--- base_model: jondurbin/bagel-8x7b-v0.2 datasets: - ai2_arc - jondurbin/airoboros-3.2 - codeparrot/apps - facebook/belebele - boolq - jondurbin/cinematika-v0.1 - drop - lmsys/lmsys-chat-1m - TIGER-Lab/MathInstruct - cais/mmlu - Muennighoff/natural-instructions - openbookqa - piqa - Vezora/Tested-22k-Python-Alpaca - cakiki/rosetta-code - Open-Orca/SlimOrca - spider - squad_v2 - migtissera/Synthia-v1.3 - datasets/winogrande - nvidia/HelpSteer - Intel/orca_dpo_pairs - unalignment/toxic-dpo-v0.1 - jondurbin/truthy-dpo-v0.1 - allenai/ultrafeedback_binarized_cleaned - Squish42/bluemoon-fandom-1-1-rp-cleaned - LDJnr/Capybara - JULIELab/EmoBank - kingbri/PIPPA-shareGPT inference: false license: apache-2.0 model_creator: Jon Durbin model_name: Bagel 8X7B v0.2 model_type: mixtral prompt_template: 'Below is an instruction that describes a task. Write a response that appropriately completes the request. ### Instruction: {prompt} ### Response: ' quantized_by: TheBloke ---
TheBlokeAI

Chat & support: TheBloke's Discord server

Want to contribute? TheBloke's Patreon page

TheBloke's LLM work is generously supported by a grant from andreessen horowitz (a16z)

# Bagel 8X7B v0.2 - GPTQ - Model creator: [Jon Durbin](https://huggingface.co/jondurbin) - Original model: [Bagel 8X7B v0.2](https://huggingface.co/jondurbin/bagel-8x7b-v0.2) # Description This repo contains GPTQ model files for [Jon Durbin's Bagel 8X7B v0.2](https://huggingface.co/jondurbin/bagel-8x7b-v0.2). Multiple GPTQ parameter permutations are provided; see Provided Files below for details of the options provided, their parameters, and the software used to create them. ## Repositories available * [AWQ model(s) for GPU inference.](https://huggingface.co/TheBloke/bagel-8x7b-v0.2-AWQ) * [GPTQ models for GPU inference, with multiple quantisation parameter options.](https://huggingface.co/TheBloke/bagel-8x7b-v0.2-GPTQ) * [2, 3, 4, 5, 6 and 8-bit GGUF models for CPU+GPU inference](https://huggingface.co/TheBloke/bagel-8x7b-v0.2-GGUF) * [Jon Durbin's original unquantised fp16 model in pytorch format, for GPU inference and for further conversions](https://huggingface.co/jondurbin/bagel-8x7b-v0.2) ## Prompt template: Alpaca ``` Below is an instruction that describes a task. Write a response that appropriately completes the request. ### Instruction: {prompt} ### Response: ``` ## Known compatible clients / servers GPTQ models are currently supported on Linux (NVidia/AMD) and Windows (NVidia only). macOS users: please use GGUF models. These GPTQ models are known to work in the following inference servers/webuis. - [text-generation-webui](https://github.com/oobabooga/text-generation-webui) - [KoboldAI United](https://github.com/henk717/koboldai) - [LoLLMS Web UI](https://github.com/ParisNeo/lollms-webui) - [Hugging Face Text Generation Inference (TGI)](https://github.com/huggingface/text-generation-inference) This may not be a complete list; if you know of others, please let me know! ## Provided files, and GPTQ parameters Multiple quantisation parameters are provided, to allow you to choose the best one for your hardware and requirements. Each separate quant is in a different branch. See below for instructions on fetching from different branches. Most GPTQ files are made with AutoGPTQ. Mistral models are currently made with Transformers.
Explanation of GPTQ parameters - Bits: The bit size of the quantised model. - GS: GPTQ group size. Higher numbers use less VRAM, but have lower quantisation accuracy. "None" is the lowest possible value. - Act Order: True or False. Also known as `desc_act`. True results in better quantisation accuracy. Some GPTQ clients have had issues with models that use Act Order plus Group Size, but this is generally resolved now. - Damp %: A GPTQ parameter that affects how samples are processed for quantisation. 0.01 is default, but 0.1 results in slightly better accuracy. - GPTQ dataset: The calibration dataset used during quantisation. Using a dataset more appropriate to the model's training can improve quantisation accuracy. Note that the GPTQ calibration dataset is not the same as the dataset used to train the model - please refer to the original model repo for details of the training dataset(s). - Sequence Length: The length of the dataset sequences used for quantisation. Ideally this is the same as the model sequence length. For some very long sequence models (16+K), a lower sequence length may have to be used. Note that a lower sequence length does not limit the sequence length of the quantised model. It only impacts the quantisation accuracy on longer inference sequences. - ExLlama Compatibility: Whether this file can be loaded with ExLlama, which currently only supports Llama and Mistral models in 4-bit.
