path: root/interference/app.py

import json
import random
import string
import time
from typing import Any
import requests
from flask import Flask, request
from flask_cors import CORS
from transformers import AutoTokenizer
from g4f import ChatCompletion

app = Flask(__name__)
CORS(app)


@app.route("/chat/completions", methods=["POST"])
def chat_completions():
    model = request.get_json().get("model", "gpt-3.5-turbo")
    stream = request.get_json().get("stream", False)
    messages = request.get_json().get("messages")

    response = ChatCompletion.create(model=model, stream=stream, messages=messages)

    completion_id = "".join(random.choices(string.ascii_letters + string.digits, k=28))
    completion_timestamp = int(time.time())

    if not stream:
        return {
            "id": f"chatcmpl-{completion_id}",
            "object": "chat.completion",
            "created": completion_timestamp,
            "model": model,
            "choices": [
                {
                    "index": 0,
                    "message": {
                        "role": "assistant",
                        "content": response,
                    },
                    "finish_reason": "stop",
                }
            ],
            "usage": {
                "prompt_tokens": None,
                "completion_tokens": None,
                "total_tokens": None,
            },
        }

    def streaming():
        for chunk in response:
            completion_data = {
                "id": f"chatcmpl-{completion_id}",
                "object": "chat.completion.chunk",
                "created": completion_timestamp,
                "model": model,
                "choices": [
                    {
                        "index": 0,
                        "delta": {
                            "content": chunk,
                        },
                        "finish_reason": None,
                    }
                ],
            }

            content = json.dumps(completion_data, separators=(",", ":"))
            yield f"data: {content}\n\n"
            time.sleep(0.1)

        end_completion_data: dict[str, Any] = {
            "id": f"chatcmpl-{completion_id}",
            "object": "chat.completion.chunk",
            "created": completion_timestamp,
            "model": model,
            "choices": [
                {
                    "index": 0,
                    "delta": {},
                    "finish_reason": "stop",
                }
            ],
        }
        content = json.dumps(end_completion_data, separators=(",", ":"))
        yield f"data: {content}\n\n"

    return app.response_class(streaming(), mimetype="text/event-stream")


#Get the embedding from huggingface
def get_embedding(input_text, token):
    huggingface_token = token
    embedding_model = "sentence-transformers/all-mpnet-base-v2"
    max_token_length = 500

    # Load the tokenizer for the "all-mpnet-base-v2" model
    tokenizer = AutoTokenizer.from_pretrained(embedding_model)
    # Tokenize the text and split the tokens into chunks of 500 tokens each
    tokens = tokenizer.tokenize(input_text)
    token_chunks = [tokens[i:i + max_token_length] for i in range(0, len(tokens), max_token_length)]

    # Initialize an empty list
    embeddings = []

    # Create embeddings for each chunk
    for chunk in token_chunks:
        # Convert the chunk tokens back to text
        chunk_text = tokenizer.convert_tokens_to_string(chunk)

        # Use the Hugging Face API to get embeddings for the chunk
        api_url = f"https://api-inference.huggingface.co/pipeline/feature-extraction/{embedding_model}"
        headers = {"Authorization": f"Bearer {huggingface_token}"}
        chunk_text = chunk_text.replace("\n", " ")
    
        # Make a POST request to get the chunk's embedding
        response = requests.post(api_url, headers=headers, json={"inputs": chunk_text, "options": {"wait_for_model": True}})
    
        # Parse the response and extract the embedding
        chunk_embedding = response.json()
        # Append the embedding to the list
        embeddings.append(chunk_embedding)

    #averaging all the embeddings
    #this isn't very effective
    #someone a better idea?
    num_embeddings = len(embeddings)
    average_embedding = [sum(x) / num_embeddings for x in zip(*embeddings)]
    embedding = average_embedding
    return embedding


@app.route("/embeddings", methods=["POST"])
def embeddings():
    input_text_list = request.get_json().get("input")
    input_text = ' '.join(map(str, input_text_list))
    token = request.headers.get('Authorization').replace("Bearer ", "")
    embedding = get_embedding(input_text, token)
    return {
  "data": [
    {
     "embedding": embedding,
     "index": 0,
     "object": "embedding"
   }
  ],
 "model": "text-embedding-ada-002",
  "object": "list",
  "usage": {
    "prompt_tokens": None,
    "total_tokens": None
  }
  }

def main():
    app.run(host="0.0.0.0", port=1337, debug=True)


if __name__ == "__main__":
    main()