Skip to content

Commit b9497f1

Browse files
committed
chore: restore B300 Kimi agentic config to both none + cpu
Reverts the temporary cpu-only trim. R3 cpu-only dispatch already in flight at the trimmed SHA; this restores the full sweep for future dispatches.
1 parent 1e9669b commit b9497f1

1 file changed

Lines changed: 1 addition & 3 deletions

File tree

.github/configs/nvidia-master.yaml

Lines changed: 1 addition & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -2638,12 +2638,10 @@ kimik2.5-fp4-b300-vllm:
26382638
- { tp: 4, ep: 1, conc-start: 4, conc-end: 64 }
26392639
# B300 has 288 GB HBM per GPU (vs B200's 192 GB) so the KV-cache cliff
26402640
# sits higher than B200's. Extend the conc sweep to probe up to 64.
2641-
# NOTE: temporarily cpu-only for the R3 retry (R2 had `none` jobs running
2642-
# cleanly at 3 TB offload but `cpu` failed; this run retries cpu at the
2643-
# corrected 2.5 TB pool size). Restore both lines once cpu is validated.
26442641
agentic-coding:
26452642
- duration: 1800
26462643
search-space:
2644+
- { tp: 8, ep: 1, offloading: none, conc-list: [1, 2, 4, 8, 16, 32, 40, 48, 56, 64] }
26472645
- { tp: 8, ep: 1, offloading: cpu, conc-list: [1, 2, 4, 8, 16, 32, 40, 48, 56, 64] }
26482646

26492647
dsr1-fp8-b200-sglang-mtp:

0 commit comments

Comments
 (0)