THUDM / THUDM/slime

Is it possible to perform validation during SFT training to monitor overfitting in Slime?

Open
#1,584 0 comments 3 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
8.5k
Forks
1.3k
Avg merge
5h 36m
Merged PRs (30d)
22

Description

Dear Slime Team,

I am using Slime for LLM SFT. I noticed that all the SFT examples you provided lack evaluation settings. I tried various configurations myself but was unable to calculate the validation dataset loss during the training process to observe if the model is overfitting.

I also reviewed your source code. It appears that your framework does not support running validation simultaneously with training, limiting observations to performance on the training dataset only.

Is my understanding correct? If I am mistaken, could you please advise on how to enable validation during SFT?

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

Start by reviewing the SFT examples and the training-related source code referenced in the issue, then trace whether a validation dataset and loss calculation can be configured during training. Done means establishing whether simultaneous validation is supported and, if not, defining the configuration or implementation needed to monitor validation loss and overfitting.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
machine-learning
Issue type
Feature
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.