Doni/CommonAutoRearsh
5
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #2 from marcinbogdanski/fix/fa3-non-hopper-fallback

Browse Source 
add fallback FA3 kernel for non-Hopper GPUs
This commit is contained in:
Andrej 2026-03-06 21:59:49 -08:00 committed by GitHub
commit bb54287479
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 4 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
train.py
View File

@ -18,7 +18,10 @@ import torch.nn as nn
import torch.nn.functional as F
from kernels import get_kernel
fa3 = get_kernel('varunneal/flash-attention-3').flash_attn_interface
cap = torch.cuda.get_device_capability()
# varunneal's FA3 is Hopper only, use kernels-community on non-Hopper GPUs
repo = "varunneal/flash-attention-3" if cap == (9, 0) else "kernels-community/flash-attn3"
fa3 = get_kernel(repo).flash_attn_interface
from constants import MAX_SEQ_LEN, TIME_BUDGET
from prepare import Tokenizer, make_dataloader, evaluate_bpb