Closed jerome-f closed 8 months ago
hmm, I'm getting a similar error:
AssertionError: if using tf gamma, only sequence length of 1536 allowed for now
@jerome-f – how did you fix this?
nevermind ... I had to add use_tf_gamma = False
to the from_pretrained()
function!
@lucidrains I think this is not getting set properly. As per the doc it says from_pretrained should set the use_tf_gamma. Then the pre trained model should be able to accept seqlength of 196_608 ? or am I missing something ?
get the AssertionError with above