SpeechGPT
SpeechGPT copied to clipboard
SpeechGPT: Empowering Large Language Models with Intrinsic Cross-Modal Conversational Abilities.
The conflict is caused by: fairseq 0.12.2 depends on omegaconf
Hello, I see number of trainable parameters in stage 2 is smaller than model parameters. Is this using PEFT? ``` Number of trainable parameters = 105,416,512 ```
Hello, I encounter this issue when running `cm_sft.py` ``` 0: File "SpeechGPT/speechgpt/src/train/cm_sft_modified.py", line 460, in 0: train() 0: File "SpeechGPT/speechgpt/src/train/cm_sft_modified.py", line 428, in train 0: train_result = trainer.train(resume_from_checkpoint=checkpoint) 0: File...
Hello, I get the following error when loading dataset for stage 2. Is there a solution? ``` 6: Traceback (most recent call last): 6: File "SpeechGPT/speechgpt/src/train/cm_sft.py", line 346, in 6:...