b200 test

Author: Oseltamivir

.github/workflows/eval-gms8k.yml

@@ -42,7 +42,7 @@ on:
     paths:
       - '.github/workflows/eval-gms8k.yml'
       - '.github/workflows/eval-tmpl.yml' 
-      - 'benchmarks/benchmark_lib.sh' 
+      - 'benchmarks/benchmark_lib.sh'  
 
 jobs:
   eval:

.github/workflows/eval-tmpl.yml

@@ -146,6 +146,7 @@ jobs:
           RUN_MODE: eval
           # Optional: structured filename if runner chooses to use it later
           EVAL_RESULT_BASENAME: ${{ env.EXP_NAME }}_${{ env.PRECISION }}_${{ env.FRAMEWORK }}_${{ runner.name }}
+        working-directory: repo
         run: |
           bash ./runners/launch_${RUNNER_NAME%%_*}.sh
 

benchmarks/gptoss_fp4_b200_docker.sh

@@ -14,7 +14,7 @@
 
 nvidia-smi
 
-rm -rf /home/nvadmin/actions-runner/_work/InferenceMAX/InferenceMAX/gpt-oss-120b/
+rm -rf gpt-oss-120b/
 
 # To improve CI stability, we patch this helper function to prevent a race condition that
 # happens 1% of the time. ref: https://github.com/flashinfer-ai/flashinfer/pull/1779