app.py

import gradio as gr
from test_prompt_generator import generate_prompt

tokenizers = [
    "google/gemma-7b",
    "mistralai/Mistral-7B-v0.1",
    "facebook/opt-2.7b",
    "microsoft/phi-2",
    "THUDM/chatglm3-6b",
    "Qwen/Qwen1.5-7B-Chat",
    "bigscience/bloom-560m",
    "ise-uiuc/Magicoder-S-DS-6.7B",
    "google/flan-t5-base",
    "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
    "google-bert/bert-base-uncased",
]


def generate(model_id, num_tokens, prefix=None, source_text=None):
    output_file = f"prompt_{num_tokens}.jsonl"
    prompt = generate_prompt(model_id, int(num_tokens), prefix=prefix, source_text=source_text, output_file=output_file)
    return prompt, output_file


demo = gr.Interface(
    fn=generate,
    title="Test Prompt Generator",
    description="Generate prompts with a given number of tokens for testing transformer models. "
    "Prompt source: https://archive.org/stream/alicesadventures19033gut/19033.txt",
    inputs=[
        gr.Dropdown(
            label="Tokenizer",
            choices=tokenizers,
            value="mistralai/Mistral-7B-v0.1",
            allow_custom_value=True,
            info="Select a tokenizer from this list or paste a model_id from a model on the Hugging Face Hub",
        ),
        gr.Number(
            label="Number of Tokens", minimum=4, maximum=2048, value=32, info="Enter a number between 4 and 2048."
        ),
        gr.Textbox(
            label="Prefix (optional)",
            info="If given, the start of the prompt will be this prefix. Example: 'Summarize the following text:'",
        ),
        gr.Textbox(
            label="Source text (optional)",
            info="By default, prompts will be generated from Alice in Wonderland. Enter text here to use that instead.",
        ),
    ],
    outputs=[gr.Textbox(label="prompt", show_copy_button=True), gr.File(label="Json file")],
    examples=[
        ["mistralai/Mistral-7B-v0.1", 32],
        ["mistralai/Mistral-7B-v0.1", 64],
        ["mistralai/Mistral-7B-v0.1", 128],
        ["mistralai/Mistral-7B-v0.1", 512],
        ["mistralai/Mistral-7B-v0.1", 1024],
        ["mistralai/Mistral-7B-v0.1", 2048],
    ],
    cache_examples=False,
    allow_flagging=False,
)

demo.launch()