pullrequests Search Results · bucket language:JavaScript language:Java language:Python language:JavaScript language:JavaScript
Filter by
1M results
Why
The repair script from #2 scans disk folders only. The production bucket holds 933,508 objects (643 GB), of which
915,311 exist only in the bucket. Local disk has 18,197, all also in the bucket ...
First demo PR. Its signed decision receipt was produced locally with the same pipeline the Action runs (see the README
section A receipt you can verify right now ). The PR stays a draft because this repository ...
Motivation
bdc1acf6cd (#2761, 2025-01-07) moved the first bucket of sglang:time_to_first_token_seconds from 0.001 s to 0.1 s and
stretched the tail from 30 s to 160 s, without saying why. Since then ...
An exception in delete_s3_files currently stops cleanup of all remaining batches and buckets.
Catch and log failures inside both loops so cleanup can continue. Batch failures include the bucket and object keys in
the error log; bucket setup failures include the bucket name.
backport-master
... . When GB200 testing
is available for the latest dev suite, remove 11 complete H100 buckets before runner allocation and route selected cases
from mixed buckets exclusively to GB200. The selection ...
Disable selective unit tests
Changed allgather_bucket_size from 5e8 to 1e8 in zero_optimization settings.
Sourcery 摘要
增强功能:
- 将 DeepSpeed ZeRO 的 all-gather bucket 大小从 5e8 减小至 1e8。
details summary Original summary in ...