Skip to content

Commit cd160ee

Browse files
committed
fix(profile): rerun Flash vLLM MTP at conc8
1 parent 2e2f876 commit cd160ee

1 file changed

Lines changed: 1 addition & 1 deletion

File tree

.github/configs/nvidia-master.yaml

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -2089,7 +2089,7 @@ dsv4-flash-fp4-b300-vllm-mtp:
20892089
- isl: 1024
20902090
osl: 1024
20912091
search-space:
2092-
- { tp: 8, ep: 8, dp-attn: true, conc-start: 64, conc-end: 64, spec-decoding: mtp }
2092+
- { tp: 8, ep: 8, dp-attn: true, conc-start: 8, conc-end: 8, spec-decoding: mtp }
20932093

20942094
# Targeted Flash MTP profile: DEP4 at the same 1k1k conc=64 point as the
20952095
# non-MTP Flash profile above. The shared SGLang MTP launcher selects the

0 commit comments

Comments
 (0)