From a5b714eee4ef09c286368c53df28443b7ac4cf4a Mon Sep 17 00:00:00 2001 From: labAxiaoming <34019940+labAxiaoming@users.noreply.github.com> Date: Fri, 11 Sep 2026 20:58:15 +0800 Subject: [PATCH] [Bugfix] Fix Qwen3-VL and Cosmos3-Edge text architectures for CPU and pipeline parallelism (#53699) Signed-off-by: xiaoming <1259730330@qq.com> --- vllm/model_executor/models/cosmos3_edge.py | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/vllm/model_executor/models/cosmos3_edge.py b/vllm/model_executor/models/cosmos3_edge.py index 8c5714248b..38e55405b8 100644 --- a/vllm/model_executor/models/cosmos3_edge.py +++ b/vllm/model_executor/models/cosmos3_edge.py @@ -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"), )