v1

app.py

@@ -13,6 +13,11 @@ from meteor.load_meteor import load_meteor
 from transformers import TextIteratorStreamer
 from torchvision.transforms.functional import pil_to_tensor
 
+# package install
+import sys
+import subprocess
+subprocess.check_call([sys.executable, '-m', 'pip', 'install', 'flash-attn', '--no-build-isolation', '-U'])
+
 # loading meteor model
 mmamba = load_mmamba('BK-Lee/Meteor-Mamba').to('cuda')
 meteor, tok_meteor = load_meteor('BK-Lee/Meteor-MLM', bits=4)

requirements.txt

@@ -9,7 +9,6 @@ wandb
 einops
 einops_exts
 sentencepiece
-flash-attn, --no-build-isolation
 timm
 shortuuid
 matplotlib