alex buz
commited on
Commit
•
e1cddb8
1
Parent(s):
767736b
test
Browse files- _app.py +60 -0
- _requirements.txt +6 -0
- app.py +12 -56
- requirements.txt +1 -5
_app.py
ADDED
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import AutoProcessor, AutoModelForCausalLM
|
2 |
+
from PIL import Image
|
3 |
+
import gradio as gr
|
4 |
+
|
5 |
+
model_id = 'microsoft/Florence-2-large'
|
6 |
+
model = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True,
|
7 |
+
torch_dtype="auto",
|
8 |
+
#device_map="auto",
|
9 |
+
cache_dir="./cache",
|
10 |
+
#attn_implementation="flash_attention_2",
|
11 |
+
).eval()
|
12 |
+
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True,
|
13 |
+
torch_dtype="auto",
|
14 |
+
#device_map="auto",
|
15 |
+
cache_dir="./cache",
|
16 |
+
#attn_implementation="flash_attention_2",
|
17 |
+
)
|
18 |
+
|
19 |
+
def run_example(task_prompt, image, text_input=None):
|
20 |
+
if text_input is None:
|
21 |
+
prompt = task_prompt
|
22 |
+
else:
|
23 |
+
prompt = task_prompt + text_input
|
24 |
+
|
25 |
+
inputs = processor(text=prompt, images=image, return_tensors="pt")
|
26 |
+
generated_ids = model.generate(
|
27 |
+
input_ids=inputs["input_ids"],
|
28 |
+
pixel_values=inputs["pixel_values"],
|
29 |
+
max_new_tokens=1024,
|
30 |
+
early_stopping=False,
|
31 |
+
do_sample=False,
|
32 |
+
num_beams=3,
|
33 |
+
)
|
34 |
+
generated_text = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
|
35 |
+
parsed_answer = processor.post_process_generation(
|
36 |
+
generated_text,
|
37 |
+
task=task_prompt,
|
38 |
+
image_size=(image.width, image.height),
|
39 |
+
#stream=True
|
40 |
+
)
|
41 |
+
|
42 |
+
return parsed_answer
|
43 |
+
|
44 |
+
def inference(image, task_prompt, text_input):
|
45 |
+
return run_example(task_prompt, image, text_input)
|
46 |
+
|
47 |
+
interface = gr.Interface(
|
48 |
+
fn=inference,
|
49 |
+
inputs=[
|
50 |
+
gr.Image(type="pil"),
|
51 |
+
gr.Textbox(label="Task Prompt", placeholder="Enter task prompt here"),
|
52 |
+
gr.Textbox(label="Additional Text Input", placeholder="Enter additional text input here (optional)", optional=True)
|
53 |
+
],
|
54 |
+
outputs="text",
|
55 |
+
title="Hugging Face Model Inference",
|
56 |
+
description="Generate text based on an image and a prompt using a Hugging Face model"
|
57 |
+
)
|
58 |
+
|
59 |
+
if __name__ == "__main__":
|
60 |
+
interface.launch()
|
_requirements.txt
ADDED
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
transformers
|
2 |
+
pillow
|
3 |
+
gradio
|
4 |
+
#flash_attn
|
5 |
+
#timm
|
6 |
+
#einops
|
app.py
CHANGED
@@ -1,60 +1,16 @@
|
|
1 |
-
from transformers import AutoProcessor, AutoModelForCausalLM
|
2 |
-
from PIL import Image
|
3 |
-
import gradio as gr
|
4 |
-
|
5 |
-
model_id = 'microsoft/Florence-2-large'
|
6 |
-
model = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True,
|
7 |
-
torch_dtype="auto",
|
8 |
-
#device_map="auto",
|
9 |
-
cache_dir="./cache",
|
10 |
-
#attn_implementation="flash_attention_2",
|
11 |
-
).eval()
|
12 |
-
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True,
|
13 |
-
torch_dtype="auto",
|
14 |
-
#device_map="auto",
|
15 |
-
cache_dir="./cache",
|
16 |
-
#attn_implementation="flash_attention_2",
|
17 |
-
)
|
18 |
-
|
19 |
-
def run_example(task_prompt, image, text_input=None):
|
20 |
-
if text_input is None:
|
21 |
-
prompt = task_prompt
|
22 |
-
else:
|
23 |
-
prompt = task_prompt + text_input
|
24 |
|
25 |
-
|
26 |
-
|
27 |
-
input_ids=inputs["input_ids"],
|
28 |
-
pixel_values=inputs["pixel_values"],
|
29 |
-
max_new_tokens=1024,
|
30 |
-
early_stopping=False,
|
31 |
-
do_sample=False,
|
32 |
-
num_beams=3,
|
33 |
-
)
|
34 |
-
generated_text = processor.batch_decode(generated_ids, skip_special_tokens=False)[0]
|
35 |
-
parsed_answer = processor.post_process_generation(
|
36 |
-
generated_text,
|
37 |
-
task=task_prompt,
|
38 |
-
image_size=(image.width, image.height),
|
39 |
-
#stream=True
|
40 |
-
)
|
41 |
-
|
42 |
-
return parsed_answer
|
43 |
|
44 |
-
|
45 |
-
return run_example(task_prompt, image, text_input)
|
46 |
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
gr.Image(type="pil"),
|
51 |
-
gr.Textbox(label="Task Prompt", placeholder="Enter task prompt here"),
|
52 |
-
gr.Textbox(label="Additional Text Input", placeholder="Enter additional text input here (optional)", optional=True)
|
53 |
-
],
|
54 |
-
outputs="text",
|
55 |
-
title="Hugging Face Model Inference",
|
56 |
-
description="Generate text based on an image and a prompt using a Hugging Face model"
|
57 |
-
)
|
58 |
|
59 |
-
|
60 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
|
2 |
+
import gradio as gr
|
3 |
+
from transformers import pipeline
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
4 |
|
5 |
+
pipeline = pipeline(task="image-classification", model="julien-c/hotdog-not-hotdog")
|
|
|
6 |
|
7 |
+
def predict(image):
|
8 |
+
predictions = pipeline(image)
|
9 |
+
return {p["label"]: p["score"] for p in predictions}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
|
11 |
+
gr.Interface(
|
12 |
+
predict,
|
13 |
+
inputs=gr.Image(label="Upload hot dog candidate", type="filepath"),
|
14 |
+
outputs=gr.Label(num_top_classes=2),
|
15 |
+
title="Hot Dog? Or Not?",
|
16 |
+
).launch()
|
requirements.txt
CHANGED
@@ -1,6 +1,2 @@
|
|
1 |
transformers
|
2 |
-
|
3 |
-
gradio
|
4 |
-
#flash_attn
|
5 |
-
#timm
|
6 |
-
#einops
|
|
|
1 |
transformers
|
2 |
+
torch
|
|
|
|
|
|
|
|