Skip to content

[executorch][native] Decode quantized weights in float32 for precise samples - #24542

Open
pytorchbot wants to merge 1 commit into
gh/digantdesai/193/origfrom
gh/digantdesai/194/orig
Open

pytorchbot wants to merge 1 commit into
gh/digantdesai/193/origfrom
gh/digantdesai/194/orig

Conversation

@pytorchbot

Copy link
Copy Markdown
Collaborator

This PR was created by the merge bot to help merge the original PR into the main branch.
ghstack PR number: #24367 by @digantdesai
^ Please use this as the source of truth for the PR details, comments, and reviews
ghstack PR base: https://github.com/pytorch/executorch/tree/gh/digantdesai/194/base
ghstack PR head: https://github.com/pytorch/executorch/tree/gh/digantdesai/194/head
Merge bot PR base: https://github.com/pytorch/executorch/tree/gh/digantdesai/193/orig
Merge bot PR head: https://github.com/pytorch/executorch/tree/gh/digantdesai/194/orig

@diff-train-skip-merge

…samples

Pull Request resolved: #24367

`precise` used weights dequantized in the method's dtype, so f16 and bf16 references held already-rounded weights while engines decode in float32. The `Gguf` and `Torchao` schemes now decode their weights in float32 before the float64 math and raise on any other torchao weight.

Authored with Claude Code.

@bypass-github-export-checks
@bypass-github-pytorch-ci-checks
@bypass-github-executorch-ci-checks
ghstack-source-id: 463657760
@exported-using-ghexport

Differential Revision: [D128254893](https://our.internmc.facebook.com/intern/diff/D128254893/)
@pytorch-bot

pytorch-bot Bot commented Nov 3, 2026 •

Copy link
Copy Markdown

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/24542

Note: Links to docs will display an error until the docs builds have been completed.

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

❌ 3 New Failures

As of commit a5508ca with merge base 0f8e096 (image):

NEW FAILURES - The following jobs have failed:

This comment was automatically generated by Dr. CI and updates every 15 minutes.

@meta-cla meta-cla Bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Nov 3, 2026

This branch was previously deployed

1 inactive deployment
cadence — a5508ca1 Deployed Nov 3, 2026 by pytorchbot via hifi5s-op-test / hifi5s #36373
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants