Commit History

Got output to match Phi2 exactly
3649bbb

BucketOfFish commited on

Fixed inference script bug and made deterministic
4f25dda

BucketOfFish commited on

Passing KV cache through iterations
c07c430

BucketOfFish commited on

Corrected rotary embedding
df388cc

BucketOfFish commited on

Got model running, but results are incorrect
0f3418e

BucketOfFish commited on

Fixed weight loading from original Phi2 model
10aca20

BucketOfFish commited on

Renaming state dict keys from Phi2
78f6f3b

BucketOfFish commited on

Updated imports in script
16cc769

BucketOfFish commited on

Just uploading entire rewritten codebase at once
a420fe7

BucketOfFish commited on

Rotary embedding correction
2380737

BucketOfFish commited on

Simplified rotary embedding
3c52426

BucketOfFish commited on

Removed all flash_attn usage
5e8c4af

BucketOfFish commited on

Revert "Removed non-flash configs"
8a5eabe

BucketOfFish commited on

Revert "Removed non-flash classes"
7c0c66d

BucketOfFish commited on

Removed non-flash configs
23b970e

BucketOfFish commited on

Removed non-flash classes
6be42f2

BucketOfFish commited on

Streaming inference script
dc6124b

BucketOfFish commited on