diff --git a/config.json b/config.json index fac72a3..195fcb1 100644 --- a/config.json +++ b/config.json @@ -68,13 +68,13 @@ "local_path": "Qwen3.5-35B-A3B-GPTQ-Int4", "dtype": "float16", "quantization": "gptq", - "ctx": "65536", + "ctx": "32768", "trust_remote": true, "valid_tp": [1, 2], "max_num_seqs": "4", - "max_tokens": "4096", + "max_tokens": "8192", "gpu_util": "0.94", - "enforce_eager": true, + "enforce_eager": false, "tool_call_parser": "qwen3_xml", "enable_auto_tool_choice": true, "served_model_name": "Qwen3.5-35B-A3B-GPTQ-Int4",