tt

app.py

@@ -1,36 +1,14 @@
 import gradio as gr
 from transformers import AutoProcessor, AutoModelForCausalLM
-import spaces
-
 import requests
 import copy
-
-from PIL import Image, ImageDraw, ImageFont 
+from PIL import Image, ImageDraw
 import io
 import matplotlib.pyplot as plt
 import matplotlib.patches as patches
-
 import random
 import numpy as np
 
-import subprocess
-subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
-
-models = {
-    'microsoft/Florence-2-large-ft': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-large-ft', trust_remote_code=True).to("cpu").eval(),
-    'microsoft/Florence-2-large': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-large', trust_remote_code=True).to("cpu").eval(),
-    'microsoft/Florence-2-base-ft': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-base-ft', trust_remote_code=True).to("cpu").eval(),
-    'microsoft/Florence-2-base': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True).to("cpu").eval(),
-}
-
-processors = {
-    'microsoft/Florence-2-large-ft': AutoProcessor.from_pretrained('microsoft/Florence-2-large-ft', trust_remote_code=True),
-    'microsoft/Florence-2-large': AutoProcessor.from_pretrained('microsoft/Florence-2-large', trust_remote_code=True),
-    'microsoft/Florence-2-base-ft': AutoProcessor.from_pretrained('microsoft/Florence-2-base-ft', trust_remote_code=True),
-    'microsoft/Florence-2-base': AutoProcessor.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True),
-}
-
-
 DESCRIPTION = "# [Florence-2 Demo](https://huggingface.co/microsoft/Florence-2-large)"
 
 colormap = ['blue','orange','green','purple','brown','pink','gray','olive','cyan','red',
@@ -42,7 +20,6 @@ def fig_to_pil(fig):
     buf.seek(0)
     return Image.open(buf)
 
-@spaces.GPU
 def run_example(task_prompt, image, text_input=None, model_id='microsoft/Florence-2-large'):
     model = models[model_id]
     processor = processors[model_id]
@@ -50,7 +27,7 @@ def run_example(task_prompt, image, text_input=None, model_id='microsoft/Florenc
         prompt = task_prompt
     else:
         prompt = task_prompt + text_input
-    inputs = processor(text=prompt, images=image, return_tensors="pt").to("cpu")
+    inputs = processor(text=prompt, images=image, return_tensors="pt")
     generated_ids = model.generate(
         input_ids=inputs["input_ids"],
         pixel_values=inputs["pixel_values"],
@@ -64,7 +41,6 @@ def run_example(task_prompt, image, text_input=None, model_id='microsoft/Florenc
         generated_text,
         task=task_prompt,
         image_size=(image.width, image.height)
-    )
     return parsed_answer
 
 def plot_bbox(image, data):
@@ -79,7 +55,6 @@ def plot_bbox(image, data):
     return fig
 
 def draw_polygons(image, prediction, fill_mask=False):
-
     draw = ImageDraw.Draw(image)
     scale = 1
     for polygons, label in zip(prediction['polygons'], prediction['labels']):
@@ -219,7 +194,22 @@ def process_image(image, task_prompt, text_input=None, model_id='microsoft/Flore
         output_image = draw_ocr_bboxes(output_image, results['<OCR_WITH_REGION>'])
         return results, output_image
     else:
-        return "", None  # Return empty string and None for unknown task prompts
+        return "", None
+
+# Load models for CPU
+models = {
+    'microsoft/Florence-2-large-ft': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-large-ft', trust_remote_code=True).eval(),
+    'microsoft/Florence-2-large': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-large', trust_remote_code=True).eval(),
+    'microsoft/Florence-2-base-ft': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-base-ft', trust_remote_code=True).eval(),
+    'microsoft/Florence-2-base': AutoModelForCausalLM.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True).eval(),
+}
+
+processors = {
+    'microsoft/Florence-2-large-ft': AutoProcessor.from_pretrained('microsoft/Florence-2-large-ft', trust_remote_code=True),
+    'microsoft/Florence-2-large': AutoProcessor.from_pretrained('microsoft/Florence-2-large', trust_remote_code=True),
+    'microsoft/Florence-2-base-ft': AutoProcessor.from_pretrained('microsoft/Florence-2-base-ft', trust_remote_code=True),
+    'microsoft/Florence-2-base': AutoProcessor.from_pretrained('microsoft/Florence-2-base', trust_remote_code=True),
+}
 
 css = """
   #output {
@@ -229,7 +219,6 @@ css = """
   }
 """
 
-
 single_task_list =[
     'Caption', 'Detailed Caption', 'More Detailed Caption', 'Object Detection',
     'Dense Region Caption', 'Region Proposal', 'Caption to Phrase Grounding',
@@ -242,15 +231,12 @@ cascased_task_list =[
     'Caption + Grounding', 'Detailed Caption + Grounding', 'More Detailed Caption + Grounding'
 ]
 
-
 def update_task_dropdown(choice):
     if choice == 'Cascased task':
-        return gr.Dropdown(choices=cascaded_task_list, value='Caption + Grounding')
+        return gr.Dropdown(choices=cascased_task_list, value='Caption + Grounding')
     else:
         return gr.Dropdown(choices=single_task_list, value='Caption')
 
-
-
 with gr.Blocks(css=css) as demo:
     gr.Markdown(DESCRIPTION)
     with gr.Tab(label="Florence-2 Image Captioning"):
@@ -258,7 +244,7 @@ with gr.Blocks(css=css) as demo:
             with gr.Column():
                 input_img = gr.Image(label="Input Picture")
                 model_selector = gr.Dropdown(choices=list(models.keys()), label="Model", value='microsoft/Florence-2-large')
-                task_type = gr.Radio(choices=['Single task', 'Cascaded task'], label='Task type selector', value='Single task')
+                task_type = gr.Radio(choices=['Single task', 'Cascased task'], label='Task type selector', value='Single task')
                 task_prompt = gr.Dropdown(choices=single_task_list, label="Task Prompt", value="Caption")
                 task_type.change(fn=update_task_dropdown, inputs=task_type, outputs=task_prompt)
                 text_input = gr.Textbox(label="Text Input (optional)")
@@ -281,5 +267,4 @@ with gr.Blocks(css=css) as demo:
 
         submit_btn.click(process_image, [input_img, task_prompt, text_input, model_selector], [output_text, output_img])
 
-# launch the demo
 demo.launch(debug=True)