d-delaurier
commited on
Commit
•
a5ebb82
1
Parent(s):
485e277
Update app.py
Browse files
app.py
CHANGED
@@ -4,23 +4,28 @@ import re
|
|
4 |
import gradio as gr
|
5 |
from threading import Thread
|
6 |
from transformers import TextIteratorStreamer, AutoTokenizer, AutoModelForCausalLM
|
7 |
-
|
8 |
import subprocess
|
|
|
9 |
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
|
10 |
|
11 |
model_id = "vikhyatk/moondream2"
|
12 |
revision = "2024-05-20"
|
13 |
tokenizer = AutoTokenizer.from_pretrained(model_id, revision=revision)
|
14 |
moondream = AutoModelForCausalLM.from_pretrained(
|
15 |
-
model_id,
|
16 |
-
|
|
|
|
|
|
|
17 |
attn_implementation="flash_attention_2"
|
18 |
)
|
19 |
moondream.eval()
|
20 |
|
21 |
-
|
22 |
@spaces.GPU(duration=10)
|
23 |
def answer_question(img, prompt):
|
|
|
|
|
|
|
24 |
image_embeds = moondream.encode_image(img)
|
25 |
streamer = TextIteratorStreamer(tokenizer, skip_special_tokens=True)
|
26 |
thread = Thread(
|
@@ -33,28 +38,26 @@ def answer_question(img, prompt):
|
|
33 |
},
|
34 |
)
|
35 |
thread.start()
|
36 |
-
|
37 |
buffer = ""
|
38 |
for new_text in streamer:
|
39 |
buffer += new_text
|
40 |
yield buffer.strip()
|
41 |
|
42 |
-
|
43 |
with gr.Blocks() as demo:
|
44 |
gr.Markdown(
|
45 |
"""
|
46 |
# myAI - AMI Vision Module
|
47 |
-
A lightweight Computer Vision model by @vikhyat
|
48 |
-
- 🌔 [moondream2](https://github.com/vikhyat/moondream)
|
49 |
"""
|
50 |
)
|
51 |
with gr.Row():
|
52 |
-
prompt = gr.Textbox(label="Input", value="Identify people in this
|
53 |
submit = gr.Button("Submit")
|
54 |
with gr.Row():
|
55 |
img = gr.Image(type="pil", label="Upload an Image")
|
56 |
output = gr.TextArea(label="Response")
|
|
|
57 |
submit.click(answer_question, [img, prompt], output)
|
58 |
prompt.submit(answer_question, [img, prompt], output)
|
59 |
|
60 |
-
demo.queue().launch()
|
|
|
4 |
import gradio as gr
|
5 |
from threading import Thread
|
6 |
from transformers import TextIteratorStreamer, AutoTokenizer, AutoModelForCausalLM
|
|
|
7 |
import subprocess
|
8 |
+
|
9 |
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
|
10 |
|
11 |
model_id = "vikhyatk/moondream2"
|
12 |
revision = "2024-05-20"
|
13 |
tokenizer = AutoTokenizer.from_pretrained(model_id, revision=revision)
|
14 |
moondream = AutoModelForCausalLM.from_pretrained(
|
15 |
+
model_id,
|
16 |
+
trust_remote_code=True,
|
17 |
+
revision=revision,
|
18 |
+
torch_dtype=torch.bfloat16,
|
19 |
+
device_map={"": "cuda"},
|
20 |
attn_implementation="flash_attention_2"
|
21 |
)
|
22 |
moondream.eval()
|
23 |
|
|
|
24 |
@spaces.GPU(duration=10)
|
25 |
def answer_question(img, prompt):
|
26 |
+
if img is None:
|
27 |
+
raise gr.Error("Please upload an image.")
|
28 |
+
|
29 |
image_embeds = moondream.encode_image(img)
|
30 |
streamer = TextIteratorStreamer(tokenizer, skip_special_tokens=True)
|
31 |
thread = Thread(
|
|
|
38 |
},
|
39 |
)
|
40 |
thread.start()
|
|
|
41 |
buffer = ""
|
42 |
for new_text in streamer:
|
43 |
buffer += new_text
|
44 |
yield buffer.strip()
|
45 |
|
|
|
46 |
with gr.Blocks() as demo:
|
47 |
gr.Markdown(
|
48 |
"""
|
49 |
# myAI - AMI Vision Module
|
50 |
+
A lightweight Computer Vision model by @vikhyat - 🌔 [moondream2](https://github.com/vikhyat/moondream)
|
|
|
51 |
"""
|
52 |
)
|
53 |
with gr.Row():
|
54 |
+
prompt = gr.Textbox(label="Input", value="Identify people in this image", scale=4)
|
55 |
submit = gr.Button("Submit")
|
56 |
with gr.Row():
|
57 |
img = gr.Image(type="pil", label="Upload an Image")
|
58 |
output = gr.TextArea(label="Response")
|
59 |
+
|
60 |
submit.click(answer_question, [img, prompt], output)
|
61 |
prompt.submit(answer_question, [img, prompt], output)
|
62 |
|
63 |
+
demo.queue().launch()
|