diff --git a/config.json b/config.json index c2e92c7..6cf7254 100644 --- a/config.json +++ b/config.json @@ -52,11 +52,11 @@ "Qwen3.5-27B-FP8": { "local_path": "Qwen3.5-27B-FP8", "dtype": "bfloat16", - "ctx": "65536", + "ctx": "32768", "trust_remote": true, "valid_tp": [1, 2], "max_num_seqs": "4", - "max_tokens": "8192", + "max_tokens": "16384", "gpu_util": "0.94", "tool_call_parser": "qwen3_xml", "enable_auto_tool_choice": true,