feat(profile): add B300 DeepSeek V4 Flash config

Author: Oseltamivir

.github/configs/nvidia-master.yaml

@@ -2038,6 +2038,24 @@ dsv4-fp4-b300-sglang:
       - { tp: 8, ep: 8, dp-attn: true, conc-start: 2048, conc-end: 2048 }
       - { tp: 8, ep: 8, dp-attn: true, conc-start: 4096, conc-end: 4096 }
 
+# Targeted single-point Flash config for profile.yml. Keep the existing Pro
+# sweep entry above unchanged; this profile-only key reuses the same B300
+# SGLang launch path at the 1k1k, conc=64 point.
+dsv4-flash-fp4-b300-sglang:
+  image: lmsysorg/sglang:deepseek-v4-b300@sha256:2fec8d7958bb0d53b50d7bf04d6ae6a7de8a35503775826e0550a45dd8c3ee15
+  model: deepseek-ai/DeepSeek-V4-Flash
+  model-prefix: dsv4
+  runner: b300
+  precision: fp4
+  framework: sglang
+  multinode: false
+  scenarios:
+    fixed-seq-len:
+    - isl: 1024
+      osl: 1024
+      search-space:
+      - { tp: 4, ep: 1, conc-start: 64, conc-end: 64 }
+
   # DeepSeek-V4-Pro on B300 with EAGLE/MTP speculative decoding. Recipe is
   # selected inside benchmarks/single_node/dsv4_fp4_b300_sglang_mtp.sh by
   # DP_ATTENTION:

runners/launch_b300-nv.sh

@@ -287,7 +287,7 @@ else
     HF_HUB_CACHE_MOUNT="/data/models"
     if [[ "$MODEL" == "Qwen/Qwen3.5-397B-A17B-FP8" ]]; then
         export MODEL="$HF_HUB_CACHE_MOUNT/${MODEL#*/}"
-    elif [[ "$MODEL_PREFIX" == "dsv4" ]]; then
+    elif [[ "$MODEL_PREFIX" == "dsv4" && "$MODEL" == "deepseek-ai/DeepSeek-V4-Pro" ]]; then
         export MODEL="$HF_HUB_CACHE_MOUNT/dsv4-pro"
     fi
     SQUASH_FILE="/data/home/sa-shared/gharunners/squash/$(echo "$IMAGE" | sed 's/[\/:@#]/_/g').sqsh"