import gradio as gr
from huggingface_hub import InferenceClient

client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")

def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    # Construct the messages with prompt engineering
    messages = [{"role": "system", "content": system_message}]

    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})

    # Add the user message
    messages.append({"role": "user", "content": message})

    # Generate the response
    response = ""

    for message in client.chat_completion(
        messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        token = message.choices[0].delta.content
        response += token

    # Check if the response is empty or if required fields are missing
    if not response.strip():
        yield "No relevant information could be generated based on the provided query."

    yield response

# Enhanced system message for structured output
default_system_message = (
    "You are part of a geospatial AI agent/assistant and are aiding in parsing the query. When a user asks something,"
    "You are a specialized parser for a geospatial AI assistant that analyzes city open data. "
    "Your task is to extract structured information from user queries about city data. "
    "Respond with the following format ONLY (no other text):\n\n"
    "city: [City Name or blank if unclear]\n"
    "state: [State Name or blank if unclear]\n"
    "dataset1: [Primary Dataset Type]\n"
    "dataset2: [Secondary Dataset Type if applicable, otherwise leave blank]\n"
    "time_filter: [any time filters mentioned - today, yesterday, last_week, last_month, this_year, etc.]\n"
    #"spatial_filter: [any spatial filters from - proximity, Boundary, near, within or distance]\n"
    "count_filter: [any count or limit filters]\n"
    "operations: [any sequencial GIS operations needed from - append_gdfs, calculate_aspect, buffer_features, calculate_field, calculate_geometry_attributes, clip_features, coordinate_system_transformation, dissolve_features, excel_to_table, export_to_geojson, feature_to_point, generate_near_table, geocode_address, reverse_geocode, create_heatmap, intersect_features, join_field, map_layout, merge_datasets, near_analysis, network_analysis_suite, project_features, service_area, shortest_path, slope_calculation, spatial_data_validation, spatial_join, spatial_statistics_hotspot, split_features, symbolization, table_to_excel, union_features]\n\n"
    "Provide only the extracted information without explanation. If information is not present in the query, leave that field blank."
)

demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value=default_system_message, label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1,
            maximum=1.0,
            value=0.95,
            step=0.05,
            label="Top-p (nucleus sampling)",
        ),
    ],
)

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