SeamlessM4TForTextToSpeech.generate not working if generation_config is passed #34811
Description
For SeamlessM4TForTextToSpeech with checkpoint "facebook/hf-seamless-m4t-medium",
model.generate(**model_inputs, tgt_lang="eng", generation_config=model.generation_config)
will fail with index error.
If not passing generation_config argument, it could run without error.
Reproduction
from transformers import AutoProcessor, SeamlessM4TForTextToSpeech
ckpt = "facebook/hf-seamless-m4t-medium"
processor = AutoProcessor.from_pretrained(ckpt)
model = SeamlessM4TForTextToSpeech.from_pretrained(ckpt)
model_inputs = processor(["This is a test"], return_tensors="pt")
# This works
outputs = model.generate(**model_inputs, tgt_lang="eng")
print(outputs)
# Index error
outputs2 = model.generate(**model_inputs, tgt_lang="eng", generation_config=model.generation_config)
print(outputs2)Expected behavior
Should not fail if generation_config is passed.