main.py

import re
from fastapi import FastAPI, HTTPException
from fastapi.responses import StreamingResponse
from fastapi.responses import JSONResponse
from pydantic import BaseModel
from huggingface_hub import InferenceClient
import uvicorn
from typing import Generator, List
import json  # Asegúrate de que esta línea esté al principio del archivo
import nltk
import os
import google.protobuf  # This line should execute without errors if protobuf is installed correctly
import sentencepiece
from transformers import pipeline, AutoTokenizer,AutoModelForSequenceClassification,AutoModel
import spacy
import numpy as np
import torch


nltk.data.path.append(os.getenv('NLTK_DATA'))

app = FastAPI()

# Initialize the InferenceClient with your model
client = InferenceClient("mistralai/Mixtral-8x7B-Instruct-v0.1")


class Item(BaseModel):
    prompt: str
    history: list
    system_prompt: str
    temperature: float = 0.8
    max_new_tokens: int = 4000
    top_p: float = 0.15
    repetition_penalty: float = 1.0

def format_prompt(current_prompt, history):
    formatted_history = "<s>"
    for entry in history:
        if entry["role"] == "user":
            formatted_history += f"[USER] {entry['content']} [/USER]"
        elif entry["role"] == "assistant":
            formatted_history += f"[ASSISTANT] {entry['content']} [/ASSISTANT]"
    formatted_history += f"[USER] {current_prompt} [/USER]</s>"
    return formatted_history


def generate_stream(item: Item) -> Generator[bytes, None, None]:
    formatted_prompt = format_prompt(f"{item.system_prompt}, {item.prompt}", item.history)
    # Estimate token count for the formatted_prompt
    input_token_count = len(nltk.word_tokenize(formatted_prompt))  # NLTK tokenization

    # Ensure total token count doesn't exceed the maximum limit
    max_tokens_allowed = 32768
    max_new_tokens_adjusted = max(1, min(item.max_new_tokens, max_tokens_allowed - input_token_count))

    generate_kwargs = {
        "temperature": item.temperature,
        "max_new_tokens": max_new_tokens_adjusted,
        "top_p": item.top_p,
        "repetition_penalty": item.repetition_penalty,
        "do_sample": True,
        "seed": 42,
    }

    # Stream the response from the InferenceClient
    for response in client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True):
        # This assumes 'details=True' gives you a structure where you can access the text like this
        chunk = {
            "text": response.token.text,
            "complete": response.generated_text is not None  # Adjust based on how you detect completion
        }
        yield json.dumps(chunk).encode("utf-8") + b"\n"


class SummarizeRequest(BaseModel):
    text: str

@app.post("/generate/")
async def generate_text(item: Item):
    # Stream response back to the client
    return StreamingResponse(generate_stream(item), media_type="application/x-ndjson")

# Define request model
class TextRequest(BaseModel):
    text: str  # Single string of long text

# Load Longformer model and tokenizer
tokenizer = AutoTokenizer.from_pretrained("allenai/longformer-base-4096")
model = AutoModel.from_pretrained("allenai/longformer-base-4096")

# Endpoint to process the document and return embeddings
@app.post("/process_document")
async def process_document(request: TextRequest):
    try:
        # Split the text into segments that fit within the model's max input size
        max_length = 4096  # Maximum token length for Longformer
        words = request.text.split()
        tokens = tokenizer.encode(request.text, add_special_tokens=True)
        input_ids = []
        current_chunk = []

        for token in tokens:
            if len(current_chunk) + len(tokenizer.convert_ids_to_tokens([token])) < max_length:
                current_chunk.append(token)
            else:
                input_ids.append(current_chunk)
                current_chunk = [token]

        if current_chunk:
            input_ids.append(current_chunk)  # Add the last chunk if any

        # Generate embeddings for each segment
        embeddings_list = []
        for ids in input_ids:
            inputs = {'input_ids': torch.tensor(ids).unsqueeze(0)}  # Batch size 1
            outputs = model(**inputs)
            embeddings = outputs.last_hidden_state.mean(dim=1).detach().numpy()
            embeddings_list.append(embeddings.tolist())  # Store embeddings for each segment

        return {
            "embeddings": embeddings_list
        }
    except Exception as e:
        print(f"Error during document processing: {e}")
        raise HTTPException(status_code=500, detail=str(e))

# @app.post("/summarize")
# async def summarize(request: TextRequest):
#     try:
#         # Preprocess and segment the text
#         processed_text = preprocess_text(request.text)
#         segments = segment_text(processed_text)

#         # Classify each segment safely
#         classified_segments = []
#         for segment in segments:
#             try:
#                 result = classifier(segment)
#                 classified_segments.append(result)
#             except Exception as e:
#                 print(f"Error classifying segment: {e}")
#                 classified_segments.append({"error": str(e)})

#         # Optional: Reduce tokens or summarize
#         reduced_texts = []
#         for segment in segments:
#             try:
#                 reduced_text, token_count = reduce_tokens(segment)
#                 reduced_texts.append((reduced_text, token_count))
#             except Exception as e:
#                 print(f"Error during token reduction: {e}")
#                 reduced_texts.append(("Error", 0))

#         return {
#             "classified_segments": classified_segments,
#             "reduced_texts": reduced_texts
#         }
    
#     except Exception as e:
#         print(f"Error during token reduction: {e}")
#         raise HTTPException(status_code=500, detail=str(e))

# if __name__ == "__main__":
#     uvicorn.run(app, host="0.0.0.0", port=8000)