Skip to content

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

Filter by

1.1M results  (458 ms)

1.1M results

Motivation get_batch_sizes_to_capture drops the top decode bucket when --cuda-graph-max-bs is below the request pool and not a multiple of the attention-TP alignment. Example: DP attention with attn ...
blackwell
lora

... preserveAspectRatio= none . It had no axes, it distorted at every width, and because the API skipped empty buckets, quiet periods disappeared and the time axis wasn t linear. Changes Server (timeline) - Zero ...

Summary Partial implementation of #6609 — blocked by the current maintainer review. Workers deployed behind a NAT or load-balancer share one peer IP, so the existing IP-keyed default bucket gives ...
core
size/m
tests

... the target weight. Every rejection restarts the entire proposal, including masked slots and final-block padding. Integer bucket masses avoid a floating-point proposal CDF. The kernel makes one ...
bug
ci/build
cpu
needs-rebase
performance
scheduler