diff --git a/config.json b/config.json index d818156..e0906c8 100644 --- a/config.json +++ b/config.json @@ -79,7 +79,7 @@ 2 ], "max_num_seqs": 32, - "gpu_util": "0.90", + "gpu_util": "0.88", "tool_call_parser": "qwen3_coder", "reasoning_parser": "qwen3", "enable_auto_tool_choice": true, @@ -93,7 +93,7 @@ "method": "dflash", "model": "Qwen3.6-27B-DFlash", "num_speculative_tokens": 8, - "draft_tensor_parallel_size": 2 + "draft_tensor_parallel_size": 1 } } }