| Branch | Bits | GS | Act Order | Damp % | GPTQ Dataset | Seq Len | Size | ExLlama | Desc | | ------ | ---- | -- | --------- | ------ | ------------ | ------- | ---- | ------- | ---- | | [main](https://huggingface.co/TheBloke/bagel-8x7b-v0.2-GPTQ/tree/main) | 4 | None | Yes | 0.1 | [VMware Open Instruct](https://huggingface.co/datasets/VMware/open-instruct/viewer/) | 8192 | 23.81 GB | No | 4-bit, with Act Order. No group size, to lower VRAM requirements. | | [gptq-4bit-128g-actorder_True](https://huggingface.co/TheBloke/bagel-8x7b-v0.2-GPTQ/tree/gptq-4bit-128g-actorder_True) | 4 | 128 | Yes | 0.1 | [VMware Open Instruct](https://huggingface.co/datasets/VMware/open-instruct/viewer/) | 8192 | 24.70 GB | No | 4-bit, with Act Order and group size 128g. Uses even less VRAM than 64g, but with slightly lower accuracy. | | [gptq-4bit-32g-actorder_True](https://huggingface.co/TheBloke/bagel-8x7b-v0.2-GPTQ/tree/gptq-4bit-32g-actorder_True) | 4 | 32 | Yes | 0.1 | [VMware Open Instruct](https://huggingface.co/datasets/VMware/open-instruct/viewer/) | 8192 | 27.42 GB | No | 4-bit, with Act Order and group size 32g. Gives highest possible inference quality, with maximum VRAM usage. | | [gptq-3bit--1g-actorder_True](https://huggingface.co/TheBloke/bagel-8x7b-v0.2-GPTQ/tree/gptq-3bit--1g-actorder_True) | 3 | None | Yes | 0.1 | [VMware Open Instruct](https://huggingface.co/datasets/VMware/open-instruct/viewer/) | 8192 | 18.01 GB | No | 3-bit, with Act Order and no group size. Lowest possible VRAM requirements. May be lower quality than 3-bit 128g. | | [gptq-3bit-128g-actorder_True](https://huggingface.co/TheBloke/bagel-8x7b-v0.2-GPTQ/tree/gptq-3bit-128g-actorder_True) | 3 | 128 | Yes | 0.1 | [VMware Open Instruct](https://huggingface.co/datasets/VMware/open-instruct/viewer/) | 8192 | 18.85 GB | No | 3-bit, with group size 128g and act-order. Higher quality than 128g-False. | | [gptq-8bit--1g-actorder_True](https://huggingface.co/TheBloke/bagel-8x7b-v0.2-GPTQ/tree/gptq-8bit--1g-actorder_True) | 8 | None | Yes | 0.1 | [VMware Open Instruct](https://huggingface.co/datasets/VMware/open-instruct/viewer/) | 8192 | 47.04 GB | No | 8-bit, with Act Order. No group size, to lower VRAM requirements. | | [gptq-8bit-128g-actorder_True](https://huggingface.co/TheBloke/bagel-8x7b-v0.2-GPTQ/tree/gptq-8bit-128g-actorder_True) | 8 | 128 | Yes | 0.1 | [VMware Open Instruct](https://huggingface.co/datasets/VMware/open-instruct/viewer/) | 8192 | 48.10 GB | No | 8-bit, with group size 128g for higher inference quality and with Act Order for even higher accuracy. | ## How to download, including from branches ### In text-generation-webui To download from the `main` branch, enter `TheBloke/bagel-8x7b-v0.2-GPTQ` in the "Download model" box. To download from another branch, add `:branchname` to the end of the download name, eg `TheBloke/bagel-8x7b-v0.2-GPTQ:gptq-4bit-128g-actorder_True` ### From the command line I recommend using the `huggingface-hub` Python library: ```shell pip3 install huggingface-hub ``` To download the `main` branch to a folder called `bagel-8x7b-v0.2-GPTQ`: ```shell mkdir bagel-8x7b-v0.2-GPTQ huggingface-cli download TheBloke/bagel-8x7b-v0.2-GPTQ --local-dir bagel-8x7b-v0.2-GPTQ --local-dir-use-symlinks False ``` To download from a different branch, add the `--revision` parameter: ```shell mkdir bagel-8x7b-v0.2-GPTQ huggingface-cli download TheBloke/bagel-8x7b-v0.2-GPTQ --revision gptq-4bit-128g-actorder_True --local-dir bagel-8x7b-v0.2-GPTQ --local-dir-use-symlinks False ```
More advanced huggingface-cli download usage If you remove the `--local-dir-use-symlinks False` parameter, the files will instead be stored in the central Hugging Face cache directory (default location on Linux is: `~/.cache/huggingface`), and symlinks will be added to the specified `--local-dir`, pointing to their real location in the cache. This allows for interrupted downloads to be resumed, and allows you to quickly clone the repo to multiple places on disk without triggering a download again. The downside, and the reason why I don't list that as the default option, is that the files are then hidden away in a cache folder and it's harder to know where your disk space is being used, and to clear it up if/when you want to remove a download model. The cache location can be changed with the `HF_HOME` environment variable, and/or the `--cache-dir` parameter to `huggingface-cli`. For more documentation on downloading with `huggingface-cli`, please see: [HF -> Hub Python Library -> Download files -> Download from the CLI](https://huggingface.co/docs/huggingface_hub/guides/download#download-from-the-cli). To accelerate downloads on fast connections (1Gbit/s or higher), install `hf_transfer`: ```shell pip3 install hf_transfer ``` And set environment variable `HF_HUB_ENABLE_HF_TRANSFER` to `1`: ```shell mkdir bagel-8x7b-v0.2-GPTQ HF_HUB_ENABLE_HF_TRANSFER=1 huggingface-cli download TheBloke/bagel-8x7b-v0.2-GPTQ --local-dir bagel-8x7b-v0.2-GPTQ --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.
