Spaces:
Runtime error
Runtime error
File size: 2,530 Bytes
2e744f0 7835ab6 2e744f0 1f7ec24 2e744f0 1f7ec24 2e744f0 dd27d00 2e744f0 1f7ec24 806f2ad 1f7ec24 2e744f0 806f2ad 1f7ec24 806f2ad 2e744f0 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 |
import os
import requests
import threading
from typing import Optional, List, Tuple
import gradio as gr
ENDPOINT_URL = "https://austrian-code-wizard--metarlaif-web-dev.modal.run"
def get_feedback_options() -> List[Tuple[str, str]]:
args = {
"C3PO_API_KEY": os.environ.get("C3PO_API_KEY"),
}
response = requests.post(f"{ENDPOINT_URL}/list_adapters", json=args)
data = response.json()["adapters"]
return [
(adapter["feedback_name"], adapter["feedback_id"])
for adapter in data]
def get_completion(prompt: str, adapter: Optional[str], enable_feedback: bool) -> str:
args = {
"C3PO_API_KEY": os.environ.get("C3PO_API_KEY"),
"prompt": prompt if enable_feedback else None,
"adapter": adapter,
}
response = requests.post(f"{ENDPOINT_URL}/completion", json=args)
data = response.json()
return data["response"]
def warmup(*args):
args = {
"C3PO_API_KEY": os.environ.get("C3PO_API_KEY"),
}
# Warmup the server but don't wait for the response
threading.Thread(target=requests.post, args=(f"{ENDPOINT_URL}/warmup"), kwargs={"json": args}, daemon=True).start()
dropdown_options = get_feedback_options()
demo = gr.Interface(
get_completion,
[
gr.Markdown(
"""
# C3PO Demo
This is a demo of Contextualized Critiques with Constrained Preference Optimization (C3PO). See the project website [here](<insert link>), repo [here](<insert link>), and the paper [here](<insert link>).
Selecting a feedback in the dropdown and enabling the "Use Feedback Adapter" checkbox will add the respective adapter to the model. The model will then use the feedback to generate the completion.
### Warning
The model is hosted on [Modal](https://www.modal.com/). While Modal provides fast cold-start times, the model container might need to spin up if it's not already running. This might take up to a minute on the first request.
"""
),
gr.Textbox(
placeholder="Enter a prompt...", label="Prompt"
),
gr.Dropdown(
choices=dropdown_options, label="Feedback", info="Will add the adapter for the respective feedback to the model.",
value=dropdown_options[0][1]
),
gr.Checkbox(
True, label="Use Feedback Adapter", info="Enables the feedback adapter."
)
],
"text",
concurrency_limit=8
)
if __name__ == "__main__":
demo.queue(max_size=32)
demo.launch()
|