Add tools/prepare_cache.py for offline GPT dataset cache preparation#4080
Merged
asolergi-nv merged 19 commits intoNVIDIA:mainfrom Apr 29, 2026
Merged
Add tools/prepare_cache.py for offline GPT dataset cache preparation#4080asolergi-nv merged 19 commits intoNVIDIA:mainfrom
asolergi-nv merged 19 commits intoNVIDIA:mainfrom
Conversation
Contributor
|
This PR has been automatically converted to draft because all PRs must start as drafts. When you are ready for review, click Ready for Review to begin the review process. This will:
See the contribution guide for more details. |
Contributor
Author
|
/ok to test d72afb6 |
Contributor
Author
|
/ok to test c87feda |
Contributor
Author
|
/ok to test 2cd79f9 |
BlendedDataset.__init__ called torch.distributed.get_rank() without first checking torch.distributed.is_initialized(), which crashes when running without distributed (e.g. tools/prepare_cache.py on a cold cache with blended datasets). Add the same is_initialized() guard already used in GPTDataset. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
deepakn94
reviewed
Apr 22, 2026
| cache_hit = False | ||
|
|
||
| if not path_to_cache or (not cache_hit and torch.distributed.get_rank() == 0): | ||
| if not path_to_cache or ( |
Contributor
There was a problem hiding this comment.
Why do we need and torch.distributed.get_rank() == 0 when we're already using log_single_rank?
Contributor
Author
There was a problem hiding this comment.
I think you're overlooking the following lines of code. Could you expand them and take a look? I'm just adding the torch.distributed.is_initialized() check for protection.
deepakn94
approved these changes
Apr 22, 2026
maanug-nv
reviewed
Apr 22, 2026
…_cache # Conflicts: # pretrain_mamba.py
…ache
- Call update_train_iters() in build_dataset_caches so jobs using
--train-samples (without --train-iters) no longer trip the
eval-samples assertion in get_train_valid_test_num_samples().
- Reject --step-batch-size-schedule in _validate_prepare_cache_args;
the tool does not initialize the microbatch calculator and cannot
support its dynamic batch-size path.
- Document the full unsupported-flags set (--mock-data, --sft,
--fim-data, --step-batch-size-schedule) in the module docstring,
docs/user-guide/data-loading.md, and megatron/core/datasets/readme.md.
- Tests: add --train-samples-only coverage and extend the
unsupported-modes parametrize with --step-batch-size-schedule.
Contributor
Author
|
/ok to test c1641a7 |
Contributor
Author
|
/ok to test ec3cb71 |
ericharper
approved these changes
Apr 29, 2026
svcnvidia-nemo-ci
added
Approved
|
🔄 Merge queue validation started! You can track the progress here: https://github.com/NVIDIA/Megatron-LM/actions/runs/25124057142 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
6 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What does this PR do ?
This PR adds a new offline cache-preparation entrypoint for GPTDataset-based training so dataset caches can be built ahead of time on a CPU-only node instead of forcing rank 0 to build them during training startup.
What this introduces
tools/prepare_cache.pyTests
Adds tests/unit_tests/data/test_prepare_cache.py covering:
Contribution process
Pre-checks
Code review
Feel free to message or comment the @mcore-oncall to help accelerate your merge into main. The less complex your PR is, the faster it will be approved and merged!
All PRs start as draft. If you open a non-draft PR, it will be automatically converted to draft.
Step 1: Mark PR as "Ready for Review"
.github/CODEOWNERS.Final Review might get declined if these requirements are not fulfilled.
Step 2: Final Review
For PRs that change
megatron/core, once all expert reviewers have approved, theFinal Reviewlabel is applied automatically and final reviewers are assigned.For PRs outside
megatron/core, this step is skipped.Step 3: Approved
Once all required reviewers have approved, the
Approvedlabel is applied automatically.Merge
Any member of mcore-engineers will be able to merge your PR.
For MRs into `dev` branch
The proposed review process for `dev` branch is under active discussion.MRs are mergable after one approval by either
eharper@nvidia.comorzijiey@nvidia.com.