### With `git` (**not** recommended) To clone a specific branch with `git`, use a command like this: ```shell git clone --single-branch --branch gptq-4bit-128g-actorder_True https://huggingface.co/TheBloke/bagel-8x7b-v0.2-GPTQ ``` Note that using Git with HF repos is strongly discouraged. It will be much slower than using `huggingface-hub`, and will use twice as much disk space as it has to store the model files twice (it stores every byte both in the intended target folder, and again in the `.git` folder as a blob.) ## How to easily download and use this model in [text-generation-webui](https://github.com/oobabooga/text-generation-webui) Please make sure you're using the latest version of [text-generation-webui](https://github.com/oobabooga/text-generation-webui). It is strongly recommended to use the text-generation-webui one-click-installers unless you're sure you know how to make a manual install. 1. Click the **Model tab**. 2. Under **Download custom model or LoRA**, enter `TheBloke/bagel-8x7b-v0.2-GPTQ`. - To download from a specific branch, enter for example `TheBloke/bagel-8x7b-v0.2-GPTQ:gptq-4bit-128g-actorder_True` - see Provided Files above for the list of branches for each option. 3. Click **Download**. 4. The model will start downloading. Once it's finished it will say "Done". 5. In the top left, click the refresh icon next to **Model**. 6. In the **Model** dropdown, choose the model you just downloaded: `bagel-8x7b-v0.2-GPTQ` 7. The model will automatically load, and is now ready for use! 8. If you want any custom settings, set them and then click **Save settings for this model** followed by **Reload the Model** in the top right. - Note that you do not need to and should not set manual GPTQ parameters any more. These are set automatically from the file `quantize_config.json`. 9. Once you're ready, click the **Text Generation** tab and enter a prompt to get started! ## Serving this model from Text Generation Inference (TGI) It's recommended to use TGI version 1.1.0 or later. The official Docker container is: `ghcr.io/huggingface/text-generation-inference:1.1.0` Example Docker parameters: ```shell --model-id TheBloke/bagel-8x7b-v0.2-GPTQ --port 3000 --quantize gptq --max-input-length 3696 --max-total-tokens 4096 --max-batch-prefill-tokens 4096 ``` Example Python code for interfacing with TGI (requires huggingface-hub 0.17.0 or later): ```shell pip3 install huggingface-hub ``` ```python from huggingface_hub import InferenceClient endpoint_url = "https://your-endpoint-url-here" prompt = "Tell me about AI" prompt_template=f'''Below is an instruction that describes a task. Write a response that appropriately completes the request. ### Instruction: {prompt} ### Response: ''' client = InferenceClient(endpoint_url) response = client.text_generation( prompt_template, max_new_tokens=128, do_sample=True, temperature=0.7, top_p=0.95, top_k=40, repetition_penalty=1.1 ) print(f"Model output: {response}") ``` ## Python code example: inference from this GPTQ model ### Install the necessary packages Requires: Transformers 4.33.0 or later, Optimum 1.12.0 or later, and AutoGPTQ 0.4.2 or later. ```shell pip3 install --upgrade transformers optimum # If using PyTorch 2.1 + CUDA 12.x: pip3 install --upgrade auto-gptq # or, if using PyTorch 2.1 + CUDA 11.x: pip3 install --upgrade auto-gptq --extra-index-url https://huggingface.github.io/autogptq-index/whl/cu118/ ``` If you are using PyTorch 2.0, you will need to install AutoGPTQ from source. Likewise if you have problems with the pre-built wheels, you should try building from source: ```shell pip3 uninstall -y auto-gptq git clone https://github.com/PanQiWei/AutoGPTQ cd AutoGPTQ git checkout v0.5.1 pip3 install . ``` ### Example Python code ```python from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline