[Hardware][MUSA] Enable Qwen3-Omni ModelOpt FP8 inference (#5671)

Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
Co-authored-by: amy-why-3459 <wuhaiyan17@huawei.com>
This commit is contained in:
authored and GitHub committed 2026-08-11 17:19:09 +08:00
1 parent 4ab1b4ff9c
commit e04210d690
7 files changed
+125 -3

No files matched your search

@@ -152,6 +152,9 @@ class Qwen3OmniMoeForConditionalGeneration(
dtype=torch.long,
)
elif self.model_stage == "talker":
# The outer wrapper exposes talker_mtp for every stage, but only
# the talker stage owns the module that the method invokes.
self.talker_mtp_graph_safe = current_omni_platform.supports_talker_mtp_graph_capture()
multimodal_config.skip_mm_profiling = True
self.has_preprocess = True
self.has_postprocess = True