add dsv4-fp4-gb300-cw-dynamo-vllm-agentic — CoreWeave sibling config

Mirrors dsv4-fp4-gb300-dynamo-vllm-agentic 1:1 except runner switches
from gb300-nv to gb300-cw. Same image, search space (conc 32/192/4096),
and recipe files. Recipe sharing works because launch_gb300-cw.sh
already has the IS_AGENTIC overlay branch (mirrors launch_gb300-nv.sh)
that copies benchmarks/multi_node/srt-slurm-recipes/vllm/deepseek-v4/
agentic into the srt-slurm clone.

Separate config (not a runner-label widening on the -nv entry) so we
can dispatch NV and CW as independent sweep runs — bundling SKUs in
one `gh workflow run` causes fault cascades per
[[feedback_separate_b200_b300_runs]].

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

Author: cquil11

.github/configs/nvidia-master.yaml

@@ -8781,6 +8781,74 @@ dsv4-fp4-gb300-dynamo-vllm-agentic:
           ep: 8
           dp-attn: true
 
+# CoreWeave sibling of dsv4-fp4-gb300-dynamo-vllm-agentic — same image,
+# recipes, and search space; only `runner` differs (gb300-cw vs gb300-nv).
+# Kept as a separate config (not a label-widening on the -nv entry)
+# because we dispatch NV and CW as independent sweep runs — bundling
+# both SKUs into one `gh workflow run` invocation lets a fault on one
+# cascade-cancel the other (see prior R20–R23 outages). The two sibling
+# configs share recipe files via the same launch_gb300-cw.sh IS_AGENTIC
+# overlay (recipes/vllm/deepseek-v4/agentic/), so a change to the recipe
+# applies to both clusters with no duplication.
+dsv4-fp4-gb300-cw-dynamo-vllm-agentic:
+  image: vllm/vllm-openai:v0.21.0-ubuntu2404
+  model: deepseek-ai/DeepSeek-V4-Pro
+  model-prefix: dsv4
+  runner: gb300-cw
+  precision: fp4
+  framework: dynamo-vllm
+  multinode: true
+  disagg: true
+  scenarios:
+    agentic-coding:
+    - duration: 1800
+      search-space:
+      # Low-latency: 1p6d at conc=32.
+      - spec-decoding: none
+        conc-list: [32]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/deepseek-v4/agentic/disagg-gb300-1p6d-dep4-tp4-agentic.yaml"
+        decode:
+          num-worker: 6
+          tp: 4
+          ep: 1
+          dp-attn: false
+      # Mid: 1p6d at conc=192.
+      - spec-decoding: none
+        conc-list: [192]
+        prefill:
+          num-worker: 1
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/deepseek-v4/agentic/disagg-gb300-1p6d-dep4-tp4-agentic.yaml"
+        decode:
+          num-worker: 6
+          tp: 4
+          ep: 1
+          dp-attn: false
+      # High-throughput: 4p1d at conc=4096.
+      - spec-decoding: none
+        conc-list: [4096]
+        prefill:
+          num-worker: 4
+          tp: 4
+          ep: 4
+          dp-attn: true
+          additional-settings:
+          - "CONFIG_FILE=recipes/vllm/deepseek-v4/agentic/disagg-gb300-4p1d-dep4-dep8-24-c4096-agentic.yaml"
+        decode:
+          num-worker: 1
+          tp: 8
+          ep: 8
+          dp-attn: true
+
 dsv4-fp4-gb300-dynamo-sglang:
   image: lmsysorg/sglang-staging:deepseek-v4-grace-blackwell-dev
   model: deepseek-ai/DeepSeek-V4-Pro