bagel-dpo-2.8b-v0.2

A bagel, with everything

Overview

An experimental fine-tune of mamba-2.8b-slimpj using 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!

########## 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

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("</s>", "").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
  • Abstraction and reasoning dataset, useful in measuring "intelligence" to a certain extent.
• airoboros
  • Variety of categories of synthetic instructions generated by gpt-4.
• apps
  • Python coding dataset with 10k problems.
• belebele
  • Multi-lingual reading comprehension dataset.
• bluemoon
  • Roleplay data scraped from Bluemoon, then cleaned and formatted as ShareGPT.
• boolq
  • Corpus of yes/no questions (which can be surprisingly difficult for AI to answer apparently?)
• capybara
  • Multi-turn dataset used to create the capybara models.
• cinematika (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
  • More reading comprehension.
• emobank
  • Emotion annotations using the Valence-Arousal-Domninance scheme.
• gutenberg (plain text)
  • Books/plain text, again to make the model less boring, only a handful of examples supported by chapterize
• 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
  • Composite dataset with a variety of math-related tasks and problem/question formats.
• mmlu
  • Massive Multitask Language Understanding - a wide variety of questions about various subject matters.
• natural_instructions
  • Millions of instructions from 1600+ task categories (sampled down substantially, stratified by task type)
• openbookqa
  • Question answering dataset.
• pippa
  • Deduped version of PIPPA in ShareGPT format.
• piqa
  • Phyiscal interaction question answering.
• python_alpaca
  • Python instruction response pairs, validated as functional.
• rosetta_code
  • Code problems and solutions in a variety of programming languages taken from rosettacode.org.
• slimorca
  • Collection of ~500k gpt-4 verified chats from OpenOrca.
• spider
  • SQL-targeted dataset.
• squad_v2
  • Contextual question answering (RAG).
• synthia
  • GPT-4 generated data using advanced prompting from Migel Tissera.
• winogrande
  • Fill in the blank style prompts.

DPO data sources

• airoboros 3.1 vs airoboros 2.2.1
  • 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
  • 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
  • Another interesting dataset by Intel, which provides various DPO pairs generated from prompts included in the SlimOrca dataset.
• toxic-dpo
  • 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
  • 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
  • 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:

{bos}<|im_start|>{role}
{text}
<|im_end|>{eos}

I just changed it to:

{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 <s> and <|im_end|> instead of </s> 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] <<SYS>>
{system}
<</SYS>>

{instruction} [/INST]

Contribute

If you're interested in new functionality/datasets, take a look at bagel repo and either make a PR or open an issue with details.

To help me with the OpenAI/compute costs:

• https://bmc.link/jondurbin
• ETH 0xce914eAFC2fe52FdceE59565Dd92c06f776fcb11
• BTC bc1qdwuth4vlg8x37ggntlxu5cjfwgmdy5zaa7pswf