feat: support DeepSeek-R1-W4AFP8 model with ep-moe mode (#7762)

Signed-off-by: yangsijia.614 <yangsijia.614@bytedance.com>
2025-07-08 05:47:21 +08:00
parent 6a6e0bb7fd
commit cb9d91ea8a
10 changed files with 1006 additions and 9 deletions
--- a/python/sglang/srt/server_args.py
+++ b/python/sglang/srt/server_args.py
@@ -708,6 +708,7 @@ class ServerArgs:
                "w8a8_fp8",
                "moe_wna16",
                "qoq",
+                "w4afp8",
            ],
            help="The quantization method.",
        )