Spaces:
Runtime error
Runtime error
Dylan
commited on
Commit
Β·
a4690cb
1
Parent(s):
98efca2
added description agents -- dummy
Browse files- agents.py +146 -0
- app.backup.py +9 -0
- app.py +59 -4
- helpers.py +9 -0
agents.py
ADDED
|
@@ -0,0 +1,146 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import torch
|
| 2 |
+
from langgraph.graph import END, StateGraph
|
| 3 |
+
from typing import TypedDict, Any
|
| 4 |
+
|
| 5 |
+
from transformers import (
|
| 6 |
+
AutoProcessor,
|
| 7 |
+
BitsAndBytesConfig,
|
| 8 |
+
Gemma3ForConditionalGeneration,
|
| 9 |
+
)
|
| 10 |
+
|
| 11 |
+
|
| 12 |
+
def get_quantization_config():
|
| 13 |
+
return BitsAndBytesConfig(
|
| 14 |
+
load_in_4bit=True,
|
| 15 |
+
bnb_4bit_compute_dtype=torch.float16,
|
| 16 |
+
bnb_4bit_quant_type="nf4",
|
| 17 |
+
bnb_4bit_use_double_quant=True,
|
| 18 |
+
)
|
| 19 |
+
|
| 20 |
+
|
| 21 |
+
# Define the state schema
|
| 22 |
+
class State(TypedDict):
|
| 23 |
+
image: Any
|
| 24 |
+
voice: str
|
| 25 |
+
caption: str
|
| 26 |
+
description: str
|
| 27 |
+
|
| 28 |
+
|
| 29 |
+
# Build the workflow graph
|
| 30 |
+
def build_graph():
|
| 31 |
+
workflow = StateGraph(State)
|
| 32 |
+
|
| 33 |
+
# Add nodes
|
| 34 |
+
workflow.add_node("caption_image", caption_image)
|
| 35 |
+
workflow.add_node("describe_with_voice", describe_with_voice)
|
| 36 |
+
|
| 37 |
+
# Add edges
|
| 38 |
+
workflow.set_entry_point("caption_image")
|
| 39 |
+
workflow.add_edge("caption_image", "describe_with_voice")
|
| 40 |
+
workflow.add_edge("describe_with_voice", END)
|
| 41 |
+
|
| 42 |
+
# Compile the graph
|
| 43 |
+
return workflow.compile()
|
| 44 |
+
|
| 45 |
+
|
| 46 |
+
model_id = "google/gemma-3-4b-it"
|
| 47 |
+
|
| 48 |
+
# Initialize processor and model
|
| 49 |
+
processor = AutoProcessor.from_pretrained(model_id)
|
| 50 |
+
model = Gemma3ForConditionalGeneration.from_pretrained(
|
| 51 |
+
model_id,
|
| 52 |
+
# quantization_config=get_quantization_config(),
|
| 53 |
+
device_map="auto",
|
| 54 |
+
torch_dtype=torch.float16,
|
| 55 |
+
)
|
| 56 |
+
|
| 57 |
+
|
| 58 |
+
def describe_with_voice(state: State) -> State:
|
| 59 |
+
state["description"] = "Dummy description"
|
| 60 |
+
return state
|
| 61 |
+
|
| 62 |
+
|
| 63 |
+
def caption_image(state: State) -> State:
|
| 64 |
+
state["caption"] = "Dummy caption"
|
| 65 |
+
|
| 66 |
+
|
| 67 |
+
def describe_with_voice2(state: State) -> State:
|
| 68 |
+
caption = state["caption"]
|
| 69 |
+
voice = state["voice"]
|
| 70 |
+
|
| 71 |
+
# Voice prompt templates
|
| 72 |
+
voice_prompts = {
|
| 73 |
+
"scurvy-ridden pirate": "You are a scurvy-ridden pirate, angry and drunk.",
|
| 74 |
+
"forgetful wizard": "You are a forgetful and easily distracted wizard.",
|
| 75 |
+
"sarcastic teenager": "You are a sarcastic and disinterested teenager.",
|
| 76 |
+
}
|
| 77 |
+
messages = [
|
| 78 |
+
{"role": "system", "content": [voice_prompts.get(voice)]},
|
| 79 |
+
{
|
| 80 |
+
"role": "user",
|
| 81 |
+
"content": [
|
| 82 |
+
{"type": "text", "text": f"Describe the following:\n\n{caption}"}
|
| 83 |
+
],
|
| 84 |
+
},
|
| 85 |
+
]
|
| 86 |
+
inputs = processor.apply_chat_template(
|
| 87 |
+
messages,
|
| 88 |
+
add_generation_prompt=True,
|
| 89 |
+
tokenize=True,
|
| 90 |
+
return_dict=True,
|
| 91 |
+
return_tensors="pt",
|
| 92 |
+
).to(model.device, dtype=torch.bfloat16)
|
| 93 |
+
input_len = inputs["input_ids"].shape[-1]
|
| 94 |
+
|
| 95 |
+
with torch.inference_mode():
|
| 96 |
+
generation = model.generate(**inputs, max_new_tokens=100, do_sample=False)
|
| 97 |
+
generation = generation[0][input_len:]
|
| 98 |
+
|
| 99 |
+
description = processor.decode(generation, skip_special_tokens=True)
|
| 100 |
+
|
| 101 |
+
state["description"] = description
|
| 102 |
+
|
| 103 |
+
return state
|
| 104 |
+
|
| 105 |
+
|
| 106 |
+
def caption_image2(state: State) -> State:
|
| 107 |
+
# image is PIL
|
| 108 |
+
image = state["image"]
|
| 109 |
+
|
| 110 |
+
# Load models (in practice, do this once and cache)
|
| 111 |
+
messages = [
|
| 112 |
+
{
|
| 113 |
+
