Vicuna: An Open-Source Chatbot Impressing GPT-4 with 90%* ChatGPT Quality},
    url = {https://vicuna.lmsys.org},
    author = {Chiang, Wei-Lin and Li, Zhuohan and Lin, Zi and Sheng, Ying and Wu, Zhanghao and Zhang, Hao and Zheng, Lianmin and Zhuang, Siyuan and Zhuang, Yonghao and Gonzalez, Joseph E. and Stoica, Ion and Xing, Eric P.},
    month = {March},
    year = {2023}
}

@misc{gpt4all,
  author = {Yuvanesh Anand and Zach Nussbaum and Brandon Duderstadt and Benjamin Schmidt and Andriy Mulyar},
  title = {GPT4All: Training an Assistant-style Chatbot with Large Scale Data Distillation from GPT-3.5-Turbo},
  year = {2023},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/nomic-ai/gpt4all}},
}

@misc{alpaca,
  author = {Rohan Taori and Ishaan Gulrajani and Tianyi Zhang and Yann Dubois and Xuechen Li and Carlos Guestrin and Percy Liang and Tatsunori B. Hashimoto },
  title = {Stanford Alpaca: An Instruction-following LLaMA model},
  year = {2023},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/tatsu-lab/stanford_alpaca}},
}

@software{leandro_von_werra_2023_7790115,
  author       = {Leandro von Werra and
                  Alex Havrilla and
                  Max reciprocated and
                  Jonathan Tow and
                  Aman cat-state and
                  Duy V. Phung and
                  Louis Castricato and
                  Shahbuland Matiana and
                  Alan and
                  Ayush Thakur and
                  Alexey Bukhtiyarov and
                  aaronrmm and
                  Fabrizio Milo and
                  Daniel and
                  Daniel King and
                  Dong Shin and
                  Ethan Kim and
                  Justin Wei and
                  Manuel Romero and
                  Nicky Pochinkov and
                  Omar Sanseviero and
                  Reshinth Adithyan and
                  Sherman Siu and
                  Thomas Simonini and
                  Vladimir Blagojevic and
                  Xu Song and
                  Zack Witten and
                  alexandremuzio and
                  crumb},
  title        = {{CarperAI/trlx: v0.6.0: LLaMa (Alpaca), Benchmark 
                   Util, T5 ILQL, Tests}},
  month        = mar,
  year         = 2023,
  publisher    = {Zenodo},
  version      = {v0.6.0},
  doi          = {10.5281/zenodo.7790115},
  url          = {https://doi.org/10.5281/zenodo.7790115}
}