Allow DCP to load Float8 MoE training weights - #4782
Open
ivy-zhou wants to merge 1 commit into
Open
Conversation
Summary: Register `Float8TrainingWeightWrapperTensor` and its serialized configuration types as safe globals so the DCP `weights_only=True` reader can restore Float8 MoE training weights. Add direct and DCP round-trip coverage. This fixes a general DCP deserialization compatibility gap exposed by TorchTitan, not a TorchTitan-specific reader bug. The quantized DeepSeek-V3.1 671B MoE configuration wraps expert weights in `Float8TrainingWeightWrapperTensor`. Both native DCP resume and DCP-to-`torch_checkpointing` compatibility loading ultimately use the DCP `FileSystemReader`, which restores payloads with `torch.load(..., weights_only=True)`. Without this registration, loading fails before state-dict adaptation with `Unsupported global`. Differential Revision: D113730611
ivy-zhou
requested review from
andrewor14,
jerryzh168 and
vkuzo
as code owners
August 19, 2026 16:48
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/ao/4782
Note: Links to docs will display an error until the docs builds have been completed. ❌ 2 New Failures, 1 Unclassified FailureAs of commit a7b88b9 with merge base 46e106b ( NEW FAILURES - The following jobs have failed:
UNCLASSIFIED FAILURE - DrCI could not classify the following job because the workflow did not run on the merge base. The failure may be pre-existing on trunk or introduced by this PR:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
@ivy-zhou has exported this pull request. If you are a Meta employee, you can view the originating Diff in D113730611. |
meta-cla
Bot
added
the
CLA Signed
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
1 participant
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.
Summary:
Register
Float8TrainingWeightWrapperTensorand its serialized configuration types as safe globals so the DCPweights_only=Truereader can restore Float8 MoE training weights. Add direct and DCP round-trip coverage.This fixes a general DCP deserialization compatibility gap exposed by TorchTitan, not a TorchTitan-specific reader bug. The quantized DeepSeek-V3.1 671B MoE configuration wraps expert weights in
Float8TrainingWeightWrapperTensor. Both native DCP resume and DCP-to-torch_checkpointingcompatibility loading ultimately use the DCPFileSystemReader, which restores payloads withtorch.load(..., weights_only=True). Without this registration, loading fails before state-dict adaptation withUnsupported global.Differential Revision: D113730611