Open
Conversation
For original authorship, see the DFM repo: https://github.com/NVIDIA-NeMo/DFM Original Major Contributors for Megatron path: @abhinavg4, @huvunvidia, @sajadn, and @suiyoubi Migrate the Megatron-based diffusion model code from the DFM repository (commit 013ceca) into Megatron-Bridge as a self-contained `diffusion/` module, following the shallow integration plan. Source mapping: - dfm/src/megatron/ -> src/megatron/bridge/diffusion/ - dfm/src/common/ (utils) -> src/megatron/bridge/diffusion/common/ - examples/megatron/ -> examples/diffusion/ - tests/unit_tests/megatron -> tests/diffusion/unit_tests/ - tests/functional_tests/mcore -> tests/diffusion/functional_tests/ Key structural changes from DFM: - model/ renamed to models/ (matches MB convention) - model/*/conversion/ extracted to top-level conversion/ (separates bridge/checkpoint-conversion code from model implementation) - All dfm.src.megatron.* imports rewritten to megatron.bridge.diffusion.* - All dfm.src.common.* imports rewritten to megatron.bridge.diffusion.common.* - dfm.src.automodel.* imports left as-is (automodel migrating separately) Models included: DiT, FLUX, WAN (video generation)
Author
|
/ok to test a63854c |
F841 (unused variables) -- removed where clearly safe: • neg_t5_embed in inference_dit_model.py • device in inference_flux.py • error in wan/inference/utils.py (cache_image) • n -> _ in wan/rope_utils.py • data_batch, x0_from_data_batch in test_edm_pipeline.py • original_base in test_flux_hf_pretrained.py • total_layers, p_size, vp_size in test_flux_provider.py F841 -- kept with `# noqa: F841` (uncertain intent): • config = get_model_config(model) in flux_step_with_automodel.py • video_latents, loss_mask in flow_matching_pipeline_wan.py D101/D103 (missing docstrings) -- added `# noqa` markers to all 36 class/function definitions across 24 files. Markdown filename -- renamed README_perf_test.md to README-perf-test.md.
Author
|
/ok to test 9242693 |
…e recipes. Introduce new test files for DiT, FLUX, and WAN pretraining, along with necessary configuration and utility files. Update test structure to enhance coverage and maintainability.
Author
|
/ok to test cb822f2 |
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
2 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 ?
Migrate DFM to MB.
PR passing all tests in DFM is here: NVIDIA-NeMo/DFM#105
Before your PR is "Ready for review"
Pre checks:
If you haven't finished some of the above items you can still open "Draft" PR.
Additional Information