mlco2 / mlco2/codecarbon

Incorrect runtimes reported in emissions files

Open
#653 5 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

question
Dominant language
Python
Stars
1.9k
Forks
323
Avg merge
1d 12h
Merged PRs (30d)
12

Description

  • CodeCarbon version: 2.6.0
  • Python version: 3.10
  • Operating System: Linux (Google Colab)
Description

I was trying to record the emissions generated while fine-tuning the Gemma-7B model on an A100 GPU on Google Colab. While the code ran without any errors, the duration reported by the library in the emissions seems to be incorrect.

In two trials, the total duration is reported as 33.5 seconds and 33.2 seconds, respectively, while the actual durations were closer to 30 minutes. I also recorded the emissions subtask-wise, and the durations for individual subtasks are reported as negative (e.g., -1723972816).

Since I did not face this issue with other notebooks I ran on Google Colab with other models and GPUs, as well as other models on the same GPU, I was wondering where the error came from in this scenario? Thanks!

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 report in Google Colab with CodeCarbon 2.6.0, Gemma-7B fine-tuning, and an A100 GPU, comparing total and subtask durations with the actual runtime. Start by tracing how runtime values are recorded for emissions files and subtasks. Done means durations are nonnegative and closely match the observed execution time.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
backend
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.