Support sliding window in triton backend (#6509)

2025-05-30 04:11:53 -04:00
parent d279d4990c
commit 22630ca242
6 changed files with 350 additions and 13 deletions
--- a/python/sglang/srt/model_executor/model_runner.py
+++ b/python/sglang/srt/model_executor/model_runner.py
@@ -1025,10 +1025,6 @@ class ModelRunner:

            return AiterAttnBackend(self)
        elif self.server_args.attention_backend == "triton":
-            assert self.sliding_window_size is None, (
-                "Window attention is not supported in the triton attention backend. "
-                "Please use `--attention-backend flashinfer`."
-            )
            assert not self.model_config.is_encoder_decoder, (
                "Cross attention is not supported in the triton attention backend. "
                "Please use `--attention-backend flashinfer`."