You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Hi @vince62s,
I'm trying to add a new language in nllb but when I try to merged lora weight to the base model I have this error :
AssertionError : An error in model's partition and checkpoint's slice was detected
When I compare of param.data.size() and ckpt_t[col_slice_start:col_slice_end, row_slice_start:row_slice_end,].size()
I have this result : torch.Size([265108, 2048]) and torch.Size([256206, 2048])
265108 corresponds to the size of the new dictionary with new tokens and 256206 to the size of the nllb dictionary.
I don't understand why is the vocabulary not updated. Yet, during training I have the good value of vocab size (265108)
Could you enlighten me?
Thanks,
Mat
The text was updated successfully, but these errors were encountered:
Hi @vince62s,
I'm trying to add a new language in nllb but when I try to merged lora weight to the base model I have this error :
AssertionError : An error in model's partition and checkpoint's slice was detected
When I compare of param.data.size() and ckpt_t[col_slice_start:col_slice_end, row_slice_start:row_slice_end,].size()
I have this result : torch.Size([265108, 2048]) and torch.Size([256206, 2048])
265108 corresponds to the size of the new dictionary with new tokens and 256206 to the size of the nllb dictionary.
I don't understand why is the vocabulary not updated. Yet, during training I have the good value of vocab size (265108)
Could you enlighten me?
Thanks,
Mat
The text was updated successfully, but these errors were encountered: