fix(profile): switch Flash vLLM MTP to DEP8

Author: Oseltamivir

.github/configs/nvidia-master.yaml

@@ -2073,8 +2073,9 @@ dsv4-flash-fp4-b300-vllm:
       search-space:
       - { tp: 4, ep: 1, conc-start: 64, conc-end: 64 }
 
-# Targeted Flash vLLM MTP profile at the same single-point profile location.
-# The shared vLLM MTP launcher selects 3 speculative tokens for this model.
+# Targeted Flash vLLM MTP DEP8 profile at the same single-point profile
+# location. The shared launcher maps dp-attn=true to DP without TP, and selects
+# 3 speculative tokens for this model.
 dsv4-flash-fp4-b300-vllm-mtp:
   image: vllm/vllm-openai:v0.21.0
   model: deepseek-ai/DeepSeek-V4-Flash
@@ -2088,7 +2089,7 @@ dsv4-flash-fp4-b300-vllm-mtp:
     - isl: 1024
       osl: 1024
       search-space:
-      - { tp: 4, ep: 1, conc-start: 64, conc-end: 64, spec-decoding: mtp }
+      - { tp: 8, ep: 8, dp-attn: true, conc-start: 64, conc-end: 64, spec-decoding: mtp }
 
 # Targeted Flash MTP profile: DEP4 at the same 1k1k conc=64 point as the
 # non-MTP Flash profile above. The shared SGLang MTP launcher selects the