pullrequests Search Results · bucket language:JavaScript language:JavaScript language:Python language:Java language:JavaScript
Filter by
1M results
Ceph addresses buckets of other tenants as tenant : bucket , which botocore s client-side bucket name validation
rejects. Unregister that validation on the upload client.
When max_num_seqs=10, CUDA Graph capture only creates buckets [1, 2, 4, 8]. A valid decode batch of 9 or 10 requests
then raises StopIteration in run_model. Other non-aligned limits, such as 17 and ...
needs-human
ready-for-review
risk/moderate
The owner deleted warehouse/raw/job_assistant/raw/ on 2026-11-04: 12 objects before, 0 after. A check from this session
confirmed it: the prefix lists 0 objects, and a former object s public r2.dev address ...
... triton.jit specialization, rows_per_block) bucket (1..4).
Claims
- Eliminates 100% of inference-time Triton JIT compilation warnings (2 - 0) reported by jit_monitor.py during
Qwen3.5-4B + DFlash ...
mrv2
qwen
Part of #109. Three fixes from the FSDP audit, one commit each.
bucket reductions start in the same order on every rank
NCCL matches collectives by call order. Each SyncGradientBucket started its ...
put_object() kept the earlier versions of a key only when versioning was Enabled. In a bucket whose versioning is
Suspended, every PUT replaced the whole version list with the new null version, so ...
Changes
- Switch rate limiter to fixed 60-second Unix-epoch-aligned windows per API key.
- Use a Redis Lua script with Redis server time, atomic INCR, and first-request expiry.
- Update in-memory ...