[Bugfix] Fix Qwen3-VL and Cosmos3-Edge text architectures for CPU and pipeline parallelism (#53699)
Signed-off-by: xiaoming <1259730330@qq.com>
labAxiaoming committed
Sep 11, 2026 at 20:58 UTC
a5b714eee4ef09c286368c53df28443b7ac4cf4a
1 file changed
+3
-1
vllm/model_executor/models/cosmos3_edge.py
+3
-1
@@ -575,7 +575,9 @@ class Cosmos3EdgeForConditionalGeneration(
575
576
with self._mark_language_model(vllm_config):
577
self.language_model = Cosmos3EdgeForCausalLM(
578
- vllm_config=vllm_config,
578
+ vllm_config=vllm_config.with_hf_config(
579
+ config.text_config, architectures=["NemotronHForCausalLM"]
580
+ ),
581
prefix=maybe_prefix(prefix, "language_model"),
582
)
583