diff --git a/config.json b/config.json index c3ec432..98dae7f 100644 --- a/config.json +++ b/config.json @@ -78,7 +78,6 @@ "Qwen3.5-35B-A3B-FP8": { "local_path": "Qwen3.5-35B-A3B-FP8", "dtype": "float16", - "quantization": "awq", "ctx": "131072", "enforce_eager": false, "trust_remote": true, @@ -86,6 +85,9 @@ "max_num_seqs": "8", "max_tokens": "16384", "gpu_util": "0.94", + "kv_cache_dtype": "fp8", + "enable_prefix_caching": true, + "max_num_batched_tokens": 16384, "tool_call_parser": "qwen3_coder", "reasoning_parser": "qwen3", "enable_auto_tool_choice": true,