transformers

cc7ab9be - Fix position_ids typo in Qwen3_5TextModel forward pass (#44399)

Fix position_ids typo in Qwen3_5TextModel forward pass (#44399) * Fix position_ids typo in Qwen3_5TextModel forward pass In `Qwen3_5TextModel.forward`, after splitting `position_ids` into `text_position_ids` (index 0, for text) and `position_ids` (indices 1:, for temporal/height/width), the decoder layer call incorrectly passed `position_ids` instead of `text_position_ids`. This caused shape mismatches in FlashAttention2 when processing non-padded data. The fix aligns with the corresponding code in `Qwen3VLTextModel` which correctly passes `text_position_ids` to decoder layers. Fixes #44384 Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> * fix: apply same position_ids typo fix to qwen3_5_moe generated file The same text_position_ids bug exists in the generated modeling_qwen3_5_moe.py since Qwen3_5MoeTextModel inherits from Qwen3_5TextModel. After make fix-repo, this will be auto-propagated from the modular fix, but fixing the generated file now for consistency. --------- Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>