Repository navigation
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/25009
Note: Links to docs will display an error until the docs builds have been completed. ❌ 134 New Failures, 2 Cancelled Jobs, 7 Unrelated Failures, 10 Unclassified FailuresAs of commit 95d13f3 with merge base c7f7931 ( NEW FAILURES - The following jobs have failed:
UNCLASSIFIED FAILURES - DrCI could not classify the following jobs because the workflow did not run on the merge base. The failures may be pre-existing on trunk or introduced by this PR:
CANCELLED JOBS - The following jobs were cancelled. Please retry:
FLAKY - The following jobs failed but were likely due to flakiness present on trunk:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
This was referenced Nov 13, 2026
metascroy
force-pushed
the
offgraph-kv-ggml-index-rows
branch
from
November 14, 2026 00:18
c0d7b51 to
7f88bcc
Compare
metascroy
force-pushed
the
offgraph-kv-selected-logits
branch
from
November 14, 2026 00:18
3f8d4b7 to
4ab7569
Compare
The off-graph export returned the last token's logits, which a forward over one sequence needs. A batched forward packs several sequences' tokens, and each wants the logits of its own last token. --logits_to_keep selected adds a third int64[K] input of token indices, passed to Hugging Face's logits_to_keep, so the method returns those rows ([1, K, vocab]) without running the LM head over every token. The package records the mode as get_logits_to_keep_mode, as MLX packages do, and Muse Glimmer's text-only module takes the index tensor too. The native runner feeds a selected method the last token's index, so these packages still run and verify their golden samples standalone.
metascroy
force-pushed
the
offgraph-kv-selected-logits
branch
from
November 14, 2026 00:38
4ab7569 to
95d13f3
Compare
metascroy
force-pushed
the
offgraph-kv-ggml-index-rows
branch
from
November 14, 2026 00:38
7f88bcc to
90a4e77
Compare
This branch had an error being deployed
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.
Stack
Off-graph KV cache for the native ggml engine, through serving Muse Glimmer on it. Review bottom-up; each PR builds and passes its tests on its own.
Summary
The off-graph export returned the last token's logits, which is what a forward over one sequence needs. A batched forward packs several sequences' tokens, and each wants the logits of its own last token.
--logits_to_keep selectedadds a thirdint64[K]input of token indices, passed to Hugging Face'slogits_to_keep, so the method returns those rows ([1, K, vocab]) without running the LM head over every token. The package records the mode asget_logits_to_keep_mode, as MLX packages do. Muse Glimmer's text-only module takes the index tensor too.Test plan
test_export_hf_llm.pyon transformers 5.0.0rc1 and 5.22.2: selected golden samples match last-token ones exactly, the export takes three inputs and returns the selected rows, and the metadata records the mode.run_hf_llm_ggmlon CPU and Metal.Authored with Claude Code.