import json

import spaces
import gradio as gr
import torch
from distilabel.llms import LlamaCppLLM
from distilabel.steps.tasks.argillalabeller import ArgillaLabeller

file_path = os.path.join(os.path.dirname(__file__), "qwen2-0_5b-instruct-fp16.gguf")
download_url = "https://huggingface.co/Qwen/Qwen2-0.5B-Instruct-GGUF/resolve/main/qwen2-0_5b-instruct-fp16.gguf?download=true"


if not os.path.exists(file_path):
    import requests
    import tqdm

    response = requests.get(download_url, stream=True)
    total_length = int(response.headers.get("content-length"))

    with open(file_path, "wb") as f:
        for chunk in tqdm.tqdm(
            response.iter_content(chunk_size=1024),
            total=total_length,
            unit="KB",
            unit_scale=True,
        ):
            f.write(chunk)


llm = LlamaCppLLM(
    model_path=file_path,
    n_gpu_layers=-1,
    n_ctx=1024 * 4,
)
task = ArgillaLabeller(llm=llm)
task.load()


def load_examples():
    with open("examples.json", "r") as f:
        return json.load(f)


# Create Gradio examples
examples = load_examples()


def process_fields(fields):
    if isinstance(fields, str):
        fields = json.loads(fields)
    if isinstance(fields, dict):
        fields = [fields]
    return [field if isinstance(field, dict) else json.loads(field) for field in fields]

@spaces.GPU
def process_records_gradio(records, example_records, fields, question):
    try:
        # Convert string inputs to dictionaries
        records = json.loads(records)
        example_records = json.loads(example_records) if example_records else None
        fields = process_fields(fields) if fields else None
        question = json.loads(question) if question else None

        print(fields)
        print(question)
        print(example_records)

        if not fields and not question:
            return "Error: Either fields or question must be provided"

        runtime_parameters = {"fields": fields, "question": question}
        if example_records:
            runtime_parameters["example_records"] = example_records
        print(runtime_parameters)
        task.set_runtime_parameters(runtime_parameters)

        results = []
        for record in records:
            output = next(task.process(inputs=[{"records": record}]))
            results.append(output[0]["suggestions"])

        return json.dumps({"results": results}, indent=2)
    except Exception as e:
        return f"Error: {str(e)}"


interface = gr.Interface(
    fn=process_records_gradio,
    inputs=[
        gr.Code(label="Records (JSON)", language="json", lines=5),
        gr.Code(label="Example Records (JSON, optional)", language="json", lines=5),
        gr.Code(label="Fields (JSON, optional)", language="json"),
        gr.Code(label="Question (JSON, optional)", language="json"),
    ],
    examples=examples,
    outputs=gr.Code(label="Suggestions", language="json", lines=10),
    title="Record Processing Interface",
    description="Enter JSON data for `rg.Record.to_dict()`, `List[rg.Record.to_dict()]`, `List[Field].serialize()`, or `List[rg.Question.serialize()]` At least one of fields or question must be provided.",
)

if __name__ == "__main__":
    interface.launch()