Fix DSA compression tail capacity for PD decode request slots (#38417)

2026-09-08 01:50:07 -07:00
parent 5ae4ccb0fd
commit ccfa120dae
1 changed files with 1 additions and 1 deletions
@@ -1246,7 +1246,7 @@ class KVCacheConfigurator:
        elif self.use_mla_backend and is_dsa_model and not self.mambaish_config:
            token_to_kv_pool = self._build_dsa_kv_pool(
                max_total_num_tokens=sizes.max_total_num_tokens,
-                max_running_requests=sizes.max_running_requests,
+                max_running_requests=req_to_token_pool.req_to_token.shape[0],
            )
        elif self.use_mla_backend and not self.mambaish_config:
            assert not is_dsa_model