thomasw21
force pushed
from
9b4a7415
to
48ed32f9
3 years ago
WIP
cc6dd1d3
Fix shared T5
fbc1d040
WIP
91a20aa6
Shared t5 tests
e55b6031
Woops
b5568fba
Woops
50cf6d9c
Hack my way into fix attn_mask
dcb2d610
DS has poor default
04cdc2e0
Maybe this is better
ffbe3bf5
Maybe this is better
90ee34c6
SEP is only defined for HFTokenizer
ed3dbeac
add --vocab-extra-ids 100
c93b5639
Hopefully this fixes MLM
ef205ddf
Hopefully this fixes MLM
c71b3799
Fix MLM datasets length
9da75889
thomasw21
force pushed
from
5fd74567
to
9da75889
3 years ago
I think GPT2 tokenizer might be missing <sep> token
07f0ce57
I think GPT2 tokenizer might be missing <sep> token
12626f61
Flatten to pass tuples of tensor instead
52c42b25
Lambda function can't unpack tuples
b25803bf
Woops
26d9dce1
More dirty fixes
fc0560ac
Woops
8578ab36
Removing erroneous trailing layers
216a3f5f
Add TODO for loading previous checkpoint
f557523e
WIP: conversion script:
e73aa58e
Assignees
No one assigned
Login to write a write a comment.
Login via GitHub