from datasets import load_from_disk, load_dataset
import pandas as pd
import os
import gradio as gr

ds_with_embeddings = load_dataset("svjack/bloom-dialogue-generate-ds-zh", split="train")
ds_with_embeddings.add_faiss_index(column='embeddings')
from sentence_transformers import SentenceTransformer
encoder = SentenceTransformer("sentence-transformers/LaBSE")

def retrieve_search_df(question = "今天天气不错。", top_k = 10):
    question_embedding = encoder.encode(question)
    scores, retrieved_examples = ds_with_embeddings.get_nearest_examples('embeddings', question_embedding, k=top_k)
    sdf = pd.DataFrame(retrieved_examples)
    sdf["scores"] = scores
    return sdf[["question", "dialogue_text", "dialogue", "repo", "scores"]]

example_sample = [
    ["今天天气不错。", 3],
    ["你想吃点什么？", 5],
]

def demo_func(prefix, max_length):
    max_length = max(int(max_length), 3)
    l = retrieve_search_df(prefix, max_length)[["dialogue", "repo"]].values.tolist()
    assert type(l) == type([])
    return {
        "Dialogue Context": l
    }

demo = gr.Interface(
        fn=demo_func,
        inputs=[gr.Text(label = "Prefix"),
                gr.Number(label = "Top K", value = 10)
        ],
        outputs="json",
        title=f"Bloom and GPT Chinese Daliy Dialogue Generator 🌸🐰 sample search demonstration",
        description = 'This _example_ was **drive** from <br/><b><h4>[https://github.com/svjack/Daliy-Dialogue](https://github.com/svjack/Daliy-Dialogue)</h4></b>\n',
        examples=example_sample if example_sample else None,
        cache_examples = False
    )

demo.launch(server_name=None, server_port=None)