Skip to content
GitLab
Explore
Sign in
Tags
Tags give the ability to mark specific points in history as being important
This project is mirrored from
https://github.com/lucidrains/x-transformers
. Pull mirroring updated
Sep 19, 2024
.
1.22.5
a3373cb9
·
just set it up so researchers can pass in a list of amateur models for contrastive decoding
·
Sep 20, 2023
1.22.4
0c7bcf15
·
clearer topk and fix topa thanks to @stas-sl
·
Sep 20, 2023
1.22.3
944e3bfb
·
address
https://github.com/lucidrains/x-transformers/issues/184
·
Sep 19, 2023
1.22.2
42a61423
·
address
https://github.com/lucidrains/x-transformers/issues/184
·
Sep 19, 2023
1.22.1
b38cca68
·
add ability to do contrastive decoding by passing in amateur_model kwarg...
·
Sep 19, 2023
1.22.0
5db8671f
·
add ability to do contrastive decoding by passing in amateur_model kwarg...
·
Sep 19, 2023
1.21.4
1a7625b0
·
1.21.4
·
Sep 19, 2023
1.21.3
b0dd2ce9
·
complete kv cache for transformer-xl inference
·
Sep 16, 2023
1.21.2
b00dbae7
·
default Attention to not return intermediates
·
Sep 15, 2023
1.21.1
3b9aeeeb
·
set things up for maybe spec decoding
·
Sep 15, 2023
1.21.0
87a0f13d
·
cache key / values during inference for sampling from causal decoder, make...
·
Sep 14, 2023
1.20.0a
9930e73e
·
use pypi api token
·
Sep 12, 2023
1.20.0
a475d87c
·
address
https://github.com/lucidrains/x-transformers/issues/182
·
Sep 12, 2023
1.19.1
612632d2
·
give separate weights per head for qk rmsnorm. seems like persimmons uses...
·
Sep 07, 2023
1.19.0
016c6e1e
·
add grouped multi-query attention, from a recent Brain paper
·
Aug 27, 2023
1.18.2
5cac36b7
·
add the auxiliary z-loss for attention logits, which hearing mixed results,...
·
Aug 23, 2023
1.18.1
2dc7c831
·
return layer hiddens for
https://github.com/lucidrains/x-transformers/issues/177
·
Aug 22, 2023
1.18.0
bba19595
·
add ability to use the multiplicative bias used on conjunction with ff geglu...
·
Aug 18, 2023
1.17.3
8c8a9743
·
address
https://github.com/lucidrains/x-transformers/issues/175
·
Aug 16, 2023
1.17.2
b0715f2e
·
fix a small issue around masking and cross attention, thanks to @pradeep-pyro
·
Aug 10, 2023
Prev
1
…
4
5
6
7
8
9
10
11
12
…
21
Next