Skip to content

Activity

add a norm for lime

lucidrainspushed 1 commit to main • 5c53e58…9b8b489 • 
7 days ago

add dynamic LIMe from Gerasimov et al., making sure it is compatible …

Force push
lucidrainsforce pushed to main • 228be5f…5c53e58 • 
7 days ago

add dynamic LIMe from Gerasimov et al., making sure it is compatible …

lucidrainspushed 1 commit to main • abee1d3…228be5f • 
7 days ago

fix for dynamic pos bias during inference

lucidrainspushed 2 commits to main • 3a42d6a…abee1d3 • 
12 days ago

oops

lucidrainspushed 1 commit to main • 6739534…3a42d6a • 
12 days ago

2.0.4

Force push
lucidrainsforce pushed to main • 7e019b2…6739534 • 
14 days ago

2.0.3

lucidrainspushed 2 commits to main • bdd8047…7e019b2 • 
14 days ago

use cautious lion for parity task

lucidrainspushed 1 commit to main • 80d5653…bdd8047 • 
23 days ago

demonstrate hybridization with a gru that only acts every 4 tokens ca…

lucidrainspushed 1 commit to main • dbc0d95…80d5653 • 
23 days ago

move examples to root

Force push
lucidrainsforce pushed to main • f655386…dbc0d95 • 
24 days ago

move examples to root

lucidrainspushed 1 commit to main • 16a743c…f655386 • 
24 days ago

release multi-latent attention

lucidrainspushed 1 commit to main • c6edc65…16a743c • 
25 days ago

the rotateable subhead keys from MLA needs to be cached

lucidrainspushed 1 commit to main • 7ad71d3…c6edc65 • 
25 days ago

complete multi-latent attention

lucidrainspushed 1 commit to main • d3420e2…7ad71d3 • 
25 days ago

prepare for decoupled rope

lucidrainspushed 1 commit to main • bb147f1…d3420e2 • 
25 days ago

first test it out without rotary

lucidrainspushed 1 commit to main • d97b7c7…bb147f1 • 
25 days ago

some driveby cleanup

Force push
lucidrainsforce pushed to main • 1bd725c…d97b7c7 • 
25 days ago

some driveby cleanup

lucidrainspushed 1 commit to main • 28818a9…1bd725c • 
25 days ago

in multi latent attention, cache the lightweight latent kv

lucidrainspushed 1 commit to main • 62237f8…28818a9 • 
25 days ago

remove resiDual, as hyperconnections is the culmination for that line…

Force push
lucidrainsforce pushed to main • 75dd02b…62237f8 • 
25 days ago

remove resiDual, as hyperconnections is the culmination for that line…

lucidrainspushed 1 commit to main • 66ab2c7…75dd02b • 
25 days ago

remove some unpopular features / research, and prepare to incorporate…

lucidrainspushed 1 commit to main • 3fe9821…66ab2c7 • 
26 days ago

allow for queries, keys, values to be derived from different combinat…

Force push
lucidrainsforce pushed to main • f254f38…3fe9821 • 
29 days ago

allow for queries, keys, values to be derived from different combinat…

lucidrainspushed 1 commit to main • 1f7ea12…f254f38 • 
29 days ago

allow each token to decide how much of input to reinject

lucidrainspushed 1 commit to main • b15815e…1f7ea12 • 
on Jan 23

1.44.5

lucidrainspushed 1 commit to main • 4cbf014…b15815e • 
on Jan 23

fix inp_inject not being used when using in_attn_cond (#307)

Pull request merge
lucidrainspushed 1 commit to main • b28d82a…4cbf014 • 
on Jan 23

if the hybrid module is an RNN, allow for folding it across the seque…

lucidrainspushed 1 commit to main • c51ecd3…b28d82a • 
on Jan 5

flexibly handle hybrid module outputs

Force push
lucidrainsforce pushed to main • f17b64a…c51ecd3 • 
on Jan 5

1.44.1

lucidrainspushed 2 commits to main • b81646f…f17b64a • 
on Jan 5