[Bugfix] Fix Qwen3-VL and Cosmos3-Edge text architectures for CPU and pipeline parallelism (#53699)
Signed-off-by: xiaoming <1259730330@qq.com>
This commit is contained in:
@@ -575,7 +575,9 @@ class Cosmos3EdgeForConditionalGeneration(
|
|||||||
|
|
||||||
with self._mark_language_model(vllm_config):
|
with self._mark_language_model(vllm_config):
|
||||||
self.language_model = Cosmos3EdgeForCausalLM(
|
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"),
|
prefix=maybe_prefix(prefix, "language_model"),
|
||||||
)
|
)
|
||||||
|
|
||||||
|
|||||||
Reference in New Issue
Block a user