Assert fp32 for rope embeddings, misc test fixes#1496
Merged
pstjohn merged 2 commits intoNVIDIA:mainfrom Mar 5, 2026
Merged
Conversation
pstjohn
requested review from
cspades,
dorotat-nv,
jomitchellnv,
jstjohn,
jwilber,
savitha-eng and
trvachov
as code owners
Contributor
|
Important Review skippedAuto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: CHILL Plan: Pro Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
savitha-eng
added a commit
that referenced
this pull request
Mar 5, 2026
Self-contained FSDP2 + TransformerEngine recipe for OpenGenome2 training, extracted from the generic llama3_native_te recipe with OG2-specific defaults: - FP32 master weights with MixedPrecisionPolicy (cast_forward_inputs=False) - Megatron-style scaled init for proj/fc2 layers - Spike-No-More embedding initialization (std=1.0) - Genomic masking for degenerate bases - Weight decay grouping (skip bias/1D params) - THD sequence packing with GQA - FP8 training with first/last layer BF16 override - RoPE fp32 assertion (from PR #1496) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
9 tasks
Signed-off-by: Peter St. John <pstjohn@nvidia.com>
Signed-off-by: Peter St. John <pstjohn@nvidia.com>
pstjohn
force-pushed
the
pstjohn/assert-fp32-for-rope-embeddings
branch
from
e609c74 to
a5deb7c
Compare
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This wouldn't have caught @savitha-eng's
cast_forward_inputs=Truebug (that casts these right as they enter the TransformerLayer), but it turns out our test suite was actually casting these to bfloat16 withmodel.to(bfloat16)calls 😬 .This also fixes a few other misc. test failures I saw locally making sure the esm2 & llama3 recipe and model tests pass.
will require #1495 for tests to pass