Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -22,11 +22,6 @@ from transformers import (
|
|
22 |
)
|
23 |
from transformers.image_utils import load_image
|
24 |
|
25 |
-
import subprocess
|
26 |
-
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
|
27 |
-
|
28 |
-
from io import BytesIO
|
29 |
-
|
30 |
# Constants for text generation
|
31 |
MAX_MAX_NEW_TOKENS = 2048
|
32 |
DEFAULT_MAX_NEW_TOKENS = 1024
|
@@ -61,16 +56,6 @@ model_o = Qwen2VLForConditionalGeneration.from_pretrained(
|
|
61 |
torch_dtype=torch.float16
|
62 |
).to(device).eval()
|
63 |
|
64 |
-
# Load SmolVLM2-2.2B-Instruct
|
65 |
-
MODEL_ID_W = "HuggingFaceTB/SmolVLM2-2.2B-Instruct"
|
66 |
-
processor_w = AutoProcessor.from_pretrained(MODEL_ID_W, trust_remote_code=True)
|
67 |
-
model_w= AutoModelForImageTextToText.from_pretrained(
|
68 |
-
MODEL_ID_W,
|
69 |
-
trust_remote_code=True,
|
70 |
-
_attn_implementation="flash_attention_2",
|
71 |
-
torch_dtype=torch.float16
|
72 |
-
).to(device).eval()
|
73 |
-
|
74 |
def downsample_video(video_path):
|
75 |
"""
|
76 |
Downsamples the video to evenly spaced frames.
|
@@ -111,9 +96,6 @@ def generate_image(model_name: str, text: str, image: Image.Image,
|
|
111 |
elif model_name == "olmOCR-7B-0225":
|
112 |
processor = processor_o
|
113 |
model = model_o
|
114 |
-
elif model_name == "SmolVLM2":
|
115 |
-
processor = processor_w
|
116 |
-
model = model_w
|
117 |
else:
|
118 |
yield "Invalid model selected."
|
119 |
return
|
@@ -168,9 +150,6 @@ def generate_video(model_name: str, text: str, video_path: str,
|
|
168 |
elif model_name == "olmOCR-7B-0225":
|
169 |
processor = processor_o
|
170 |
model = model_o
|
171 |
-
elif model_name == "SmolVLM2":
|
172 |
-
processor = processor_w
|
173 |
-
model = model_w
|
174 |
else:
|
175 |
yield "Invalid model selected."
|
176 |
return
|
@@ -269,7 +248,7 @@ with gr.Blocks(css=css, theme="bethecloud/storj_theme") as demo:
|
|
269 |
with gr.Column():
|
270 |
output = gr.Textbox(label="Output", interactive=False, lines=2, scale=2)
|
271 |
model_choice = gr.Radio(
|
272 |
-
choices=["VIREX-062225-exp", "DREX-062225-exp", "olmOCR-7B-0225"
|
273 |
label="Select Model",
|
274 |
value="VIREX-062225-exp"
|
275 |
)
|
|
|
22 |
)
|
23 |
from transformers.image_utils import load_image
|
24 |
|
|
|
|
|
|
|
|
|
|
|
25 |
# Constants for text generation
|
26 |
MAX_MAX_NEW_TOKENS = 2048
|
27 |
DEFAULT_MAX_NEW_TOKENS = 1024
|
|
|
56 |
torch_dtype=torch.float16
|
57 |
).to(device).eval()
|
58 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
59 |
def downsample_video(video_path):
|
60 |
"""
|
61 |
Downsamples the video to evenly spaced frames.
|
|
|
96 |
elif model_name == "olmOCR-7B-0225":
|
97 |
processor = processor_o
|
98 |
model = model_o
|
|
|
|
|
|
|
99 |
else:
|
100 |
yield "Invalid model selected."
|
101 |
return
|
|
|
150 |
elif model_name == "olmOCR-7B-0225":
|
151 |
processor = processor_o
|
152 |
model = model_o
|
|
|
|
|
|
|
153 |
else:
|
154 |
yield "Invalid model selected."
|
155 |
return
|
|
|
248 |
with gr.Column():
|
249 |
output = gr.Textbox(label="Output", interactive=False, lines=2, scale=2)
|
250 |
model_choice = gr.Radio(
|
251 |
+
choices=["VIREX-062225-exp", "DREX-062225-exp", "olmOCR-7B-0225"],
|
252 |
label="Select Model",
|
253 |
value="VIREX-062225-exp"
|
254 |
)
|