Update MiniMax MXFP4 benchmark to M2.5 with vLLM v0.17.1

- Model: amd/MiniMax-M2.1-MXFP4 → amd/MiniMax-M2.5-MXFP4
- Image: vllm/vllm-openai-rocm v0.16.0 → v0.17.1
- Rename config key and script from m2.1 to m2.5
- Update perf-changelog entry

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: functionstackx

.github/configs/amd-master.yaml

@@ -384,10 +384,10 @@ minimaxm2.5-fp8-mi355x-vllm:
     - { tp: 2, conc-start: 4, conc-end: 64 }
     - { tp: 4, conc-start: 4, conc-end: 64 }
 
-minimaxm2.1-fp4-mi355x-vllm:
-  image: vllm/vllm-openai-rocm:v0.16.0
-  model: amd/MiniMax-M2.1-MXFP4
-  model-prefix: minimaxm2.1
+minimaxm2.5-fp4-mi355x-vllm:
+  image: vllm/vllm-openai-rocm:v0.17.1
+  model: amd/MiniMax-M2.5-MXFP4
+  model-prefix: minimaxm2.5
   runner: mi355x
   precision: fp4
   framework: vllm

…ks/single_node/minimaxm2.1_fp4_mi355x.sh …ks/single_node/minimaxm2.5_fp4_mi355x.shbenchmarks/single_node/minimaxm2.1_fp4_mi355x.sh renamed to benchmarks/single_node/minimaxm2.5_fp4_mi355x.sh benchmarks/single_node/minimaxm2.1_fp4_mi355x.sh renamed to benchmarks/single_node/minimaxm2.5_fp4_mi355x.sh

@@ -981,12 +981,11 @@
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/914
 
 - config-keys:
-    - minimaxm2.1-fp4-mi355x-vllm
+    - minimaxm2.5-fp4-mi355x-vllm
   description:
-    - "Add MiniMax M2.1 MXFP4 vLLM benchmark for MI355X"
-    - "Model: amd/MiniMax-M2.1-MXFP4 with --trust-remote-code and --block-size=32"
-    - "Image: vllm/vllm-openai-rocm:v0.16.0"
+    - "Add MiniMax M2.5 MXFP4 vLLM benchmark for MI355X"
+    - "Model: amd/MiniMax-M2.5-MXFP4 with --trust-remote-code and --block-size=32"
+    - "Image: vllm/vllm-openai-rocm:v0.17.1"
     - "Environment: VLLM_ROCM_USE_AITER=1"
-    - "TP=2 only (TP=4 disabled due to vLLM bug https://github.com/vllm-project/vllm/issues/35637)"
-    - "Concurrency 4-64 for 1k1k, 1k8k, and 8k1k sequence lengths"
+    - "TP=2 and TP=4, concurrency 4-64 for 1k1k, 1k8k, and 8k1k sequence lengths"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/827