diff --git a/config.json b/config.json index bc99569..b2c92b1 100644 --- a/config.json +++ b/config.json @@ -73,11 +73,11 @@ "local_path": "Qwen3.5-35B-A3B-GPTQ-Int4", "dtype": "bfloat16", "quantization": "gptq", - "ctx": "65536", + "ctx": "131072", "trust_remote": true, "valid_tp": [1, 2], "max_num_seqs": "4", - "max_tokens": "4096", + "max_tokens": "8192", "gpu_util": "0.90", "enforce_eager": false, "tool_call_parser": "qwen3_coder",