diff --git a/config.json b/config.json index a82c64b..af4ee1c 100644 --- a/config.json +++ b/config.json @@ -78,16 +78,17 @@ 1, 2 ], - "max_num_seqs": 256, - "gpu_util": "0.92", - "kv_cache_dtype": "fp8", - "enable_prefix_caching": true, + "max_num_seqs": 32, + "gpu_util": "0.90", "tool_call_parser": "qwen3_coder", "reasoning_parser": "qwen3", "enable_auto_tool_choice": true, "language_model_only": true, "served_model_name": "Qwen3.6-27B-FP8", "hf_model_id": "Qwen/Qwen3.6-27B-FP8", + "env": { + "VLLM_RPC_TIMEOUT": "300" + }, "speculative": { "method": "dflash", "model": "Qwen3.6-27B-DFlash",