#!/usr/bin/env python
"""05 -- streaming output from a generator, with progress.

Yielding from the handler streams partial results to the browser. This is the
pattern behind token-by-token LLM output (see 07_chat_local_llm.py).
"""
import time
import gradio as gr
from _common import launch

TEXT = ("Streaming works by yielding instead of returning. "
        "Gradio keeps the connection open and pushes each yielded value.")


def stream(sentence, delay, progress=gr.Progress()):
    words = (sentence or TEXT).split()
    acc = []
    for i, w in enumerate(words):
        acc.append(w)
        progress((i + 1) / len(words), desc="streaming")
        time.sleep(float(delay))
        yield " ".join(acc)


with gr.Blocks(title="Streaming") as demo:
    gr.Markdown("## Streaming a generator")
    inp = gr.Textbox(label="Sentence", value=TEXT, lines=2)
    delay = gr.Slider(0.0, 0.5, value=0.08, step=0.01, label="Delay per word (s)")
    out = gr.Textbox(label="Streamed output", lines=4)
    gr.Button("Stream", variant="primary").click(
        stream, [inp, delay], out, api_name="stream")

if __name__ == "__main__":
    launch(demo)