Default Branch

main 919a6502fe · ci(wheels): notify Lark when builds or publishing fail (#3881) · Updated 2026-10-02 06:32:18 +08:00

ZIP  TAR.GZ

Branches

tom/refactor-miles/op30-135 558b660361 · Split the inference deployment layout out of SglangConfig into SglangScalingConfig · Updated 2026-10-01 15:42:57 +08:00	364 719		ZIP  TAR.GZ
tom/refactor-miles/deliver4-phase6-fix 15e51eaeeb · P6 round 1 / P6-24: Require a transport failure during receiver fault dispatch · Updated 2026-10-01 15:33:00 +08:00	15 237		ZIP  TAR.GZ
tom/refactor-miles/op47-87 15e51eaeeb · P6 round 1 / P6-24: Require a transport failure during receiver fault dispatch · Updated 2026-10-01 15:33:00 +08:00	15 237		ZIP  TAR.GZ
tom/refactor-miles/op30-137 668b3a35a0 · Wait for the declared engine cell count before configuring the dumper · Updated 2026-10-01 15:32:05 +08:00	364 718		ZIP  TAR.GZ
tom/refactor-miles/op30-120 8f376ce757 · Carry num_workers_per_cell on the worker launch context · Updated 2026-10-01 15:32:04 +08:00	364 714		ZIP  TAR.GZ
tom/refactor-miles/op30-121 40b4b2d086 · Decide the eval fleet from the inference controller's servers · Updated 2026-10-01 15:32:04 +08:00	364 717		ZIP  TAR.GZ
tom/refactor-miles/op30-133 e5498f54e5 · Gate the trainer controller's startup on --trainer-init-expected-num-cells · Updated 2026-10-01 15:32:04 +08:00	364 716		ZIP  TAR.GZ
tom/refactor-miles/op30-134 76f92a9e12 · Take the trainer rank's world size from the launch context · Updated 2026-10-01 15:32:04 +08:00	364 715		ZIP  TAR.GZ
tom/refactor-miles/op30-132 e96a44dbc8 · Compute a spec's scheduling from the ScalingConfig instead of storing it · Updated 2026-10-01 15:31:58 +08:00	364 713		ZIP  TAR.GZ
tom/refactor-miles/op30-113 48ab4fdf48 · Describe the eval fleet by its observed engine GPU counts · Updated 2026-10-01 15:30:19 +08:00	364 709		ZIP  TAR.GZ
tom/refactor-miles/op30-116 66588835af · Move the scaling fields into a ScalingConfig only AllConfig owns · Updated 2026-10-01 15:30:19 +08:00	364 710		ZIP  TAR.GZ
tom/refactor-miles/op30-119 35e3b8338a · Bound generation concurrency by the current engine count · Updated 2026-10-01 15:30:19 +08:00	364 708		ZIP  TAR.GZ
tom/refactor-miles/op30-131 5bbba0ea21 · Hand a ScalingConfig to every reader of a spec's scheduling · Updated 2026-10-01 15:30:19 +08:00	364 711		ZIP  TAR.GZ
tom/refactor-miles/op30-136 98fd54e61a · Extract the trainer's GPU count and cell count into shared helpers · Updated 2026-10-01 15:30:19 +08:00	364 712		ZIP  TAR.GZ
tom/refactor-miles/op30-112 ff89d7fa01 · Size the http client and throughput metrics from the observed engine topology · Updated 2026-10-01 15:29:41 +08:00	364 707		ZIP  TAR.GZ
tom/refactor-miles/op47-86 66947413be · P6 round 1 / P6-27: Enable all-gather fault hooks for the explicit TP2 mode · Updated 2026-10-01 15:28:51 +08:00	15 236		ZIP  TAR.GZ
tom/refactor-miles/op30-108 030d773332 · Size weight transfer from observed engine GPU counts and offsets · Updated 2026-10-01 15:28:27 +08:00	364 702		ZIP  TAR.GZ
tom/refactor-miles/op30-109 0fd98916eb · Stop requiring external engines to add up to --rollout-num-gpus · Updated 2026-10-01 15:28:27 +08:00	364 703		ZIP  TAR.GZ
tom/refactor-miles/op30-110 8b9877997b · Resolve the initial expected engine cell count when parsing arguments · Updated 2026-10-01 15:28:27 +08:00	364 705		ZIP  TAR.GZ
tom/refactor-miles/op30-111 e99f9b68be · Carry the observed inference topology on the rollout configuration · Updated 2026-10-01 15:28:27 +08:00	364 706		ZIP  TAR.GZ