ABQ-LLM icon indicating copy to clipboard operation
ABQ-LLM copied to clipboard

Published 20 hours ago • verified publisher icon bytedance
→

Metadata

An acceleration library that supports arbitrary bit-width combinatorial quantization operations

Results 13 ABQ-LLM issues
Sort by recently updated
recently updated
newest added

Do you have any benchmark compared with MARLIN?

1 comment

Juelianqvq avatar Juelianqvq

What is the difference between pack4 and pack, and why is pack4 used in the file test_w4a4_int_wmma.cu?

2 comment

luliyucoordinate avatar luliyucoordinate

The methods in this article are really cool !!!

1 comment

luliyucoordinate avatar luliyucoordinate

← Metadata

82
Stars
4
Forks
Watchers

Owner

verified publisher icon bytedance

Metadata

An acceleration library that supports arbitrary bit-width combinatorial quantization operations

Back