transformers

7623aa3e - Fix `Qwen2AudioForConditionalGeneration.forward()` and `test_flash_attn_kernels_inference_equivalence` (#39503)

Fix `Qwen2AudioForConditionalGeneration.forward()` and `test_flash_attn_kernels_inference_equivalence` (#39503) * Add missing cache_position argument. * Pass cache_position to language model. * Overwrite prepare_inputs_for_generation. * Set model to half precision for Flash Attention test. * Cast model to bfloat16.