x

config.json

@@ -20,7 +20,7 @@
   "revision": "",
   "models": {
     "default": "Qwen3.5-35B-A3B-GPTQ-Int4",
-    "selected": "Gemma-4-26B-A4B",
+    "selected": "Qwen3.5-35B-A3B-FP8",
     "profiles": {
       "GLM-4.7-Flash-AWQ": {
         "local_path": "GLM-4.7-Flash-AWQ",
@@ -78,12 +78,13 @@
       "Qwen3.5-35B-A3B-FP8": {
         "local_path": "Qwen3.5-35B-A3B-FP8",
         "dtype": "float16",
-        "ctx": "65536",
+        "ctx": "131072",
+        "enforce_eager": false,
         "trust_remote": true,
         "valid_tp": [1, 2],
         "max_num_seqs": "8",
-        "max_tokens": "4096",
-        "gpu_util": "0.92",
+        "max_tokens": "16384",
+        "gpu_util": "0.94",
         "tool_call_parser": "qwen3_coder",
         "enable_auto_tool_choice": true,
         "served_model_name": "Qwen3.5-35B-A3B-FP8",

docker-compose.yml

@@ -24,12 +24,9 @@ services:
       PYTHONUNBUFFERED: "1"
       # 启用 AITER 加速（R9700 属于 gfx1201 架构）
       VLLM_ROCM_USE_AITER: "1"
-      VLLM_USE_AITER_UNIFIED_ATTENTION: "1"
       VLLM_ROCM_USE_AITER_MHA: "0"
       # 增强 All-Reduce 性能（双卡通信优化）
       VLLM_ROCM_QUICK_REDUCE_QUANTIZATION: "INT4"
-      # 使用 vLLM V1 引擎（性能更好）
-      VLLM_USE_V1: "1"
     devices:
       - /dev/kfd
       - /dev/dri