--- inference: false license: other language: - zh - en pipeline_tag: text-generation ---
TheBlokeAI

Chat & support: my new Discord server

Want to contribute? TheBloke's Patreon page

# Fire Balloon's Baichuan Llama 7B GGML These files are GGML format model files for [Fire Balloon's Baichuan Llama 7B](https://huggingface.co/fireballoon/baichuan-llama-7b). GGML files are for CPU + GPU inference using [llama.cpp](https://github.com/ggerganov/llama.cpp) and libraries and UIs which support this format, such as: * [text-generation-webui](https://github.com/oobabooga/text-generation-webui) * [KoboldCpp](https://github.com/LostRuins/koboldcpp) * [LoLLMS Web UI](https://github.com/ParisNeo/lollms-webui) * [llama-cpp-python](https://github.com/abetlen/llama-cpp-python) * [ctransformers](https://github.com/marella/ctransformers) This model is a Llama conversion of [Baichuan Inc's Baichuan 7B]https://huggingface.co/baichuan-inc/baichuan-7B). It contains the same data, but rewritten by Fire Balloon into the familiar Llama format. ## Repositories available * [4-bit GPTQ models for GPU inference](https://huggingface.co/TheBloke/baichuan-llama-7B-GPTQ) * [2, 3, 4, 5, 6 and 8-bit GGML models for CPU+GPU inference](https://huggingface.co/TheBloke/baichuan-llama-7B-GGML) * [Unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/fireballoon/baichuan-llama-7b) ## Prompt template A general prompt template is unknown at this point. The example given in the README is a 1-shot categorisation: ``` Hamlet->Shakespeare\nOne Hundred Years of Solitude-> ``` ## Compatibility ### Original llama.cpp quant methods: `q4_0, q4_1, q5_0, q5_1, q8_0` I have quantized these 'original' quantisation methods using an older version of llama.cpp so that they remain compatible with llama.cpp as of May 19th, commit `2d5db48`. These are guaranteed to be compatbile with any UIs, tools and libraries released since late May. ### New k-quant methods: `q2_K, q3_K_S, q3_K_M, q3_K_L, q4_K_S, q4_K_M, q5_K_S, q6_K` These new quantisation methods are compatible with llama.cpp as of June 6th, commit `2d43387`. They are now also compatible with recent releases of text-generation-webui, KoboldCpp, llama-cpp-python and ctransformers. Other tools and libraries may or may not be compatible - check their documentation if in doubt. ## Explanation of the new k-quant methods 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 * GGML_TYPE_Q8_K - "type-0" 8-bit quantization. Only used for quantizing intermediate results. The difference to the existing Q8_0 is that the block size is 256. All 2-6 bit dot products are implemented for this quantization type. 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 | | ---- | ---- | ---- | ---- | ---- | ----- | | baichuan-llama-7b.ggmlv3.q2_K.bin | q2_K | 2 | 3.02 GB | 5.52 GB | New k-quant method. Uses GGML_TYPE_Q4_K for the attention.vw and feed_forward.w2 tensors, GGML_TYPE_Q2_K for the other tensors. | | baichuan-llama-7b.ggmlv3.q3_K_L.bin | q3_K_L | 3 | 3.76 GB | 6.26 GB | New k-quant method. Uses GGML_TYPE_Q5_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else GGML_TYPE_Q3_K | | baichuan-llama-7b.ggmlv3.q3_K_M.bin | q3_K_M | 3 | 3.45 GB | 5.95 GB | New k-quant method. Uses GGML_TYPE_Q4_K for the attention.wv, attention.wo, and feed_forward.w2 tensors, else GGML_TYPE_Q3_K | | baichuan-llama-7b.ggmlv3.q3_K_S.bin | q3_K_S | 3 | 3.11 GB | 5.61 GB | New k-quant method. Uses GGML_TYPE_Q3_K for all tensors | | baichuan-llama-7b.ggmlv3.q4_0.bin | q4_0 | 4 | 3.94 GB | 6.44 GB | Original llama.cpp quant method, 4-bit. | | baichuan-llama-7b.ggmlv3.q4_1.bin | q4_1 | 4 | 4.38 GB | 6.88 GB | Original llama.cpp quant method, 4-bit. Higher accuracy than q4_0 but not as high as q5_0. However has quicker inference than q5 models. | | baichuan-llama-7b.ggmlv3.q4_K_M.bin | q4_K_M | 4 | 4.26 GB | 6.76 GB | New k-quant method. Uses GGML_TYPE_Q6_K for half of the attention.wv and feed_forward.w2 tensors, else GGML_TYPE_Q4_K | | baichuan-llama-7b.ggmlv3.q4_K_S.bin | q4_K_S | 4 | 4.01 GB | 6.51 GB | New k-quant method. Uses GGML_TYPE_Q4_K for all tensors | | baichuan-llama-7b.ggmlv3.q5_0.bin | q5_0 | 5 | 4.81 GB | 7.31 GB | Original llama.cpp quant method, 5-bit. Higher accuracy, higher resource usage and slower inference. | | baichuan-llama-7b.ggmlv3.q5_1.bin | q5_1 | 5 | 5.25 GB | 7.75 GB | Original llama.cpp quant method, 5-bit. Even higher accuracy, resource usage and slower inference. | | baichuan-llama-7b.ggmlv3.q5_K_M.bin | q5_K_M | 5 | 4.98 GB | 7.48 GB | New k-quant method. Uses GGML_TYPE_Q6_K for half of the attention.wv and feed_forward.w2 tensors, else GGML_TYPE_Q5_K | | baichuan-llama-7b.ggmlv3.q5_K_S.bin | q5_K_S | 5 | 4.85 GB | 7.35 GB | New k-quant method. Uses GGML_TYPE_Q5_K for all tensors | | baichuan-llama-7b.ggmlv3.q6_K.bin | q6_K | 6 | 5.74 GB | 8.24 GB | New k-quant method. Uses GGML_TYPE_Q8_K - 6-bit quantization - for all tensors | | baichuan-llama-7b.ggmlv3.q8_0.bin | q8_0 | 8 | 7.44 GB | 9.94 GB | Original llama.cpp quant method, 8-bit. Almost indistinguishable from float16. High resource use and slow. Not recommended for most users. | **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 run in `llama.cpp` I use the following command line; adjust for your tastes and needs: ``` ./main -t 10 -ngl 32 -m baichuan-llama-7b.ggmlv3.q5_0.bin --color -c 2048 --temp 0.7 --repeat_penalty 1.1 -n -1 -p "### Instruction: Write a story about llamas\n### Response:" ``` If you're able to use full GPU offloading, you should use `-t 1` to get best performance. If not able to fully offload to GPU, you should use more cores. Change `-t 10` to the number of physical CPU cores you have, or a lower number depending on what gives best performance. Change `-ngl 32` to the number of layers to offload to GPU. Remove it if you don't have GPU acceleration. If you want to have a chat-style conversation, replace the `-p ` argument with `-i -ins` ## How to run in `text-generation-webui` 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). ## Discord For further support, and discussions on these models and AI in general, join us at: [TheBloke AI's Discord server](https://discord.gg/Jq4vkcDakD) ## Thanks, and how to contribute. Thanks to the [chirper.ai](https://chirper.ai) team! I've had a lot of people ask if they can contribute. I enjoy providing models and helping people, and would love to be able to spend even more time doing it, as well as expanding into new projects like fine tuning/training. If you're able and willing to contribute it will be most gratefully received and will help me to keep providing more models, and to start work on new AI projects. Donaters will get priority support on any and all AI/LLM/model questions and requests, access to a private Discord room, plus other benefits. * Patreon: https://patreon.com/TheBlokeAI * Ko-Fi: https://ko-fi.com/TheBlokeAI **Special thanks to**: Luke from CarbonQuill, Aemon Algiz, Dmitriy Samsonov. **Patreon special mentions**: Mano Prime, Fen Risland, Derek Yates, Preetika Verma, webtim, Sean Connelly, Alps Aficionado, Karl Bernard, Junyu Yang, Nathan LeClaire, Chris McCloskey, Lone Striker, Asp the Wyvern, Eugene Pentland, Imad Khwaja, trip7s trip, WelcomeToTheClub, John Detwiler, Artur Olbinski, Khalefa Al-Ahmad, Trenton Dambrowitz, Talal Aujan, Kevin Schuppel, Luke Pendergrass, Pyrater, Joseph William Delisle, terasurfer , vamX, Gabriel Puliatti, David Flickinger, Jonathan Leane, Iucharbius , Luke, Deep Realms, Cory Kujawski, ya boyyy, Illia Dulskyi, senxiiz, Johann-Peter Hartmann, John Villwock, K, Ghost , Spiking Neurons AB, Nikolai Manek, Rainer Wilmers, Pierre Kircher, biorpg, Space Cruiser, Ai Maven, subjectnull, Willem Michiel, Ajan Kanaga, Kalila, chris gileta, Oscar Rangel. Thank you to all my generous patrons and donaters! # Original model card: Fire Balloon's Baichuan Llama 7B # baichuan-llama-7B 使用[LLaMA](https://huggingface.co/huggyllama/llama-7b)格式保存的[baichuan-7B](https://huggingface.co/baichuan-inc/baichuan-7B)。可以直接使用LlamaForCausalLM和LlamaTokenizer加载。 [baichuan-7B](https://huggingface.co/baichuan-inc/baichuan-7B) model saved in the format of the [LLaMA](https://huggingface.co/huggyllama/llama-7b) model. You can directly use LlamaForCausalLM and LlamaTokenizer to load the model. ```python from transformers import AutoModelForCausalLM, AutoTokenizer tokenizer = AutoTokenizer.from_pretrained("fireballoon/baichuan-llama-7b", use_fast=False) model = AutoModelForCausalLM.from_pretrained("fireballoon/baichuan-llama-7b", device_map="auto") ``` The following is from the original repo [baichuan-7B](https://huggingface.co/baichuan-inc/baichuan-7B). # baichuan-7B baichuan-7B是由百川智能开发的一个开源的大规模预训练模型。基于Transformer结构，在大约1.2万亿tokens上训练的70亿参数模型，支持中英双语，上下文窗口长度为4096。在标准的中文和英文权威benchmark（C-EVAL/MMLU）上均取得同尺寸最好的效果。 如果希望使用baichuan-7B（如进行推理、Finetune等），我们推荐使用配套代码库[baichuan-7B](https://github.com/baichuan-inc/baichuan-7B)。 baichuan-7B is an open-source large-scale pre-trained model developed by Baichuan Intelligent Technology. Based on the Transformer architecture, it is a model with 7 billion parameters trained on approximately 1.2 trillion tokens. It supports both Chinese and English, with a context window length of 4096. It achieves the best performance of its size on standard Chinese and English authoritative benchmarks (C-EVAL/MMLU). If you wish to use baichuan-7B (for inference, finetuning, etc.), we recommend using the accompanying code library [baichuan-7B](https://github.com/baichuan-inc/baichuan-7B). ## Why use baichuan-7B - 在同尺寸模型中baichuan-7B达到了目前SOTA的水平，参考下面MMLU指标 - baichuan-7B使用自有的中英文双语语料进行训练，在中文上进行优化，在C-Eval达到SOTA水平 - 不同于LLaMA完全禁止商业使用，baichuan-7B使用更宽松的开源协议，允许用于商业目的 - Among models of the same size, baichuan-7B has achieved the current state-of-the-art (SOTA) level, as evidenced by the following MMLU metrics. - baichuan-7B is trained on proprietary bilingual Chinese-English corpora, optimized for Chinese, and achieves SOTA performance on C-Eval. - Unlike LLaMA, which completely prohibits commercial use, baichuan-7B employs a more lenient open-source license, allowing for commercial purposes. ## How to Get Started with the Model 如下是一个使用baichuan-7B进行1-shot推理的任务，根据作品给出作者名，正确输出为"夜雨寄北->李商隐" ```python from transformers import AutoModelForCausalLM, AutoTokenizer tokenizer = AutoTokenizer.from_pretrained("fireballoon/baichuan-llama-7b", use_fast=False) model = AutoModelForCausalLM.from_pretrained("fireballoon/baichuan-llama-7b", device_map="auto") inputs = tokenizer('登鹳雀楼->王之涣\n夜雨寄北->', return_tensors='pt') inputs = inputs.to('cuda:0') pred = model.generate(**inputs, max_new_tokens=64,repetition_penalty=1.1) print(tokenizer.decode(pred.cpu()[0], skip_special_tokens=True)) ``` The following is a task of performing 1-shot inference