vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

finish bitnet i2 e2e

Browse source
This commit is contained in:
Eddie-Wang 2024-06-08 12:44:13 +00:00
parent 2a01a7ce0d
commit 4e1ab50628
6 changed files with 16 additions and 637 deletions
Download patch file Download diff file Expand all files Collapse all files

2
convert-hf-to-gguf.py
View file

@ -1431,7 +1431,7 @@ class BitnetModel(Model):
if x[i] != 0:
scale = x[i]
break
x = np.divide(x, scale)
x = np.where(x * scale > 0, 1, np.where(x * scale < 0, -1, x))
x = x.astype(np.uint8)
x = np.reshape(x, [x.shape[0] // 4, 4])
keep_bit = {0:192, 1:48, 2:12, 3:3}