ci: add DSpark and DFlash2 GPU smoke coverage - #179
Conversation
Signed-off-by: torchspec-bot <262938024+torchspec-bot@users.noreply.github.com>
Signed-off-by: torchspec-bot <262938024+torchspec-bot@users.noreply.github.com>
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e8b044597c
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| step_count_pattern = re.compile(r"num_train_steps=(\d+)") | ||
| step_counts = [int(match.group(1)) for line in lines if (match := step_count_pattern.search(line))] | ||
| if not step_counts: | ||
| raise SystemExit(f"{lane}: training log did not report num_train_steps") | ||
| expected_step_count = step_counts[-1] |
There was a problem hiding this comment.
For the normal non-profile run, deriving expected_step_count from the training process's own num_train_steps log makes the completion check circular. If preprocessing, dataset loading, or step calculation regresses and retains only one of the 1,001 fixture records, the process reports one step and this check accepts it. Preserve the independent fixture-based expectation when TORCHSPEC_CI_PROFILE_NUM_STEPS is unset, using the profile override only for intentionally shortened runs.
Useful? React with 👍 / 👎.
Signed-off-by: Yubo Wang <yubowang2019@gmail.com>
2 participants
Summary
Validation
bash -nfor CI shell scriptspre-commit validate-configgit diff --checkGPU Slurm integration is intended to run through the repository CI lane.