[WIP] Add mammut #641
gpucce
wants to merge
53 commits into
mlfoundations:main
from
gpucce:add_mammut
initial commits
8d2589f3
typos
4e218498
first run
2c967a91
t arg change
53cf4ed2
Merge remote-tracking branch 'upstream/main' into add_mammut
46a89a47
better generation logic
5827a10e
Merge remote-tracking branch 'upstream/main' into add_mammut
fcee7c23
adjust notation
997b7868
make as with cls at the end
852b2f47
transformer.py: MultimodalTransformer not using init_parameters
88f70baa
Merge branch 'fix_multimodal_transformer' of https://github.com/iejMa…
ad716d75
test CI
dcb18d72
move back ci
491bb5ce
split pooling
dce72e8c
Merge remote-tracking branch 'origin/double_att_pool' into fix_multim…
623226c3
Merge branch 'fix_multimodal_transformer' of https://github.com/iejMa…
ee982c91
Merge branch 'fix_multimodal_transformer' into add_mammut
e191018c
Merge branch 'main' into add_mammut
20dc72bc
fix padding and 0 loss
4b95dd48
Merge branch 'main' into add_mammut
1c386220
revert multiple attn pooler
826e7995
rm white space
b25d6f92
duplicated init_parameters
5a55161f
missing is_decoder
26f39c0f
small improvements
fe6bc7bd
move mage projection
0ea74acd
remove useless kwargs
e4258f11
remove useless kwargs
ddb647ce
refactor mammut into MultimodalTransformer
a3d32fd5
fix args
5a29467c
small improvements
805ad2d5
inherit init
46da15f7
make equal
53c6d392
fix typo
1d46cfa0
output latents and logits
2abdf120
add mammut L/14 config
4c8dcee9
allow text=None in forward
2ca74c10
better decoder
52e7b3e8
better decoder
4f0b8321
Merge branch 'main' into add_mammut
b980b57b
Merge branch 'main' into add_mammut
dc8d8413
integrate transformers generate changes
f13b230e
Merge branch 'main' into merge_main_mammut
29a052fb
stash pop
e83ac00f
update generation
717fe7ac
uniform generation
f64c9952
make equal
f3912dc5
add context legnth to mammut
c4bb1986
add older poolign
ef9071ba
Merge branch 'merge_main_mammut' into add_mammut
cb619dc2
Merge branch 'main' into add_mammut
1e14e4c5
typo and print
b34393ee
uniform everything
cc503674
Reviewers
No reviews
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub