Repository navigation
Conversation
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/24951
Note: Links to docs will display an error until the docs builds have been completed. ❌ 136 New Failures, 7 Unrelated Failures, 10 Unclassified FailuresAs of commit 9093078 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:
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. |
This was referenced Nov 12, 2026
meta-cla
Bot
added
the
CLA Signed
This was referenced Nov 12, 2026
export_hf_llm loads models with AutoModelForCausalLM, which has no Muse Glimmer: Transformers builds it only as MuseGlimmerForConditionalGeneration, with a vision tower and a scan of input_ids for image and video tokens that the ggml engine cannot run. Build it with AutoModelForImageTextToText and export MuseGlimmerTextLM instead: its language model and LM head under their Hugging Face parameter names, with the output multiplier and logit soft cap. --layers also truncates its per-layer layer_rope_theta. It exports only with --offgraph_cache, which carries its sliding windows. With --gguf, a GGUF checkpoint loads through the muse-glimmer mapping. The first four layers of Muse-Glimmer-30B-KQuant-17GB-Q4_K_M.gguf export with every weight packed, and their golden samples pass on the ggml engine on CPU and Metal. The test needs a transformers with Muse Glimmer and skips otherwise.
metascroy
force-pushed
the
offgraph-kv-verify-without-tokenizer
branch
from
November 14, 2026 00:18
c09bc0d to
b5c3fb1
Compare
metascroy
force-pushed
the
offgraph-kv-mg-export
branch
from
November 14, 2026 00:18
e701e1d to
9093078
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
Exports Muse Glimmer through the native backend.
MuseGlimmerForConditionalGeneration, not throughAutoModelForCausalLM. That class carries a vision tower and scansinput_idsfor image and video tokens, which the ggml engine can't run. The export builds it withAutoModelForImageTextToTextand exportsMuseGlimmerTextLMinstead: its language model and LM head under their HF parameter names, with the output multiplier and logit soft cap.--layers: also truncates its per-layerlayer_rope_theta.--offgraph_cache, which carries its sliding windows.Test plan
test_export_hf_llm.py: off-graph golden samples of a tiny Muse Glimmer match HF's full multimodal model; oneupdate_and_attendper layer and noaten.eq. Skips without a transformers that has Muse Glimmer..ptnwith every weight packed. Its golden samples pass on Metal (max logit error 0.04–0.20) and CPU (0.43–0.62). Greedy continuation of "The capital of France is": " Paris. It is the". About 24 decode steps/s on an M4 Max.Authored with Claude Code.