My attempts at a double-dequant kernel for QLoRA weights (see: Unsloth Puzzles). Results compared against bitsandbytes CUDA dequantization. See my blog post for more.
| Performance | Speedup |
|---|---|
 |
 |
-
Notifications
You must be signed in to change notification settings - Fork 0
QLoRA dequantization kernels in Triton.
lweitkamp/qlora_dequantize_triton
Folders and files
| Name | Name | Last commit message | Last commit date | |
|---|---|---|---|---|
Repository files navigation
About
QLoRA dequantization kernels in Triton.