[Feat] QWen-1M context support[2/2]: Update block sparse attention backend (#5949)

2025-08-07 14:49:36 +08:00
parent a69b637014
commit b7cd743038
15 changed files with 2121 additions and 4 deletions
--- a/python/sglang/srt/model_executor/cuda_graph_runner.py
+++ b/python/sglang/srt/model_executor/cuda_graph_runner.py
@@ -589,6 +589,7 @@ class CudaGraphRunner:
            req_pool_indices=req_pool_indices,
            seq_lens=seq_lens,
            next_token_logits_buffer=next_token_logits_buffer,
+            orig_seq_lens=seq_lens,
            req_to_token_pool=self.model_runner.req_to_token_pool,
            token_to_kv_pool=self.model_runner.token_to_kv_pool,
            attn_backend=self.model_runner.attn_backend,