Hi,
I am newbie to SpeechBrain and hope i can get help!

I trained a transformer model, and found in the results/.../save directory, there are some files saved: model.ckpt. Is this the final models needed for inference? There are some other .ckpt files in the save directory.

My trained models are not big, performance is good. I used the huggingface lm.ckpt, which is huge.
I am trying to get a small set of model, including a small acoustic model and LM (LM with <10M parameters).
How can i train a small LM with Librispeech?
If i use a small LM, will the accuracy drop greatly? What WER i can expect for Librispeech test-clean?

Can anyone give me some guidance on this? Any documents / tutorials i can read?

Another question: Can i convert the model.ckpt file and the lm.ckpt files to TFLite format?
If yes, is there a tool i can use to do this?

Thanks,
Willy