Skip to content

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

Filter by

897k results  (432 ms)

897k results

Adds an opt-in way for the MLPerf DeepSWE recipes to write trainer checkpoints to a zonal GCS Rapid bucket through a GCSFuse mount (the GKE GCS FUSE CSI driver), using colocated_python checkpointing ...

Daily One Minute Curiosity: the token bucket rate limiter with lazy refill (O(1) time and space per user). Includes code.py, code.js (both run and verified) and explanation.md, plus a README index ...

Summary - Adds Episode 06: a daily One Minute Curiosity post on the token bucket rate-limiting algorithm — covering the problem, real-world use cases (APIs, cybersecurity, networking ...

What does this PR do? Reduce overlapping tensor lifetimes during SGLang bucketed weight sync. A non-fused view that overflows the current bucket is now compacted after that bucket has been consumed ...

Summary - Added RateLimitExceededError (models) and a RateLimiter token bucket (rate_limit.py) with acquire() and reset(). - IngestSink.append now admits one token per record before buffering ...
area/cli
area/connectors
area/docs
👤 human
needs a look
size/M

What to check: (1) status.simulations in status.py, which now calls job.job_spec for a project on a bucket, and the ValueError the status command in cli.py turns into exit 2; (2) it is wrong if ...

Split out of #2027 on review (its commit 7). What it saves - bucket_status makes one HeadBucket per probe (anonymous, then authenticated) instead of s3fs _info calls, which can take several ...

What Follow-up to #1281. The shared rate window can now refuse, but only on buckets an operator lists: - TREG_PROVIDER_RATE_ENFORCE takes comma-separated view.rate_bucket keys without the ...
area:ci
area:dashboard
area:docs