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
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
1{bos}<|im_start|>{role}
2{text}
3<|im_end|>{eos}
I just changed it to:
text
1{bos}{role}
2{text}
3{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.
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 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):
The base model is tinyllama, 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.
