[recipe] feat: Add DeepSeek-V4 training smoke configs#3923
Open
Meirtz wants to merge 2 commits into
Open
Conversation
Signed-off-by: Lingrui Mei <lmei@nvidia.com>
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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 ?
Adds minimal DeepSeek-V4 training smoke configs for exercising selected Megatron Core DSv4 architecture paths through Megatron Bridge. These configs are intended for short validation runs, not full production pretraining recipes.
Changelog
deepseek_v4_tiny_pretrain_configfor a 1-layer DSv4 hybrid-attention smoke path.deepseek_v4_flash_proxy_pretrain_configfor DSv4 hybrid attention with mHC and hash-routed MoE.deepseek_v4_flash_mtp_proxy_pretrain_configfor DSv4 hybrid attention with mHC and MTP.GitHub Actions CI
See the CI section in the Contributing doc for how to trigger the CI. A Nvidia developer will need to approve and trigger the CI for external contributors.
Local validation:
uv run --no-sync --with pre-commit pre-commit run --all-filesBefore your PR is "Ready for review"
Pre checks:
Additional Information
This PR intentionally limits the scope to smoke configs and tests. It does not claim full DeepSeek-V4 training support.
The recipe configs currently depend on the Megatron Core dev pin because the required DSv4 primitives are not yet fully available in the Megatron Core main pin.