model_name_or_path = "TheBloke/bagel-8x7b-v0.2-GPTQ" # To use a different branch, change revision # For example: revision="gptq-4bit-128g-actorder_True" model = AutoModelForCausalLM.from_pretrained(model_name_or_path, device_map="auto", trust_remote_code=False, revision="main") tokenizer = AutoTokenizer.from_pretrained(model_name_or_path, use_fast=True) prompt = "Write a story about llamas" system_message = "You are a story writing assistant" prompt_template=f'''Below is an instruction that describes a task. Write a response that appropriately completes the request. ### Instruction: {prompt} ### Response: ''' print("\n\n*** Generate:") input_ids = tokenizer(prompt_template, return_tensors='pt').input_ids.cuda() output = model.generate(inputs=input_ids, temperature=0.7, do_sample=True, top_p=0.95, top_k=40, max_new_tokens=512) print(tokenizer.decode(output[0])) # Inference can also be done using transformers' pipeline print("*** Pipeline:") pipe = pipeline( "text-generation", model=model, tokenizer=tokenizer, max_new_tokens=512, do_sample=True, temperature=0.7, top_p=0.95, top_k=40, repetition_penalty=1.1 ) print(pipe(prompt_template)[0]['generated_text']) ``` ## Compatibility The files provided are tested to work with Transformers. For non-Mistral models, AutoGPTQ can also be used directly. [ExLlama](https://github.com/turboderp/exllama) is compatible with Llama architecture models (including Mistral, Yi, DeepSeek, SOLAR, etc) in 4-bit. Please see the Provided Files table above for per-file compatibility. For a list of clients/servers, please see "Known compatible clients / servers", above. ## Discord For further support, and discussions on these models and AI in general, join us at: [TheBloke AI's Discord server](https://discord.gg/theblokeai) ## Thanks, and how to contribute Thanks to the [chirper.ai](https://chirper.ai) team! Thanks to Clay from [gpus.llm-utils.org](llm-utils)! 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**: Aemon Algiz. **Patreon special mentions**: Michael Levine, 阿明, Trailburnt, Nikolai Manek, John Detwiler, Randy H, Will Dee, Sebastain Graf, NimbleBox.ai, Eugene Pentland, Emad Mostaque, Ai Maven, Jim Angel, Jeff Scroggin, Michael Davis, Manuel Alberto Morcote, Stephen Murray, Robert, Justin Joy, Luke @flexchar, Brandon Frisco, Elijah Stavena, S_X, Dan Guido, Undi ., Komninos Chatzipapas, Shadi, theTransient, Lone Striker, Raven Klaugh, jjj, Cap'n Zoog, Michel-Marie MAUDET (LINAGORA), Matthew Berman, David, Fen Risland, Omer Bin Jawed, Luke Pendergrass, Kalila, OG, Erik Bjäreholt, Rooh Singh, Joseph William Delisle, Dan Lewis, TL, John Villwock, AzureBlack, Brad, Pedro Madruga, Caitlyn Gatomon, K, jinyuan sun, Mano Prime, Alex, Jeffrey Morgan, Alicia Loh, Illia Dulskyi, Chadd, transmissions 11, fincy, Rainer Wilmers, ReadyPlayerEmma, knownsqashed, Mandus, biorpg, Deo Leter, Brandon Phillips, SuperWojo, Sean Connelly, Iucharbius, Jack West, Harry Royden McLaughlin, Nicholas, terasurfer, Vitor Caleffi, Duane Dunston, Johann-Peter Hartmann, David Ziegler, Olakabola, Ken Nordquist, Trenton Dambrowitz, Tom X Nguyen, Vadim, Ajan Kanaga, Leonard Tan, Clay Pascal, Alexandros Triantafyllidis, JM33133, Xule, vamX, ya boyyy, subjectnull, Talal Aujan, Alps Aficionado, wassieverse, Ari Malik, James Bentley, Woland, Spencer Kim, Michael Dempsey, Fred von Graf, Elle, zynix, William Richards, Stanislav Ovsiannikov, Edmond Seymore, Jonathan Leane, Martin Kemka, usrbinkat, Enrico Ros Thank you to all my generous patrons and donaters! And thank you again to a16z for their generous grant. # Original model card: Jon Durbin's Bagel 8X7B v0.2 # A bagel, with everything (except DPO) ![bagel](bagel.png) ## Overview An experimental fine-tune of [mixtral-8x7b-v0.1](https://huggingface.co/mistralai/Mixtral-8x7B-v0.1) using [bagel](https://github.com/jondurbin/bagel) This is the model after the SFT phase, before DPO has been applied. Hardware kindly provided by [Massed Compute](https://massedcompute.com/?utm_source=huggingface&utm_creative_format=model_card&utm_content=creator_jon) ### Data sources *Yes, you will see benchmark names in the list, but this only uses the train splits, and a decontamination by cosine similarity is performed at the end as a sanity check* - [ai2_arc](https://huggingface.co/datasets/ai2_arc) - Abstraction and reasoning dataset, useful in measuring "intelligence" to a certain