Trying out gpt-v1 and facing an error around tensor size mismatch

[sshenoy7]

Disclaimer: I am very new to this domain and trying to grasp concepts and gain understanding. So pardon if my questions are noob-like :)

I am trying out the gpt-v1.py (using the wizard_of_oz.txt not the vocab one) following the YouTube video as-is and facing following error:

"RuntimeError: The size of tensor a (32) must match the size of tensor b (33) at non-singleton dimension 2" for line wei.masked_fill(self.tril[:T, :T] == 0, float('-inf'))

Any help or guidance would be appreciated.

[sshenoy7]

I found the error.

I missed to add the following line that crops the token generation process to block size when max new tokens are bigger than the block_size:
index_cond = index[:, -block_size:]