Update app.py
Browse files
app.py
CHANGED
@@ -11,6 +11,14 @@ from transformers import (
|
|
11 |
TextIteratorStreamer,
|
12 |
LlamaTokenizer,
|
13 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
|
15 |
MAX_MAX_NEW_TOKENS = 1024
|
16 |
DEFAULT_MAX_NEW_TOKENS = 50
|
|
|
11 |
TextIteratorStreamer,
|
12 |
LlamaTokenizer,
|
13 |
)
|
14 |
+
import subprocess
|
15 |
+
|
16 |
+
subprocess.run(
|
17 |
+
"pip install flash-attn --no-build-isolation",
|
18 |
+
env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
|
19 |
+
shell=True,
|
20 |
+
)
|
21 |
+
|
22 |
|
23 |
MAX_MAX_NEW_TOKENS = 1024
|
24 |
DEFAULT_MAX_NEW_TOKENS = 50
|