b200 test

Author: Oseltamivir

.github/workflows/eval-gms8k.yml

@@ -49,10 +49,10 @@ jobs:
     uses: ./.github/workflows/eval-tmpl.yml
     secrets: inherit
     with:
-      runner: b200-nvd_2
-      image: ${{ inputs.image || 'nvcr.io#nvidia/tensorrt-llm/release:1.2.0rc0.post1' }}
+      runner: b200-nvd_3
+      image: ${{ inputs.image || 'vllm/vllm-openai:v0.11.0' }}
       model: ${{ inputs.model || 'openai/gpt-oss-120b' }}
-      framework: trt
+      framework: vllm
       precision: fp4
       exp-name: ${{ inputs.exp-name || 'gptoss_gsm8k_poc' }}
       tp: '4'

benchmarks/gptoss_fp4_b200_docker.sh

@@ -79,7 +79,9 @@ run_benchmark_serving \
 
 # After throughput, run evaluation only if RUN_EVAL is true
 if [ "${RUN_EVAL}" = "true" ]; then
-    run_eval --framework lm-eval --port "$PORT" --concurrent-requests $(( $CONC * 2 ))
+    run_eval --framework lm-eval --port "$PORT" --concurrent-requests $(( $CONC * 2 ))  --results-dir /workspace/
     append_lm_eval_summary
 fi
+
+ls -lt /workspace/
 set +x