--- license: apache-2.0 datasets: - ai2_arc - unalignment/spicy-3.1 - 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 --- # A bagel, with everything ![bagel](bagel.png) ## Overview An experimental fine-tune of [mamba-2.8b-slimpj](https://hf.co/state-spaces/mamba-2.8b-slimpj) using [bagel](https://github.com/jondurbin/bagel) Default recommended system prompt: ``` You are a helpful, unbiased, uncensored assistant. ``` Supports several prompt formats, but you can also use `tokenizer.apply_chat_template` This model did surprisingly well on MT-Bench, for a 2.8b that was only pre-trained on the slimpajama dataset! ```text ########## First turn ########## score model turn bagel-dpo-2.8b-v0.2 1 5.10625 ########## Second turn ########## score model turn bagel-dpo-2.8b-v0.2 2 4.7375 ########## Average ########## score model bagel-dpo-2.8b-v0.2 4.921875 ``` ## Example chat script ```python import torch from transformers import AutoTokenizer, AutoModelForCausalLM from mamba_ssm.models.mixer_seq_simple import MambaLMHeadModel device = "cuda" tokenizer = AutoTokenizer.from_pretrained("bagel-final-2.8b-v0.2") model = MambaLMHeadModel.from_pretrained("bagel-final-2.8b-v0.2", device="cuda", dtype=torch.float32) messages = [{"role": "system", "content": "You are a helpful, unbiased, uncensored assistant."}] while True: user_message = input("[INST] ") messages.append({"role": "user", "content": user_message}) input_ids = tokenizer.apply_chat_template(messages, return_tensors="pt", add_generation_prompt=True).to("cuda") out = model.generate(input_ids=input_ids, max_length=2000, temperature=0.9, top_p=0.7, eos_token_id=tokenizer.eos_token_id, repetition_penalty=1.07) decoded = tokenizer.batch_decode(out)[0].split("[/INST]")[-1].replace("", "").strip() messages.append({"role": "assistant", "content": decoded}) print("[/INST]", decoded) ``` ## SFT 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. ## DPO data sources - [airoboros 3.1](https://huggingface.co/datasets/unalignment/spicy-3.1) vs [airoboros 2.2.1](https://huggingface.co/datasets/jondurbin/airoboros-gpt4-1.4.1) - The creative/writing tasks from airoboros-2.2.1 were re-generated using gpt4-0314 and a custom prompt to get longer, more creative, less clichè responses for airoboros 3.1, so we can use the shorter/boring version as the "rejected" value and the rerolled response as "chosen" - [helpsteer](https://huggingface.co/datasets/nvidia/HelpSteer) - Really neat dataset provided by the folks at NVidia with human annotation across a variety of metrics. Only items with the highest "correctness" value were used for DPO here, with the highest scoring output as "chosen" and random lower scoring value as "rejected" - [orca_dpo_pairs](https://huggingface.co/datasets/Intel/orca_dpo_pairs) - Another interesting dataset by Intel, which provides various DPO pairs generated from prompts included in the SlimOrca dataset. - [toxic-dpo](https://huggingface.co/datasets/unalignment/toxic-dpo-v0.1) - __*highly toxic and potentially illegal content!*__ De-censorship, for academic and lawful purposes only, of course. Generated by llama-2-70b via prompt engineering. - [truthy](https://huggingface.co/datasets/jondurbin/truthy-dpo-v0.1) - DPO pairs meant to increase truthfulness of the model, e.g. common misconceptions, differentiate between AI assistants and roleplayed human in terms of corporeal awareness/locality/etc. - [ultrafeedback](https://huggingface.co/datasets/allenai/ultrafeedback_binarized_cleaned) - One of the bits of magic behind the Zephyr model. Only the items with a chosen score of 8 or higher were included. 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). ## 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] ```