"role": "system",
|
| 114 |
+
"content": [
|
| 115 |
+
{
|
| 116 |
+
"type": "text",
|
| 117 |
+
"text": "You are a helpful assistant that will describe images in 3-5 sentences.",
|
| 118 |
+
}
|
| 119 |
+
],
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"role": "user",
|
| 123 |
+
"content": [
|
| 124 |
+
{"type": "image", "image": image},
|
| 125 |
+
{"type": "text", "text": "Describe this image."},
|
| 126 |
+
],
|
| 127 |
+
},
|
| 128 |
+
]
|
| 129 |
+
inputs = processor.apply_chat_template(
|
| 130 |
+
messages,
|
| 131 |
+
add_generation_prompt=True,
|
| 132 |
+
tokenize=True,
|
| 133 |
+
return_dict=True,
|
| 134 |
+
return_tensors="pt",
|
| 135 |
+
).to(model.device, dtype=torch.bfloat16)
|
| 136 |
+
input_len = inputs["input_ids"].shape[-1]
|
| 137 |
+
|
| 138 |
+
with torch.inference_mode():
|
| 139 |
+
generation = model.generate(**inputs, max_new_tokens=100, do_sample=False)
|
| 140 |
+
generation = generation[0][input_len:]
|
| 141 |
+
|
| 142 |
+
caption = processor.decode(generation, skip_special_tokens=True)
|
| 143 |
+
|
| 144 |
+
state["caption"] = caption
|
| 145 |
+
|
| 146 |
+
return state
|
app.backup.py
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import gradio as gr
|
| 2 |
+
|
| 3 |
+
|
| 4 |
+
def greet(name):
|
| 5 |
+
return "Hello " + name + "!!"
|
| 6 |
+
|
| 7 |
+
|
| 8 |
+
demo = gr.Interface(fn=greet, inputs="text", outputs="text")
|
| 9 |
+
demo.launch()
|
app.py
CHANGED
|
@@ -1,7 +1,62 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
|
| 3 |
-
|
| 4 |
-
return "Hello " + name + "!!"
|
| 5 |
|
| 6 |
-
|
| 7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
|
| 3 |
+
from agents import build_graph
|
|
|
|
| 4 |
|
| 5 |
+
# Initialize the graph
|
| 6 |
+
graph = build_graph()
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
def process_and_display(image, voice):
|
| 10 |
+
# Initialize state
|
| 11 |
+
state = {"image": image, "voice": voice, "caption": "", "description": ""}
|
| 12 |
+
|
| 13 |
+
# Run the graph
|
| 14 |
+
result = graph.invoke(state)
|
| 15 |
+
|
| 16 |
+
# Return the caption and description
|
| 17 |
+
return result["caption"], result["description"]
|
| 18 |
+
|
| 19 |
+
|
| 20 |
+
def create_interface():
|
| 21 |
+
with gr.Blocks() as demo:
|
| 22 |
+
gr.Markdown("# Image Description with Voice Personas")
|
| 23 |
+
gr.Markdown("""
|
| 24 |
+
This app takes an image and generates a description using a selected voice persona.
|
| 25 |
+
|
| 26 |
+
1. Upload an image
|
| 27 |
+
2. Select a voice persona from the dropdown
|
| 28 |
+
3. Click "Generate Description" to see the results
|
| 29 |
+
""")
|
| 30 |
+
|
| 31 |
+
with gr.Row():
|
| 32 |
+
with gr.Column():
|
| 33 |
+
image_input = gr.Image(type="pil", label="Upload an Image")
|
| 34 |
+
voice_dropdown = gr.Dropdown(
|
| 35 |
+
choices=[
|
| 36 |
+
"scurvy-ridden pirate",
|
| 37 |
+
"forgetful wizard",
|
| 38 |
+
"sarcastic teenager",
|
| 39 |
+
],
|
| 40 |
+
label="Select a Voice",
|
| 41 |
+
value="scurvy-ridden pirate",
|
| 42 |
+
)
|
| 43 |
+
submit_button = gr.Button("Generate Description")
|
| 44 |
+
|
| 45 |
+
with gr.Column():
|
| 46 |
+
caption_output = gr.Textbox(label="Image Caption")
|
| 47 |
+
description_output = gr.Textbox(label="Voice Description")
|
| 48 |
+
|
| 49 |
+
submit_button.click(
|
| 50 |
+
fn=process_and_display,
|
| 51 |
+
inputs=[image_input, voice_dropdown],
|
| 52 |
+
outputs=[caption_output, description_output],
|
| 53 |
+
)
|
| 54 |
+
|
| 55 |
+
return demo
|
| 56 |
+
|
| 57 |
+
|
| 58 |
+
# Launch the app
|
| 59 |
+
demo = create_interface()
|
| 60 |
+
|
| 61 |
+
if __name__ == "__main__":
|
| 62 |
+
demo.launch()
|
helpers.py
ADDED
|
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import base64
|
| 2 |
+
import io
|
| 3 |
+
|
| 4 |
+
|
| 5 |
+
def image_to_base64(image):
|
| 6 |
+
"""Convert PIL Image to base64 encoded string"""
|
| 7 |
+
img_byte_arr = io.BytesIO()
|
| 8 |
+
image.save(img_byte_arr, format="JPEG")
|
| 9 |
+
return base64.b64encode(img_byte_arr.getvalue()).decode("utf-8")
|