[Bugfix][Models] Accept Qwen3_5MoeTextConfig in Qwen3_5MoeProcessingInfo for transformers 5.x compatibility (#50704)

Signed-off-by: loulanyue <loulanyue@users.noreply.github.com>
Co-authored-by: loulanyue <loulanyue@users.noreply.github.com>
This commit is contained in:
Rafli Surya Wijaya
2026-08-02 05:50:40 +00:00
committed by GitHub
co-authored by loulanyue
parent e2fa28594f
commit 0601850791
+5 -1
View File
@@ -108,7 +108,11 @@ class Qwen3_5ProcessingInfo(Qwen3VLProcessingInfo):
class Qwen3_5MoeProcessingInfo(Qwen3VLProcessingInfo):
def get_hf_config(self):
return self.ctx.get_hf_config(Qwen3_5MoeConfig)
# transformers 5.x renames the top-level Qwen3.5-MoE config class to
# Qwen3_5MoeTextConfig for text-only models, while transformers ≤4.x
# returns Qwen3_5MoeConfig (the multimodal wrapper). Accept both so
# that vLLM works regardless of which transformers version is installed.
return self.ctx.get_hf_config((Qwen3_5MoeConfig, Qwen3_5MoeTextConfig))
class Qwen3_5DecoderLayer(Qwen3NextDecoderLayer):