0.2.2

0.2.2

v0.2.1

fix alibi in palm lite

v0.2.0a

release palm lite version, thanks to @conceptofmind

v0.2.0

release palm lite version, thanks to @conceptofmind

0.1.0

fuse the attention and feedforward projections, thanks to @feifeibear…

… for alerting me to this

0.0.12

fix rotary embedding caching

0.0.11

start chipping away at Triton version of PaLM, use causal numerically…

… stable softmax (no need for causal mask) + bias-less layernorm, modified from Phil Tillets layernorm tutorial, cite Triton

0.0.10

0.0.10a

cache causal mask and rotary embeddings within attention module

0.0.9

fix prelayernorm in attention

0.0.8

add enwik8 training