using baichuan-7B, where the author's name is given based on the work, with the correct output being "One Hundred Years of Solitude->Gabriel Garcia Marquez" ```python from transformers import AutoModelForCausalLM, AutoTokenizer tokenizer = AutoTokenizer.from_pretrained("fireballoon/baichuan-llama-7b", use_fast=False) model = AutoModelForCausalLM.from_pretrained("fireballoon/baichuan-llama-7b", device_map="auto") inputs = tokenizer('Hamlet->Shakespeare\nOne Hundred Years of Solitude->', return_tensors='pt') inputs = inputs.to('cuda:0') pred = model.generate(**inputs, max_new_tokens=64,repetition_penalty=1.1) print(tokenizer.decode(pred.cpu()[0], skip_special_tokens=True)) ``` ## Model Details ### Model Description - **Developed by:** 百川智能(Baichuan Intelligent Technology) - **Email**: opensource@baichuan-inc.com - **Language(s) (NLP):** Chinese/English - **License:** [baichuan-7B License](https://huggingface.co/baichuan-inc/baichuan-7B/blob/main/baichuan-7B%20%E6%A8%A1%E5%9E%8B%E8%AE%B8%E5%8F%AF%E5%8D%8F%E8%AE%AE.pdf) ### Model Sources 整体模型基于标准的Transformer结构，我们采用了和LLaMA一样的模型设计 - **Position Embedding**：采用rotary-embedding，是现阶段被大多数模型采用的位置编码方案，具有很好的外推性。 - **Feedforward Layer**：采用SwiGLU，Feedforward变化为(8/3)倍的隐含层大小，即11008。 - **Layer Normalization**: 基于[RMSNorm](https://arxiv.org/abs/1910.07467)的Pre-Normalization。 具体参数和见下表 | Hyperparameter | Value | |----------------|-------| |n_parameters | 7000559616 | |n_layers | 32 | | n_heads | 32 | | d_model | 4096 | | vocab size | 64000 | | sequence length | 4096 | The overall model is based on the standard Transformer structure, and we have adopted the same model design as LLaMA: - Position Embedding: We use rotary-embedding, which is the position encoding scheme adopted by most models at this stage, and it has excellent extrapolation capabilities. - Feedforward Layer: We use SwiGLU. The feedforward changes to (8/3) times the size of the hidden layer, that is, 11008. - Layer Normalization: Pre-Normalization based on [RMSNorm](https://arxiv.org/abs/1910.07467). The specific parameters are as follows: | Hyperparameter | Value | |----------------|-------| |n_parameters | 7000559616 | |n_layers | 32 | | n_heads | 32 | | d_model | 4096 | | vocab size | 64000 | | sequence length | 4096 | ## Uses ### Downstream Use 我们同时开源出了和本模型配套的训练代码，允许进行高效的Finetune用于下游任务，具体参见[baichuan-7B](https://github.com/baichuan-inc/baichuan-7B)。 We have also open-sourced the training code that accompanies this model, allowing for efficient finetuning for downstream tasks. For more details, please refer to [baichuan-7B](https://github.com/baichuan-inc/baichuan-7B). ### Out-of-Scope Use 在没有充分评估风险和采取缓解措施的情况下投入生产使用；任何可能被视为不负责任或有害的使用案例。 Production use without adequate assessment of risks and mitigation; any use cases which may be considered irresponsible or harmful. ## Bias, Risks, and Limitations baichuan-7B可能会产生事实上不正确的输出，不应依赖它产生事实上准确的信息。baichuan-7B是在各种公共数据集上进行训练的。尽管我们已经做出了巨大的努力来清洗预训练数据，但这个模型可能会生成淫秽、偏见或其他冒犯性的输出。 baichuan-7B can produce factually incorrect output, and should not be relied on to produce factually accurate information. baichuan-7B was trained on various public datasets. While great efforts have been taken to clean the pretraining data, it is possible that this model could generate lewd, biased or otherwise offensive outputs. ## Training Details 训练具体设置参见[baichuan-7B](https://github.com/baichuan-inc/baichuan-7B)。 For specific training settings, please refer to [baichuan-7B](https://github.com/baichuan-inc/baichuan-7B). ## Evaluation ### 中文评测 #### C-Eval [CEval数据集](https://cevalbenchmark.com/index.html)是一个全面的中文基础模型评测数据集，涵盖了52个学科和四个难度的级别。