问题 / #3567

#3567 [Other] Resumable model download stalls silently (~69 % of shard 2): no progress, no message, no running process

open · @ogerly · 2 评论 · 去 GitHub 看

Setup & installWindows

说明

During the ~58 GB model download (two shards, resumable) Shard 1 completed and verified. Shard 2 then stood at 69 % for hours:

- no process running anymore,
- no error message,
- `.part.chunks` did not move further,
- the command had exited normally (EXITCODE 0) - so nothing looked broken.

Re-running exactly the same setup.py command resumed cleanly from the existing chunk files and finished (Shard 2 verified afterwards).

So the resume-mechanism is good - the STALL and the silent exit are the problem.

Suggestion: after the resumable download, verify completion and
a) on incomplete state: print "download NOT finished - rerun this exact command to continue" instead of exiting silently,
b) optionally a heartbeat (e.g. "stalled 5 min at 69 % - retry chunk x/y") or automatic retry for chunks that received no bytes within a timeout.

Environment: Windows 11, E: drive (~165 MB/s write), network-bound download via Python, `--no-start` mode.

Strata version: 0.1.45.3 (engine built from source)

本站相关内容

相关页面的快捷入口。