Qwen3-Coder-Next fixes and updated recipe

recipes/qwen3-coder-next-fp8.yaml

@@ -1,30 +1,30 @@
 # Recipe: Qwen3-Coder-Next-FP8
 # Qwen3-Coder-Next model in native FP8 format
-# Currently can only be run in solo mode, cluster mode fails with error - tracking https://github.com/vllm-project/vllm/issues/33857
+
 
 recipe_version: "1"
 name: Qwen3-Coder-Next-FP8
-description: vLLM serving Qwen3-Coder-Next-FP8 on a SINGLE NODE ONLY!
+description: vLLM serving Qwen3-Coder-Next-FP8
 
 # HuggingFace model to download (optional, for --download-model)
 model: Qwen/Qwen3-Coder-Next-FP8
 
-# This model can only run on single node (solo)
-solo_only: true
+#solo_only: true
 
 # Container image to use
 container: vllm-node
 
-# No mods required
-mods: []
+# Mod required to fix slowness and crash in the cluster (tracking https://github.com/vllm-project/vllm/issues/33857)
+mods:
+  - mods/fix-qwen3-coder-next
 
 # Default settings (can be overridden via CLI)
 defaults:
   port: 8000
   host: 0.0.0.0
-  tensor_parallel: 1
+  tensor_parallel: 2
   gpu_memory_utilization: 0.7
-  max_model_len: 131072
+  max_model_len: 262144
 
 # Environment variables
 env: {}
@@ -40,4 +40,7 @@ command: |
     --load-format fastsafetensors \
     --attention-backend flashinfer \
     --enable-prefix-caching \
-    --max-model-len {max_model_len} 
+    --max-model-len {max_model_len} \
+    -tp {tensor_parallel} \
+    --distributed-executor-backend ray
+