extent. - [airoboros](https://huggingface.co/datasets/unalignment/spicy-3.1) - Variety of categories of synthetic instructions generated by gpt-4. - [apps](https://huggingface.co/datasets/codeparrot/apps) - Python coding dataset with 10k problems. - [belebele](https://huggingface.co/datasets/facebook/belebele) - Multi-lingual reading comprehension dataset. - [bluemoon](https://huggingface.co/datasets/Squish42/bluemoon-fandom-1-1-rp-cleaned) - Roleplay data scraped from Bluemoon, then cleaned and formatted as ShareGPT. - [boolq](https://huggingface.co/datasets/boolq) - Corpus of yes/no questions (which can be surprisingly difficult for AI to answer apparently?) - [capybara](https://huggingface.co/datasets/LDJnr/Capybara) - Multi-turn dataset used to create the capybara models. - [cinematika](https://huggingface.co/datasets/jondurbin/cinematika-v0.1) (instruction and plain text) - RP-style data synthesized from movie scripts so the model isn't quite as boring as it otherwise would be. - [drop](https://huggingface.co/datasets/drop) - More reading comprehension. - [emobank](https://github.com/JULIELab/EmoBank) - Emotion annotations using the Valence-Arousal-Domninance scheme. - [gutenberg](https://www.gutenberg.org/) (plain text) - Books/plain text, again to make the model less boring, only a handful of examples supported by [chapterize](https://github.com/JonathanReeve/chapterize) - [lmsys_chat_1m](https://huggingface.co/datasets/lmsys/lmsys-chat-1m) (only gpt-4 items, also used for DPO) - Chats collected by the lmsys chat arena, containing a wide variety of chats with various models. - [mathinstruct](https://huggingface.co/datasets/TIGER-Lab/MathInstruct) - Composite dataset with a variety of math-related tasks and problem/question formats. - [mmlu](https://huggingface.co/datasets/cais/mmlu) - Massive Multitask Language Understanding - a wide variety of questions about various subject matters. - [natural_instructions](https://huggingface.co/datasets/Muennighoff/natural-instructions) - Millions of instructions from 1600+ task categories (sampled down substantially, stratified by task type) - [openbookqa](https://huggingface.co/datasets/openbookqa) - Question answering dataset. - [pippa](https://huggingface.co/datasets/kingbri/PIPPA-shareGPT) - Deduped version of [PIPPA](https://huggingface.co/datasets/PygmalionAI/PIPPA) in ShareGPT format. - [piqa](https://huggingface.co/datasets/piqa) - Phyiscal interaction question answering. - [python_alpaca](https://huggingface.co/datasets/Vezora/Tested-22k-Python-Alpaca) - Python instruction response pairs, validated as functional. - [rosetta_code](https://huggingface.co/datasets/cakiki/rosetta-code) - Code problems and solutions in a variety of programming languages taken from rosettacode.org. - [slimorca](https://huggingface.co/datasets/Open-Orca/SlimOrca) - Collection of ~500k gpt-4 verified chats from OpenOrca. - [spider](https://huggingface.co/datasets/spider) - SQL-targeted dataset. - [squad_v2](https://huggingface.co/datasets/squad_v2) - Contextual question answering (RAG). - [synthia](https://huggingface.co/datasets/migtissera/Synthia-v1.3) - GPT-4 generated data using advanced prompting from Migel Tissera. - [winogrande](https://huggingface.co/datasets/winogrande) - Fill in the blank style prompts. Only the train splits were used (if a split was provided), and an additional pass of decontamination is performed using approximate nearest neighbor search (via faiss). ## How to easily download and use this model [Massed Compute](https://massedcompute.com/?utm_source=huggingface&utm_creative_format=model_card&utm_content=creator_jon) has created a Virtual Machine (VM) pre-loaded with TGI and Text Generation WebUI. 1) For this model rent the [Jon Durbin 4xA6000](https://shop.massedcompute.com/products/jon-durbin-4x-a6000?utm_source=huggingface&utm_creative_format=model_card&utm_content=creator_jon) Virtual Machine 2) After you start your rental you will receive an email with instructions on how to Login to the VM 3) Once inside the VM, open the terminal and run `conda activate text-generation-inference` 4) Then `cd Desktop/text-generation-inference/` 5) Run `volume=$PWD/data` 6) Run`model=jondurbin/bagel-8x7b-v0.2` 7) `sudo docker run --gpus '"device=0,1,2,3"' --shm-size 1g -p 8080:80 -v $volume:/data ghcr.io/huggingface/text-generation-inference:1.3 --model-id $model` 8) The model will take some time to load... 9) Once loaded the model will be available on port 8080 Sample command within the VM ``` curl 0.0.0.0:8080/generate \ -X POST \ -d '{"inputs":"<|system|>You are a friendly chatbot.