diff --git a/config.json b/config.json index 5052673..370df5b 100644 --- a/config.json +++ b/config.json @@ -92,21 +92,14 @@ "trust_remote": true, "valid_tp": [2], "max_num_seqs": "6", - "max_tokens": "4096", + "max_tokens": "8096", "gpu_util": "0.90", "enforce_eager": false, - "env": { - "VLLM_USE_TRITON_AWQ": "1" - }, "tool_call_parser": "qwen3_coder", "reasoning_parser": "qwen3", "enable_auto_tool_choice": true, "served_model_name": "Qwen3-Next-80B-A3B-Instruct-AWQ-4bit", "hf_model_id": "cpatonn/Qwen3-Next-80B-A3B-Instruct-AWQ-4bit", - "temperature": 0.7, - "top_p": 0.8, - "top_k": 20, - "min_p": 0 } } }