README.md

metadata

model-index:
  - name: abacaj/mistral-7b-sft
    results:
      - task:
          type: text-generation
        dataset:
          type: openai_humaneval
          name: HumanEval
        metrics:
          - name: pass@1
            type: pass@1
            value: 54.27
            verified: false
      - task:
          type: text-generation
        dataset:
          type: mbpp
          name: MBPP
        metrics:
          - name: pass@1
            type: pass@1
            value: 38
            verified: false
      - task:
          type: text-generation
        dataset:
          type: mmlu
          name: MMLU
        metrics:
          - name: pass@1
            type: pass@1
            value: 45.89
            verified: false
language:
  - en

How to run inference:

import transformers
import torch


def fmt_prompt(prompt: str) -> str:
    return f"""[Instructions]:\n{prompt}\n\n[Response]:"""


if __name__ == "__main__":
    model_name = "abacaj/mistral-7b-sft"
    tokenizer = transformers.AutoTokenizer.from_pretrained(model_name)

    model = (
        transformers.AutoModelForCausalLM.from_pretrained(
            model_name,
        )
        .to("cuda:0")
        .eval()
    )

    prompt = "If A is greater than B and B is greater than C does that make A greater than C?"
    prompt_input = fmt_prompt(prompt)
    inputs = tokenizer(prompt_input, return_tensors="pt").to(model.device)
    input_ids_cutoff = inputs.input_ids.size(dim=1)

    with torch.no_grad():
        generated_ids = model.generate(
            **inputs,
            use_cache=True,
            max_new_tokens=512,
            temperature=0.2,
            top_p=0.95,
            do_sample=True,
            eos_token_id=tokenizer.eos_token_id,
            pad_token_id=tokenizer.pad_token_id,
        )

    completion = tokenizer.decode(
        generated_ids[0][input_ids_cutoff:],
        skip_special_tokens=True,
    )

    print(completion)

Evals:

Code to train model: https://github.com/abacaj/train-with-fsdp