\n<|user|>What type of model are you?\n<|assistant|>","parameters":{"do_sample": true, "max_new_tokens": 100, "repetition_penalty": 1.15, "temperature": 0.7, "top_k": 20, "top_p": 0.9, "best_of": 1}}'\ -H 'Content-Type: application/json' ``` You can also access the model from outside the VM ``` curl IP_ADDRESS_PROVIDED_BY_MASSED_COMPUTE_VM:8080/generate \ -X POST \ -d '{"inputs":"<|system|>You are a friendly chatbot.\n<|user|>What type of model are you?\n<|assistant|>","parameters":{"do_sample": true, "max_new_tokens": 100, "repetition_penalty": 1.15, "temperature": 0.7, "top_k": 20, "top_p": 0.9, "best_of": 1}}'\ -H 'Content-Type: application/json ``` For assistance with the VM join the [Massed Compute Discord Server](https://discord.gg/Mj4YMQY3DA) ## Prompt formatting In sticking with the theme of the bagel, I didn't want to use a single prompt format, so I used 4 - vicuna, llama-2, alpaca, and chat-ml (sorta). I also didn't want to randomly select a single prompt format for each item (hoping each instruction would generalize more when used in a variety of prompt formats), so each instruction is actually converted into every prompt format. This means each epoch of our fine-tune is really basically 4 epochs. So, for the fine-tunes, I would recommend only doing 1 epoch (or 0.75 epochs). I am testing with a single epoch using a relatively low learning rate. ### Alpaca (sort of) ``` Below is an instruction that describes a task. Write a response that appropriately completes the request. ### Instruction: {system prompt, if provided} {instruction} ### Response: ``` The main difference here is that because of the dataset formatting and variety of data sources, it would have been much to tedious to add an `### Input:` block, so the inputs are just in the instruction section. ### Vicuna ``` {system prompt, if provided, randomly defaulting to "A chat between a user and an unbiased, uncensored assistant."} USER: {instruction} ASSISTANT: ``` ### ChatML (sort of) I don't really understand the point of having special tokens for `<|im_start|>` and `<|im_end|>`, because in practice they just act as BOS and EOS tokens (but, please correct me if I'm wrong). So, instead of: ```text {bos}<|im_start|>{role} {text} <|im_end|>{eos} ``` I just changed it to: ```text {bos}{role} {text} {eos} ``` If you *really* want to use `<|im_start|>` and `<|im_end|>`, just update your `tokenizer_config.json` to use `<|im_start|>` instead of `` and `<|im_end|>` instead of `` and when tokenizing. And if you still don't like what I've done to this chat-ml-ish format, feel free to cry into your pillow or fork the code and do a new fine-tune. ### Llama-2 chat ``` [INST] <> {system} <> {instruction} [/INST] ``` ### Default via chat template The model's `tokenizer_config.json` includes the default chat template (llama-2), so you can simply use the `apply_chat_template` method to build the full prompt. ``` import transformers tokenizer = transformers.AutoTokenizer.from_pretrained('jondurbin/bagel-8x7b-v0.2') chat = [ {"role": "system", "content": "You are Bob, a friendly AI assistant."}, {"role": "user", "content": "Hello, how are you?"}, {"role": "assistant", "content": "I'm doing great. How can I help you today?"}, {"role": "user", "content": "I'd like to show off how chat templating works!"}, ] print(tokenizer.apply_chat_template(chat, tokenize=False)) ``` ### Contribute If you're interested in new functionality/datasets, take a look at [bagel repo](https://github.com/jondurbin/bagel) and either make a PR or open an issue with details. To help me with the fine-tuning costs (which are extremely expensive for these large combined datasets): - https://bmc.link/jondurbin - ETH 0xce914eAFC2fe52FdceE59565Dd92c06f776fcb11 - BTC bc1qdwuth4vlg8x37ggntlxu5cjfwgmdy5zaa7pswf ### Guide for certain tasks #### RA(G)/contextual question answering The model was trained to ignore what it thinks it knows, and uses the context to answer the questions, when using the format below. The model was also tuned to limit the values to the provided context as much as possible to reduce hallucinations. The format for a contextual prompt is as follows: ``` BEGININPUT BEGINCONTEXT [key0: value0] [key1: value1] ... other metdata ... ENDCONTEXT [insert your text blocks here] ENDINPUT [add as many other blocks, in the exact same format] BEGININSTRUCTION [insert your instruction(s). The model was tuned with single questions, paragraph format, lists, etc.] ENDINSTRUCTION ``` I know it's a bit verbose and annoying, but after much trial and error, using these explicit delimiters helps the model understand where to find the responses and how to associate specific sources with it. - `BEGININPUT` - denotes a new input block - `BEGINCONTEXT` - denotes the block of context (metadata key/value pairs) to associate with the current input block - `ENDCONTEXT` - denotes the end of the metadata block for the current input - [text] - Insert whatever text you want for the input block, as many paragraphs as can fit in the context. - `ENDINPUT` - denotes the end of the current input block - [repeat as many input blocks in this format as you want] - `BEGININSTRUCTION` - denotes the start of the list (or one) instruction(s) to respond to for all of the input blocks above. - [instruction(s)] - `ENDINSTRUCTION` - denotes the end of instruction set __Use a very low temperature!__ Here's a trivial, but important example to prove the point: ``` BEGININPUT BEGINCONTEXT date: 2021-01-01 url: https://web.site/123 ENDCONTEXT In a shocking turn of events, blueberries are now green, but will be sticking with the same name. ENDINPUT BEGININSTRUCTION What color are bluberries? Source? ENDINSTRUCTION ``` And the response: ``` Blueberries are now green. Source: date: 2021-01-01 url: https://web.site/123 ``` #### Summarization 500 samples have been included from [this dataset](https://huggingface.co/datasets/mattpscott/airoboros-summarization), using the same format as contextual question answering, for example: ``` BEGININPUT {text to summarize} ENDINPUT BEGININSTRUCTION Summarize the input in around 130 words. ENDINSTRUCTION ``` #### Agent/function calling The dataset includes many examples of function/args generation based on input criteria. This is somewhat similar to the OpenAI function calling, but the output is either JSON or YAML. Example prompt: ``` As an AI assistant, please select the most suitable function and parameters from the list of available functions below, based on the user's input. Provide your response in JSON format. Input: I want to know how many times 'Python' is mentioned in my text file. Available functions: file_analytics: description: This tool performs various operations on a text file. params: action: The operation we want to perform on the data, such as "count_occurrences", "find_line", etc. filters: keyword: The word or phrase we want to search for. ``` Response: ```json { "function": "file_analytics", "params": { "action": "count_occurrences", "filters": { "keyword": "Python" } } } ``` #### reWOO style execution planning The model now supports execution planning for complex instructions that would require making use of several tools. The output is just the plan, you must implement a mechanism to parse the output and actually call the functions! Example prompt: ``` Please construct a systematic plan to generate an optimal response to the user instruction, utilizing a set of provided tools. Each plan will correspond to an evidence value, which will be the output of one of the available functions given an input string that could be the user's question, one or more prior evidence values, or a combination of both. Here are the tools available to be called: DuckDuckGo[input]: Tool that enables the user to search for information using the DuckDuckGo search engine. This tool is helpful in locating a suitable list of sites that may contain the answer to the user's query. It does not directly answer the question but finds a list of sites that might have the answer. TextScraper[input]: Load one or more websites from the input string containing newline delimited links, where input is one or more links, and produces plain text output containing the content of the links. HyperlinkExtractor[input]: Extract links from plain text and produces a plain text, newline delimited response of links. KnowledgeModel[input]: Question answering language model, particularly useful in answering questions based on an input passage of text. The input must be a text question that references an :evidence[n]: variable, e.g. What color is the cat, given :evidence1:? The output should be in format: Plan: [first