Uploaded using `kernel-builder`.
Browse files
README.md
CHANGED
|
@@ -7,12 +7,30 @@ This is the repository card of kernels-community/flash-attn2 that has been pushe
|
|
| 7 |
|
| 8 |
## How to use
|
| 9 |
|
| 10 |
-
|
|
|
|
|
|
|
| 11 |
|
| 12 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
|
| 14 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 15 |
|
| 16 |
## Benchmarks
|
| 17 |
|
| 18 |
-
Benchmarking script is available for this kernel. Run `kernels benchmark kernels-community/flash-attn2 --version
|
|
|
|
| 7 |
|
| 8 |
## How to use
|
| 9 |
|
| 10 |
+
```python
|
| 11 |
+
# make sure `kernels` is installed: `pip install -U kernels`
|
| 12 |
+
from kernels import get_kernel
|
| 13 |
|
| 14 |
+
kernel_module = get_kernel("kernels-community/flash-attn2", version=2)
|
| 15 |
+
flash_attn_func = kernel_module.flash_attn_func
|
| 16 |
+
|
| 17 |
+
flash_attn_func(...)
|
| 18 |
+
```
|
| 19 |
|
| 20 |
+
## Available functions
|
| 21 |
+
- `flash_attn_func`
|
| 22 |
+
- `flash_attn_kvpacked_func`
|
| 23 |
+
- `flash_attn_qkvpacked_func`
|
| 24 |
+
- `flash_attn_varlen_func`
|
| 25 |
+
- `flash_attn_varlen_kvpacked_func`
|
| 26 |
+
- `flash_attn_varlen_qkvpacked_func`
|
| 27 |
+
- `flash_attn_with_kvcache`
|
| 28 |
+
- `fwd`
|
| 29 |
+
- `varlen_fwd`
|
| 30 |
+
- `bwd`
|
| 31 |
+
- `varlen_bwd`
|
| 32 |
+
- `fwd_kvcache`
|
| 33 |
|
| 34 |
## Benchmarks
|
| 35 |
|
| 36 |
+
Benchmarking script is available for this kernel. Run `kernels benchmark kernels-community/flash-attn2 --version 2`.
|