Update repo-id references after migration to SNC org
Browse files
README.md
CHANGED
|
@@ -66,7 +66,7 @@ pip install -U kernels
|
|
| 66 |
```python
|
| 67 |
from kernels import get_kernel
|
| 68 |
|
| 69 |
-
flash_attn_4 = get_kernel("
|
| 70 |
```
|
| 71 |
|
| 72 |
`kernels` will download this repository, resolve dependencies, and make the package importable without any manual build step.
|
|
@@ -76,7 +76,7 @@ flash_attn_4 = get_kernel("blake-snc/flash-attn-4-sm120-sncbl")
|
|
| 76 |
If you prefer not to use the `kernels` library, you can install from source:
|
| 77 |
|
| 78 |
```bash
|
| 79 |
-
git clone https://huggingface.co/
|
| 80 |
cd flash-attn-4-sm120-sncbl
|
| 81 |
pip install -e ./torch-ext
|
| 82 |
```
|
|
@@ -89,7 +89,7 @@ pip install -e ./torch-ext
|
|
| 89 |
import torch
|
| 90 |
from kernels import get_kernel
|
| 91 |
|
| 92 |
-
flash_attn_4 = get_kernel("
|
| 93 |
|
| 94 |
B, S, H, D = 1, 1024, 16, 128
|
| 95 |
q = torch.randn(B, S, H, D, device="cuda", dtype=torch.bfloat16)
|
|
@@ -225,4 +225,4 @@ For bundle-specific issues (the dispatch logic, validation gaps, packaging), ope
|
|
| 225 |
|
| 226 |
## See also
|
| 227 |
|
| 228 |
-
- [`CONFLICTS_LOG.md`](https://huggingface.co/
|
|
|
|
| 66 |
```python
|
| 67 |
from kernels import get_kernel
|
| 68 |
|
| 69 |
+
flash_attn_4 = get_kernel("SecondNatureComputing/flash-attn-4-sm120")
|
| 70 |
```
|
| 71 |
|
| 72 |
`kernels` will download this repository, resolve dependencies, and make the package importable without any manual build step.
|
|
|
|
| 76 |
If you prefer not to use the `kernels` library, you can install from source:
|
| 77 |
|
| 78 |
```bash
|
| 79 |
+
git clone https://huggingface.co/SecondNatureComputing/flash-attn-4-sm120
|
| 80 |
cd flash-attn-4-sm120-sncbl
|
| 81 |
pip install -e ./torch-ext
|
| 82 |
```
|
|
|
|
| 89 |
import torch
|
| 90 |
from kernels import get_kernel
|
| 91 |
|
| 92 |
+
flash_attn_4 = get_kernel("SecondNatureComputing/flash-attn-4-sm120")
|
| 93 |
|
| 94 |
B, S, H, D = 1, 1024, 16, 128
|
| 95 |
q = torch.randn(B, S, H, D, device="cuda", dtype=torch.bfloat16)
|
|
|
|
| 225 |
|
| 226 |
## See also
|
| 227 |
|
| 228 |
+
- [`CONFLICTS_LOG.md`](https://huggingface.co/SecondNatureComputing/flash-attn-4-sm120/blob/main/CONFLICTS_LOG.md) — detailed log of every conflict encountered while stacking the six PRs, with resolution and per-PR backport guidance
|