Add phi3 128K model support #7225
ggerganov
merged
22 commits into
ggml-org:master
from
liuwei-git:master
ggerganov
commented
on 2024-05-12
slaren
commented
on 2024-05-16
add phi3 128k support in convert-hf-to-gguf
8fa413d8
add phi3 128k support in cuda
56d9fa72
address build warnings on llama.cpp
cc19780a
adjust index value in cuda long rope freq factors
9f871298
add long rope support in ggml cpu backend
c5569311
make freq factors only depend on ctx size
6333ed1a
remove unused rope scaling type 'su' frin gguf converter
5683db3b
fix flint warnings on convert-hf-to-gguf.py
b1f491a2
set to the short freq factor when context size is small than trained …
d05ae12e
add one line of comments
8a9c897f
metal : support rope freq_factors
2d473a4a
ggml : update ggml_rope_ext API to support freq. factors
471d8170
backends : add dev messages to support rope freq. factors
352c3859
minor : style
f4cb482c
tests : update to use new rope API
e7c7d8ca
backends : fix pragma semicolons
4f787ead
minor : cleanup
d93b5cad
llama : move rope factors from KV header to tensors
600896b8
llama : remove tmp assert
23b72b87
ggerganov
approved these changes
on 2024-05-21
cuda : fix compile warning
e9acbce6
convert : read/write n_head_kv
92711138
llama : fix uninitialized tensors
7528c705
Login to write a write a comment.
Login via GitHub