我们使用该数据集的dev集作为few-shot的来源，在test集上进行了5-shot测试。 | Model 5-shot | Average | Avg(Hard) | STEM | Social Sciences | Humanities | Others | |-----------------------------|---------|-----------|------|-----------------|------------|--------| | GPT-4 | 68.7 | 54.9 | 67.1 | 77.6 | 64.5 | 67.8 | | ChatGPT | 54.4 | 41.4 | 52.9 | 61.8 | 50.9 | 53.6 | | Claude-v1.3 | 54.2 | 39.0 | 51.9 | 61.7 | 52.1 | 53.7 | | Claude-instant-v1.0 | 45.9 | 35.5 | 43.1 | 53.8 | 44.2 | 45.4 | | moss-moon-003-base (16B) | 27.4 | 24.5 | 27.0 | 29.1 | 27.2 | 26.9 | | Ziya-LLaMA-13B-pretrain | 30.2 | 22.7 | 27.7 | 34.4 | 32.0 | 28.9 | | LLaMA-7B-hf | 27.1 | 25.9 | 27.1 | 26.8 | 27.9 | 26.3 | | ChatGLM-6B | 34.5 | 23.1 | 30.4 | 39.6 | 37.4 | 34.5 | | Falcon-7B | 25.8 | 24.3 | 25.8 | 26.0 | 25.8 | 25.6 | | Open-LLaMA-v2-pretrain (7B) | 24.0 | 22.5 | 23.1 | 25.3 | 25.2 | 23.2 | | TigerBot-7B-base | 25.7 | 27.0 | 27.3 | 24.7 | 23.4 | 26.1 | | Aquila-7B* | 25.5 | 25.2 | 25.6 | 24.6 | 25.2 | 26.6 | | BLOOM-7B | 22.8 | 20.2 | 21.8 | 23.3 | 23.9 | 23.3 | | BLOOMZ-7B | 35.7 | 25.8 | 31.3 | 43.5 | 36.6 | 35.6 | | **baichuan-7B** | 42.8 | 31.5 | 38.2 | 52.0 | 46.2 | 39.3 | #### Gaokao [Gaokao](https://github.com/ExpressAI/AI-Gaokao) 是一个以中国高考题作为评测大语言模型能力的数据集，用以评估模型的语言能力和逻辑推理能力。 我们只保留了其中的单项选择题，并对所有模型进行统一5-shot测试。 以下是测试的结果。 | Model | Average | |-------------------------|-----------------| | Open-LLaMA-v2-pretrain | 21.41 | | Ziya-LLaMA-13B-pretrain | 23.17 | | Falcon-7B | 23.98 | | TigerBot-7B-base | 25.94 | | LLaMA-7B | 27.81 | | ChatGLM-6B | 21.41 | | BLOOM-7B | 26.96 | | BLOOMZ-7B | 28.72 | | Aquila-7B* | 24.39 | | **baichuan-7B** | **36.24** | #### AGIEval [AGIEval](https://github.com/microsoft/AGIEval) 旨在评估模型的认知和解决问题相关的任务中的一般能力。 我们只保留了其中的四选一单项选择题，随机划分后对所有模型进行了统一5-shot测试。 | Model | Average | |-------------------------|-----------------| | Open-LLaMA-v2-pretrain | 23.49 | | Ziya-LLaMA-13B-pretrain | 27.64 | | Falcon-7B | 27.18 | | TigerBot-7B-base | 25.19 | | LLaMA-7B | 28.17 | | ChatGLM-6B | 23.49 | | BLOOM-7B | 26.55 | | BLOOMZ-7B | 30.27 | | Aquila-7B* | 25.58 | | **baichuan-7B** | **34.44** | *其中Aquila模型来源于[智源官方网站](https://model.baai.ac.cn/model-detail/100098)，仅做参考 ### English Leaderboard In addition to Chinese, we also tested the model's performance in English. #### MMLU [MMLU](https://arxiv.org/abs/2009.03300) is an English evaluation dataset that includes 57 multiple-choice tasks, covering elementary mathematics, American history, computer science, law, etc. The difficulty ranges from high school level to expert level, making it a mainstream LLM evaluation dataset. We adopted the [open-source]((https://github.com/hendrycks/test)) evaluation scheme, and the final 5-shot results are as follows: | Model | Humanities | Social Sciences | STEM | Other | Average | |----------------------------------------|-----------:|:---------------:|:----:|:-----:|:-------:| | LLaMA-7B2 | 34.0 | 38.3 | 30.5 | 38.1 | 35.1 | | Falcon-7B1 | - | - | - | - | 35.0 | | mpt-7B1 | - | - | - | - | 35.6 | | ChatGLM-6B0 | 35.4 | 41.0 | 31.3 | 40.5 | 36.9 | | BLOOM 7B0 | 25.0 | 24.4 | 26.5 | 26.4 | 25.5 | | BLOOMZ 7B0 | 31.3 | 42.1 | 34.4 | 39.0 | 36.1 | | moss-moon-003-base (16B)0 | 24.2 | 22.8 | 22.4 | 24.4 | 23.6 | | moss-moon-003-sft (16B)0 | 30.5 | 33.8 | 29.3 | 34.4 | 31.9 | | **baichuan-7B0** | 38.4 | 48.9 | 35.6 | 48.1 | 42.3 | The superscript in the Model column indicates the source of the results. ``` 0:reimplemented 1:https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard 2:https://paperswithcode.com/sota/multi-task-language-understanding-on-mmlu ```