Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -6,6 +6,9 @@ import gradio as gr
|
|
| 6 |
from threading import Thread
|
| 7 |
from transformers import TextIteratorStreamer, AutoTokenizer, AutoModelForCausalLM
|
| 8 |
|
|
|
|
|
|
|
|
|
|
| 9 |
parser = argparse.ArgumentParser()
|
| 10 |
|
| 11 |
if torch.cuda.is_available():
|
|
@@ -17,7 +20,8 @@ model_id = "vikhyatk/moondream2"
|
|
| 17 |
revision = "2024-04-02"
|
| 18 |
tokenizer = AutoTokenizer.from_pretrained(model_id, revision=revision)
|
| 19 |
moondream = AutoModelForCausalLM.from_pretrained(
|
| 20 |
-
model_id, trust_remote_code=True, revision=revision
|
|
|
|
| 21 |
).to(device=device, dtype=dtype)
|
| 22 |
moondream.eval()
|
| 23 |
|
|
|
|
| 6 |
from threading import Thread
|
| 7 |
from transformers import TextIteratorStreamer, AutoTokenizer, AutoModelForCausalLM
|
| 8 |
|
| 9 |
+
import subprocess
|
| 10 |
+
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
|
| 11 |
+
|
| 12 |
parser = argparse.ArgumentParser()
|
| 13 |
|
| 14 |
if torch.cuda.is_available():
|
|
|
|
| 20 |
revision = "2024-04-02"
|
| 21 |
tokenizer = AutoTokenizer.from_pretrained(model_id, revision=revision)
|
| 22 |
moondream = AutoModelForCausalLM.from_pretrained(
|
| 23 |
+
model_id, trust_remote_code=True, revision=revision,
|
| 24 |
+
attn_implementation="flash_attention_2"
|
| 25 |
).to(device=device, dtype=dtype)
|
| 26 |
moondream.eval()
|
| 27 |
|