action to take based on input question] :evidence0: = [function to call with input parameter] Plan: [next action to take, based on result of :evidence0:] :evidence1: = [next function to call and input parameter, which may include reference to previous evidence, e.g. "Given context :evidence0"] ... Answer: [:evidence[n]: containing the final answer.] Question: Who won the Nobel Prize in Literature in 2020 and what is their most famous work? ``` Response: ``` Plan: Begin by conducting a web search to find out who won the Nobel Prize in Literature in 2020. :evidence0: = DuckDuckGo[Nobel Prize in Literature 2020 winner] Plan: Utilize the KnowledgeModel to interpret the search results and find the name of the laureate. :evidence1: = KnowledgeModel[Who won the Nobel Prize in Literature in 2020, given :evidence0:?] Plan: Conduct another web search to find the most famous work of the identified laureate. :evidence2: = DuckDuckGo[Most famous work of :evidence1:] Plan: Extract the relevant links from the DuckDuckGo search results for a more focused search. :evidence3: = HyperlinkExtractor[:evidence2:] Plan: Use the TextScraper tool to extract information from the relevant links. :evidence4: = TextScraper[:evidence3:] Plan: Finally, utilize the KnowledgeModel to identify and summarize the most famous work of the laureate from the extracted information. :evidence5: = KnowledgeModel[What is the most famous work of :evidence1:, given :evidence4:?] Answer: :evidence5: ``` For this to be useful, you'd have to parse the output plan text, and implement/call each of the functions. This is just pseudo-code, completely untested off the top of my head, and obviously would requiring full implementation + hardening: ```python import re import requests def inject_context(input_text, **context): for ref in set(re.findall(r"(:evidence[0-9]+:)", input_text, re.I)): input_text = input_text.replace(ref, context.get(ref, "")) return input_text def duckduckgo(input_text, **context): search_string = inject_context(input_text, **context) ... search via duck duck go using search_string ... return text content def link_extractor(input_text, **context): input_text = inject_context(input_text, **context) return "\n".join(list(set(re.findall(r"(https?://[^\s]+?\.?)", input_text, re.I)))) def scrape(input_text, **context): input_text = inject_context(input_text, **context) text = [] for link in input_text.splitlines(): text.append(requests.get(link).text) return "\n".join(text) def infer(input_text, **context) prompt = inject_context(input_text, **context) ... call model with prompt, return output def parse_plan(plan): method_map = { "DuckDuckGo": duckduckgo, "HyperlinkExtractor": link_extractor, "KnowledgeModel": infer, "TextScraper": scrape, } context = {} for line in plan.strip().splitlines(): if line.startswith("Plan:"): print(line) continue parts = re.match("^(:evidence[0-9]+:)\s*=\s*([^\[]+])(\[.*\])\s$", line, re.I) if not parts: if line.startswith("Answer: "): return context.get(line.split(" ")[-1].strip(), "Answer couldn't be generated...") raise RuntimeError("bad format: " + line) context[parts.group(1)] = method_map[parts.group(2)](parts.group(3), **context) ``` ### Fine-tuning information You can find charts, and the full configuration used to fine-tune this model on [weights and biases](https://wandb.ai/jondurbin/bagel-8x7b-v0.2/runs/agxjjdso?workspace=user-jondurbin) The model was fine-tuned on an 8x a6000 instance, for 4 days, 15 hours, 6 minutes and 42 seconds. ### Licence and usage restrictions The base model is mixtral-8x7b-v0.1, which is licensed as apache-2.0 - no issues there. The fine-tuning data, however, includes several datasets that have data generated at least in part by OpenAI's gpt-4. I am not a lawyer, so I can't help determine if this is actually commercially viable, but some questions that often come up are: - Does the OpenAI ToS apply only to the user who created the dataset initially, and not subsequent models? - If the dataset was released under a permissive license, but actually includes OpenAI generated data, does that ToS supersede the license? - Does the dataset fall completely under fair use anyways, since the model isn't really capable of reproducing the entire training set verbatim? Use your best judgement and seek legal advice if you are concerned about the terms. In any case, by using this model, you agree to completely indemnify me.