mlfoundations / mlfoundations/evalchemy

[Bug] google/codegemma-7b-it Evaluation Results All Zero

Open
#147 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
HTML
Stars
610
Forks
87
PR merge metrics
No merged PRs in 30d

Description

Hi there,

I'm encountering an issue when evaluating the google/codegemma-7b-it model using evalchemy.

Problem Description:
When running evaluations for google/codegemma-7b-it, all the resulting metrics consistently show a value of 0.0, regardless of the dataset or evaluation setup. This suggests that the model might not be generating valid outputs or that there's an issue with how its outputs are being processed and scored by evalchemy.

Steps to Reproduce:

  1. Evaluation script
    CUDA_VISIBLE_DEVICES=0 python -m eval.eval \
      --model vllm \
      --tasks "LiveCodeBenchv5_official,HumanEval,HumanEvalPlus,MBPP,MBPPPlus,CruxEval,CodeElo,BigCodeBench" \
      --model_args "pretrained=/personal/opensource/models/codegemma-7b-it,tensor_parallel_size=1" \
      --batch_size 64 \
      --max_tokens 2048 \
      --apply_chat_template True \
      --output_path logs
    
  2. Observe the evaluation results. All metrics are reported as 0.0, except CruxEval.
Image

Any insights or guidance on resolving this would be greatly appreciated. Thank you!

Bests,
John

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Reproduce the issue with the provided python -m eval.eval command, using the vllm model and listed tasks. Start at the eval.eval entry point and compare generated outputs with the scoring results, including the task that does not report zero. Done means identifying the cause of the zero metrics and verifying corrected results across the affected evaluations.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning, testing-qa
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
30/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.