app.py

import gradio as gr
import re
import requests
import json
import os
from screenshot import BG_COMP, BOX_COMP, GENERATION_VAR, PROMPT_VAR, main
from pathlib import Path

title = "BLOOM"
description = """Gradio Demo for BLOOM. To use it, simply add your text, or click one of the examples to load them.
Tips:
- Do NOT talk to BLOOM as an entity, it's not a chatbot but a webpage/blog/article completion model.
- For the best results: MIMIC a few sentences of a webpage similar to the content you want to generate.
Start a paragraph as if YOU were writing a blog, webpage, math post, coding article and BLOOM will generate a coherent follow-up. Longer prompts usually give more interesting results.
Options:
- sampling: imaginative completions (may be not super accurate e.g. math/history)
- greedy: accurate completions (may be more boring or have repetitions)
"""

API_URL = os.getenv("API_URL")
TOKEN = os.getenv("TOKEN")

examples = [
    [
        'A "whatpu" is a small, furry animal native to Tanzania. An example of a sentence that uses the word whatpu is: We were traveling in Africa and we saw these very cute whatpus. To do a "farduddle" means to jump up and down really fast. An example of a sentence that uses the word farduddle is:',
        32,
        "Sample",
        False,
        "Sample 1",
    ],
    [
        "A poem about the beauty of science by Alfred Edgar Brittle\nTitle: The Magic Craft\nIn the old times",
        50,
        "Sample",
        False,
        "Sample 1",
    ],
    ["استخراج العدد العاملي في لغة بايثون:", 30, "Greedy", False, "Sample 1"],
    ["Pour déguster un ortolan, il faut tout d'abord", 32, "Sample", False, "Sample 1"],
    [
        "Traduce español de España a español de Argentina\nEl coche es rojo - el auto es rojo\nEl ordenador es nuevo - la computadora es nueva\nel boligrafo es negro -",
        16,
        "Sample",
        False,
        "Sample 1",
    ],
    [
        "Estos ejemplos quitan vocales de las palabras\nEjemplos:\nhola - hl\nmanzana - mnzn\npapas - pps\nalacran - lcrn\npapa -",
        16,
        "Sample",
        False,
        "Sample 1",
    ],
    [
        "Question: If I put cheese into the fridge, will it melt?\nAnswer:",
        32,
        "Sample",
        False,
        "Sample 1",
    ],
    ["Math exercise - answers:\n34+10=44\n54+20=", 16, "Greedy", False, "Sample 1"],
    [
        "Question: Where does the Greek Goddess Persephone spend half of the year when she is not with her mother?\nAnswer:",
        24,
        "Greedy",
        False,
        "Sample 1",
    ],
    [
        "spelling test answers.\nWhat are the letters in « language »?\nAnswer: l-a-n-g-u-a-g-e\nWhat are the letters in « Romanian »?\nAnswer:",
        24,
        "Greedy",
        False,
        "Sample 1",
    ],
]


def query(payload):
    print(payload)
    response = requests.request("POST", API_URL, json=payload, headers={"Authorization": f"Bearer {TOKEN}"})
    print(response)
    return json.loads(response.content.decode("utf-8"))


def inference(input_sentence, max_length, sample_or_greedy, raw_text=False, seed=42):
    if sample_or_greedy == "Sample":
        parameters = {
            "max_new_tokens": max_length,
            "top_p": 0.9,
            "do_sample": True,
            "seed": seed,
            "early_stopping": False,
            "length_penalty": 0.0,
            "eos_token_id": None,
        }
    else:
        parameters = {
            "max_new_tokens": max_length,
            "do_sample": False,
            "seed": seed,
            "early_stopping": False,
            "length_penalty": 0.0,
            "eos_token_id": None,
        }

    payload = {"inputs": input_sentence, "parameters": parameters}

    data = query(payload)

    if raw_text:
        return None, data[0]["generated_text"]

    width, height = 3246, 3246
    assets_path = "assets"
    font_mapping = {
        "latin characters (faster)": "DejaVuSans.ttf",
        "complete alphabet (slower)": "GoNotoCurrent.ttf",
    }
    working_dir = Path(__file__).parent.resolve()
    font_path = str(working_dir / font_mapping["complete alphabet (slower)"])
    img_save_path = str(working_dir / "output.jpeg")
    colors = {
        BG_COMP: "#000000",
        PROMPT_VAR: "#FFFFFF",
        GENERATION_VAR: "#FF57A0",
        BOX_COMP: "#120F25",
    }

    new_string = data[0]["generated_text"].split(input_sentence, 1)[1]

    return data[0]["generated_text"]


gr.Interface(
    inference,
    [
        gr.inputs.Textbox(label="Input"),
        gr.inputs.Slider(1, 64, default=32, step=1, label="Tokens to generate"),
        gr.inputs.Radio(
            ["Sample", "Greedy"], label="Sample or greedy", default="Sample"
        ),
        gr.Checkbox(label="Just output raw text"),
        gr.inputs.Radio(
            ["Sample 1", "Sample 2", "Sample 3", "Sample 4", "Sample 5"],
            default="Sample 1",
            label="Sample other generations (only work in 'Sample' mode",
            type="index",
        ),
    ],
    ["text"],
    examples=examples,
    # article=article,
    cache_examples=False,
    title=title,
    description=description,
).launch()