nodes/hao-ai-lab-FastVideo
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

[feat] Add native SM103a VSA support (#1812)

Browse Source 
Signed-off-by: lishunyang12 <lishunyang12@163.com>
This commit is contained in:
SYLAR
2026-09-03 15:00:28 -07:00
committed by GitHub
parent 0bd19a976b
commit 7bb76b5ec9
18 changed files with 94 additions and 81 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastvideo-kernel/python/fastvideo_kernel/vsa_utils.py
+2 -1
View File
@@ -14,7 +14,8 @@ import math
import torch
VSA_TILE_SIZE = (4, 4, 4)
# 128 is served by the sm_100a CUDA backend (block_sparse_attn_sm100a); 64 and 256 by
# 128 is served by the sm_100a/sm_103a CUDA backend (legacy API name
# block_sparse_attn_sm100a); 64 and 256 by
# Triton and the CuTe-DSL path. A volume here only needs a backend that accepts it.
_SUPPORTED_VSA_BLOCK_VOLUMES = (64, 128, 256)