Kernels
kernels-bot commited on
Commit
88fe4d9
·
verified ·
1 Parent(s): b945a2a

Uploaded using `kernel-builder`.

Browse files
Files changed (1) hide show
  1. README.md +24 -16
README.md CHANGED
@@ -1,24 +1,32 @@
1
  ---
2
- tags:
3
- - kernels
4
- - flash-mla
5
- - deepseek
6
- - kernel-builder
7
  ---
8
 
9
- ![Status](https://hubwebhook.dholtz.com/shield?repo=kernels-community/flash-mla)
10
 
11
- ## flash-mla
12
 
13
- This repo builds Deepseeks [FlashMLA](https://github.com/deepseek-ai/FlashMLA) kernel via the HF [kernel-builder](https://github.com/huggingface/kernel-builder)
 
 
14
 
15
- ### Dev
16
- ```bash
17
- nix develop -L
18
- pytest -vv tests/
19
- ```
20
 
21
- ### Build
22
- ```bash
23
- nix build .#bundle -L
24
  ```
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ library_name: kernels
3
+ license: mit
 
 
 
4
  ---
5
 
6
+ This is the repository card of kernels-community/flash-mla that has been pushed on the Hub. It was built to be used with the [`kernels` library](https://github.com/huggingface/kernels). This card was automatically generated.
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-mla")
15
+ __version__ = kernel_module.__version__
 
 
 
16
 
17
+ __version__(...)
 
 
18
  ```
19
+
20
+ ## Available functions
21
+ - `__version__`
22
+ - `FlashMLASchedMeta`
23
+ - `get_mla_metadata`
24
+ - `flash_mla_with_kvcache`
25
+ - `flash_attn_varlen_func`
26
+ - `flash_attn_varlen_qkvpacked_func`
27
+ - `flash_attn_varlen_kvpacked_func`
28
+ - `flash_mla_sparse_fwd`
29
+
30
+ ## Benchmarks
31
+
32
+ Benchmarking script is available for this kernel. Run `kernels benchmark kernels-community/flash-mla`.