[Bugfix] Fix Qwen3-VL and Cosmos3-Edge text architectures for CPU and pipeline parallelism (#53699)

Signed-off-by: xiaoming <1259730330@qq.com>
2026-09-11 05:58:15 -07:00
parent 912dfb3758
commit a5b714eee4
1 changed files with 3 additions and 1 deletions
@@ -575,7 +575,9 @@ class Cosmos3EdgeForConditionalGeneration(

        with self._mark_language_model(vllm_config):
            self.language_model = Cosmos3EdgeForCausalLM(
-                vllm_config=vllm_config,
+                vllm_config=vllm_config.with_hf_config(
+                    config.text_config, architectures=["NemotronHForCausalLM"]
+                ),
                prefix=maybe_prefix(prefix, "language_model"),
            )