--- language: - en tags: - causal-lm - llama license: cc-by-nc-sa-4.0 datasets: - OpenAssistant/oasst1 - nomic-ai/gpt4all_prompt_generations - tatsu-lab/alpaca inference: false --- # StableVicuna-13B-GGML This is GGML format quantised 4bit and 5bit models of [CarperAI's StableVicuna 13B](https://huggingface.co/CarperAI/stable-vicuna-13b-delta). It is the result of merging the deltas from the above repository with the original Llama 13B weights, and then quantising to 4bit and 5bit GGML for CPU inference using [llama.cpp](https://github.com/ggerganov/llama.cpp). ## Repositories available * [4bit GPTQ models for GPU inference](https://huggingface.co/TheBloke/stable-vicuna-13B-GPTQ). * [4bit and 5bit GGML models for CPU inference](https://huggingface.co/TheBloke/stable-vicuna-13B-GGML). * [Unquantised 16bit model in HF format](https://huggingface.co/TheBloke/stable-vicuna-13B-HF). ## PROMPT TEMPLATE This model works best with the following prompt template: ``` ### Human: your prompt here ### Assistant: ``` ## REQUIRES LATEST LLAMA.CPP (May 12th 2023 - commit b9fd7ee)! llama.cpp recently made a breaking change to its quantisation methods. I have re-quantised the GGML files in this repo. Therefore you will require llama.cpp compiled on May 12th or later (commit `b9fd7ee` or later) to use them. The previous files, which will still work in older versions of llama.cpp, can be found in branch `previous_llama`. ## Provided files | Name | Quant method | Bits | Size | RAM required | Use case | | ---- | ---- | ---- | ---- | ---- | ----- | `stable-vicuna-13B.ggml.q4_0.bin` | q4_0 | 4bit | 8.14GB | 10.5GB | 4bit. | `stable-vicuna-13B.ggml.q5_0.bin` | q5_0 | 5bit | 8.95GB | 11.0GB | 5bit. Higher accuracy, higher resource usage and slower inference. | `stable-vicuna-13B.ggml.q5_1.bin` | q5_1 | 5bit | 9.76GB | 12.25GB | 5bit. Higher accuracy than q5_0, but again higher resource usage and slower inference. | ## How to run in `llama.cpp` I use the following command line; adjust for your tastes and needs: ``` ./main -t 18 -m stable-vicuna-13B.ggml.q4_0.bin --color -c 2048 --temp 0.7 --repeat_penalty 1.1 -n -1 -r "### Human:" -p "### Human: write a story about llamas ### Assistant:" ``` Change `-t 18` to the number of physical CPU cores you have. For example if your system has 8 cores/16 threads, use `-t 8`. ## How to run in `text-generation-webui` GGML models can be loaded into text-generation-webui by installing the llama.cpp module, then placing the ggml model file in a model folder as usual. Further instructions here: [text-generation-webui/docs/llama.cpp-models.md](https://github.com/oobabooga/text-generation-webui/blob/main/docs/llama.cpp-models.md). Note: at this time text-generation-webui may not support the May 12th updated quantisation methods. **Thireus** has written a [great guide on how to update it to the latest llama.cpp code](https://huggingface.co/TheBloke/wizardLM-7B-GGML/discussions/5) which may help with getting the files working in text-gen-ui sooner. # Original StableVicuna-13B model card ## Model Description StableVicuna-13B is a [Vicuna-13B v0](https://huggingface.co/lmsys/vicuna-13b-delta-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](https://github.com/PhungVanDuy) of [CarperAI](https://carper.ai) * **Model type:** **StableVicuna-13B** is an auto-regressive language model based on the LLaMA transformer architecture. * **Language(s)**: English * **Library**: [trlX](https://github.com/CarperAI/trlx) * **License for delta weights**: [CC-BY-NC-SA-4.0](https://creativecommons.org/licenses/by-nc-sa/4.0/) * *Note*: License for the base LLaMA model's weights is Meta's [non-commercial bespoke license](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md). * **Contact**: For questions and comments about the model, visit the [CarperAI](https://discord.com/invite/KgfkCVYHdu) and [StableFoundation](https://discord.gg/stablediffusion) Discord servers. | Hyperparameter | Value | |---------------------------|-------| | \\(n_\text{parameters}\\) | 13B | | \\(d_\text{model}\\) | 5120 | | \\(n_\text{layers}\\) | 40 | | \\(n_\text{heads}\\) | 40 | ## Training ### Training Dataset StableVicuna-13B is fine-tuned on a mix of three datasets. [OpenAssistant Conversations Dataset (OASST1)](https://huggingface.co/datasets/OpenAssistant/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](https://huggingface.co/datasets/nomic-ai/gpt4all_prompt_generations), a dataset of 400k prompts and responses generated by GPT-4; and [Alpaca](https://huggingface.co/datasets/tatsu-lab/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)](https://huggingface.co/datasets/OpenAssistant/oasst1) along with two other datasets: [Anthropic HH-RLHF](https://huggingface.co/datasets/Anthropic/hh-rlhf), a dataset of preferences about AI assistant helpfulness and harmlessness; and [Stanford Human Preferences Dataset](https://huggingface.co/datasets/stanfordnlp/SHP) 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`](https://github.com/CarperAI/trlx/blob/main/trlx/trainer/accelerate_ppo_trainer.py) 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](https://creativecommons.org/licenses/by-nc/4.0/). ### 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](https://arxiv.org/abs/2302.13971). 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](https://stability.ai/). ## Citations ```bibtex @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} } ``` ```bibtex @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} } ``` ```bibtex @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}}, } ``` ```bibtex @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}}, } ``` ```bibtex @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} } ```