[Model] Support Meituan LongCat-Flash && LongCat-Flash-MTP (#9824)

2025-08-31 14:29:21 +08:00
parent fd5ce576a4
commit 5e194b2143
10 changed files with 1940 additions and 11 deletions
--- a/python/sglang/srt/model_executor/model_runner.py
+++ b/python/sglang/srt/model_executor/model_runner.py
@@ -307,7 +307,10 @@ class ModelRunner:
        model_num_layers = (
            self.model_config.num_nextn_predict_layers
            if self.is_draft_worker and model_has_mtp_layers
-            else self.model_config.num_hidden_layers
+            else max(
+                self.model_config.num_hidden_layers,
+                self.model_config.num_attention_layers,
+            )
        )
        self.start_layer = getattr(self.model, "start_layer", 0)
        self.end_layer = getattr(self.model, "end_layer", model_num_layers)