Eval run for Minimax M2.5

benchmarks/benchmark_lib.sh

@@ -470,6 +470,8 @@ def _le_parse_generations(outputs, **kwargs):
                   tmp[idx] = content
           except Exception:
               tmp = [""]
+          if not tmp:
+              tmp = [""]
           res.extend(tmp)
       return res
 

benchmarks/single_node/minimaxm2.5_fp8_h200.sh

@@ -49,7 +49,7 @@ 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 $CONC
+    run_eval --framework lm-eval --port "$PORT" --concurrent-requests 8
     append_lm_eval_summary
 fi
 set +x

perf-changelog.yaml

@@ -790,3 +790,10 @@
     - "Image: vllm/vllm-openai-rocm:v0.15.1"
     - "TP=8, concurrency 4-64 for 1k1k, 1k8k, and 8k1k sequence lengths"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/825
+
+- config-keys:
+  - minimaxm2.5-fp8-h200-vllm
+  description:
+    - GSM8k eval results for minimax h200
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/838
+  evals-only: true