x

config.json

@@ -69,11 +69,11 @@
         "local_path": "Qwen3.6-27B-FP8",
         "dtype": "auto",
         "quantization": "fp8",
-        "ctx": "131072",
+        "ctx": "65536",
         "max_tokens": "32768",
         "max_num_batched_tokens": 16384,
         "trust_remote": true,
-        "enforce_eager": false,
+        "enforce_eager": true,
         "valid_tp": [
           1,
           2

docker-compose.yml

@@ -19,10 +19,12 @@ services:
       HF_HUB_OFFLINE: "1"
       TRANSFORMERS_OFFLINE: "1"
       HF_DATASETS_OFFLINE: "1"
-      VLLM_ROCM_USE_AITER: "1"  # 打开总开关
+      VLLM_ROCM_USE_AITER: "0"  # 打开总开关
       # 明确禁用不稳定的MoE后端，让其他AITER优化（如MHA）生效
       VLLM_ROCM_USE_AITER_MOE: "0"
       VLLM_ROCM_MOE_BACKEND: "TRITON"  # 强制MoE使用Triton
+      TORCHINDUCTOR_FX_GRAPH_CACHE: "0" # 禁用 torch.compile 缓存
+      VLLM_DISABLE_COMPILE_CACHE: "1"
     devices:
       - /dev/kfd
       - /dev/dri