liangsu9988 commited on
Commit
68bd0b8
·
verified ·
1 Parent(s): ae4d556

Initialize legacy kernels compatibility mirror

Browse files
Files changed (1) hide show
  1. README.md +6 -32
README.md CHANGED
@@ -1,35 +1,9 @@
1
- # masked-mha-runtime
2
 
3
- FlashRT native SM110 masked FP16/BF16 MHA for fixed-shape CUDA Graph runtimes.
4
- It masks padded logits during softmax, supports fused-QKV token strides, and
5
- keeps caller-owned logits/output buffers stable across graph replay.
6
 
7
- ```python
8
- from kernels import get_kernel
9
- import torch
10
 
11
- ops = get_kernel("flashrt/masked-mha-runtime", version=1)
12
- logits = ops.allocate_workspace(q, k)
13
- out = torch.empty_like(q, memory_format=torch.contiguous_format)
14
- ops.attention_mha_fp16_masked(q, k, v, logits=logits, out=out)
15
- ```
16
-
17
- Public functions are `allocate_workspace`, `attention_mha_fp16_masked`,
18
- `attention_mha_bf16_masked`, `forward_static`, `forward`, and
19
- `forward_seqused_static`. The explicit masked names are additive; the legacy
20
- `forward_static` entry dispatches the same kernels.
21
- Inputs use `(sequence, heads, head_dim)`. The production GROOT gate covers
22
- DiT `(41, 32, 48)`, ViT/LLM sequence lengths `277/1024`, padded boundaries
23
- `1025/2048`, FP16 and BF16, fused strides, and bitwise CUDA Graph replay.
24
- `forward_seqused_static` is the FP16 shared-KV path used by static VLA
25
- decoders: Q is `(Sq,H,D)`, K/V are `(Sk_max,D)`, and a CUDA int32 `valid_k`
26
- limits softmax without a standalone `-inf` fill. It supports `Sk_max <= 1024`.
27
-
28
- The BF16 masked entry accepts `qkv_token_stride=` and validates it against the
29
- actual Tensor token stride. This permits direct views into fused QKV GEMM
30
- output while rejecting a mismatched layout before launch. Both masked entries
31
- only read and write the valid `S_kv` columns, so padded workspace bytes never
32
- participate in softmax or CUDA Graph replay.
33
-
34
- See [CARD.md](CARD.md) for the complete contract. Source provenance is FlashRT
35
- commit `24df793f4fa2d50780aea03b644208c6e0cb4162`.
 
1
+ # flashrt/masked-mha-runtime
2
 
3
+ This repository is a compatibility mirror for older `kernels` clients
4
+ that resolve repositories through the default Hugging Face model repo API.
 
5
 
6
+ Canonical Kernel Hub repo: https://huggingface.co/kernels/flashrt/masked-mha-runtime
 
 
7
 
8
+ Do not edit this mirror by hand. It is generated from the Kernel Hub
9
+ `vN` branches and contains the same `build/**` artifacts.