[DO NOT MERGE] ci: instrument HF download paths to diagnose stalls#19352
Closed
rascani wants to merge 1 commit intopytorch:mainfrom
Closed
[DO NOT MERGE] ci: instrument HF download paths to diagnose stalls#19352rascani wants to merge 1 commit intopytorch:mainfrom
rascani wants to merge 1 commit intopytorch:mainfrom
Conversation
CUDA jobs that download HuggingFace model weights have been silently hanging mid-download until the 90-min job timeout fires, with no exception logged and no progress-bar updates. Running theories (library version, hf_xet, empty-chunk filtering) are speculation without instrumentation. Add diagnostic instrumentation to the three known surfaces: - `export_model_artifact.sh`'s two `snapshot_download` python -c calls plus the surrounding env (HF_HUB_VERBOSITY=debug, PYTHONUNBUFFERED=1) - `test_huggingface_optimum_model.py`'s top-level setup, before any import that transitively pulls in huggingface_hub - `mlx.yml`'s inline Voxtral snapshot_download Each Python entry point installs `faulthandler.dump_traceback_later (60, repeat=True)` so a hung process surfaces a stack trace every 60s — pinpointing exactly where execution is stuck. DEBUG-level logging on `huggingface_hub`, `httpx`, `httpcore`, `urllib3` shows the protocol-level conversation. Revert once we have signal from a stalled run. Authored with Claude Code.
meta-cla
Bot
added
the
CLA Signed
This PR needs a
|
digantdesai
added a commit
that referenced
this pull request
May 7, 2026
HuggingFace's Xet storage backend stalls mid-download on CI runners, causing the 90-minute job timeout to fire before model weights finish downloading. Force standard HTTP downloads instead. (from debug logs in #19352)
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.
CUDA jobs that download HuggingFace model weights have been silently hanging mid-download until the 90-min job timeout fires, with no exception logged and no progress-bar updates. Running theories (library version, hf_xet, empty-chunk filtering) are speculation without instrumentation.
Add diagnostic instrumentation to the three known surfaces:
export_model_artifact.sh's twosnapshot_downloadpython -c calls plus the surrounding env (HF_HUB_VERBOSITY=debug, PYTHONUNBUFFERED=1)test_huggingface_optimum_model.py's top-level setup, before any import that transitively pulls in huggingface_hubmlx.yml's inline Voxtral snapshot_downloadEach Python entry point installs
faulthandler.dump_traceback_later (60, repeat=True)so a hung process surfaces a stack trace every 60s — pinpointing exactly where execution is stuck. DEBUG-level logging onhuggingface_hub,httpx,httpcore,urllib3shows the protocol-level conversation.Revert once we have signal from a stalled run.
Authored with Claude Code.