Skip to content

pullrequests Search Results · bucket language:JavaScript language:JavaScript language:JavaScript language:Shell language:Python

Filter by

971k results  (513 ms)

971k results

dependencies
python:uv

Summary - compile SM90 sparse indexer-score and sparse attention-score kernels during plan preparation from symbolic tensors - bound specialization to 17 physical Top-K capacity buckets while ...
area:sparse_attention
cat-enhancements
op: DSA
orig-external

... that refuse to overwrite an existing object reject that second upload, so every backup ends as FAILED even though all of its data was uploaded. Two examples: a Google Cloud Storage bucket with a ...

Summary - Add an Apache CSP header file under the Vite public directory so it is copied into the production build. - Replace process-local rate-limit buckets with atomic PostgreSQL counters ...

... ) | 1.31 | 1.20 | ≈ | | | | | BiDijkstra (ms) | 1.04 | 0.962 | ≈ | | | | | A* Haversine (ms) | 0.609 | 0.567 | ≈ | | | | | Buckets (ms) | 0.413 | 0.379 | ≈ | | | | | BiBuckets (ms) | 0.302 ...

... bucket, documented with Cloudflare R2. It uses boto3, loaded only when backups are on. - Off by default. All four BACKUP_S3_* values turn it on, and setting only some raises ConfigError at ...

... | | --- | --- | | VERL_WEIGHT_SYNC_DOUBLE_BUFFER=1 | BucketedWeightSender uses two CUDA IPC buffers and keeps one bucket in flight, so vLLM load_weights of bucket i overlaps the gather and packing of bucket i+1 ...

... W instead of 395 W. The reading and count now use the same total or per-phase bucket, including phase-D participants. This also exposed a live CT002 response mismatch: an individual active-control ...

Summary Opt-in MiniMax-M3 prefill token graphs (ATOM_MINIMAX_M3_PREFILL_TOKEN_GRAPH=1, off by default). A prefill no longer than the largest ATOM_TOKEN_GRAPH_SIZES bucket pads to the smallest bucket ...