facebook

/

mms-1b-all

like 131

Follow

AI at Meta 5.05k

Could you kindly clarify the differences in the pre-trained weights of the wav2vec2 sub-model in mms-1b, mms-1b-all, and mms-1b-fl102? And what causes these differences? Or could you explain how the latter two models are trained based on the first one?

Thanks.