|
import os |
|
import torch |
|
import multiprocessing |
|
from fastapi import FastAPI, Request |
|
from pydantic import BaseModel |
|
from transformers import AutoModelForCausalLM, AutoTokenizer |
|
from dotenv import load_dotenv |
|
from accelerate import Accelerator |
|
|
|
|
|
load_dotenv() |
|
|
|
|
|
html_content = """ |
|
<!DOCTYPE html> |
|
<html> |
|
<head> |
|
<title>Llama-3.2-1B-Instruct-API</title> |
|
</head> |
|
<body> |
|
<div style="text-align: center;"> |
|
<a href="https://buymeacoffee.com/xxparthparekhxx" target="_blank"> |
|
<img src="https://cdn.buymeacoffee.com/buttons/v2/default-yellow.png" |
|
alt="Buy Me A Coffee" |
|
height="40px"> |
|
</a> |
|
<h2>Please Chill Out! 😎</h2> |
|
<p>This API takes around <strong>5.62 minutes</strong> to process a single request due to current hardware limitations.</p> |
|
<h3>Want Faster Responses? Help Me Out! 🚀</h3> |
|
<p>If you'd like to see this API running faster on high-performance <strong>A100</strong> hardware, please consider buying me a coffee. ☕ Your support will go towards upgrading to <strong>Hugging Face Pro</strong>, which will allow me to run A100-powered spaces for everyone! 🙌</p> |
|
<h4>Instructions to Clone and Run Locally:</h4> |
|
<ol> |
|
<li><strong>Clone the Repository:</strong> |
|
<pre> |
|
git clone https://huggingface.co/spaces/xxparthparekhxx/llama-3.2-1B-FastApi |
|
cd llama-3.2-1B-FastApi |
|
</pre> |
|
</li> |
|
<li><strong>Run the Docker container:</strong> |
|
<pre> |
|
docker build -t llama-api . |
|
docker run -p 7860:7860 llama-api |
|
</pre> |
|
</li> |
|
<li><strong>Access the API locally:</strong> |
|
<p>Open <a href="http://localhost:7860">http://localhost:7860</a> to access the API docs locally.</p> |
|
</li> |
|
</ol> |
|
</div> |
|
</body> |
|
</html> |
|
""" |
|
|
|
|
|
app = FastAPI( |
|
title="Llama-3.2-1B-Instruct-API", |
|
description= html_content, |
|
docs_url="/", |
|
redoc_url="/doc" |
|
) |
|
|
|
HF_TOKEN = os.getenv("HF_TOKEN") |
|
|
|
MODEL = "meta-llama/Llama-3.2-1B-Instruct" |
|
|
|
|
|
device = "cuda" if torch.cuda.is_available() else "cpu" |
|
print(f"Using device: {device}") |
|
|
|
|
|
torch.set_num_threads(multiprocessing.cpu_count()) |
|
|
|
|
|
accelerator = Accelerator() |
|
|
|
|
|
tokenizer = AutoTokenizer.from_pretrained(MODEL, token=HF_TOKEN, use_fast=True) |
|
model = AutoModelForCausalLM.from_pretrained( |
|
MODEL, |
|
token=HF_TOKEN, |
|
torch_dtype=torch.float16, |
|
device_map="auto" |
|
) |
|
|
|
|
|
model, tokenizer = accelerator.prepare(model, tokenizer) |
|
|
|
|
|
class PromptRequest(BaseModel): |
|
prompt: str |
|
max_new_tokens: int = 100 |
|
temperature: float = 0.7 |
|
|
|
@app.post("/generate/") |
|
async def generate_text(request: PromptRequest): |
|
inputs = tokenizer(request.prompt, return_tensors="pt").to(device) |
|
|
|
with torch.no_grad(): |
|
outputs = model.generate( |
|
**inputs, |
|
max_new_tokens=request.max_new_tokens, |
|
temperature=request.temperature, |
|
do_sample=False, |
|
pad_token_id=tokenizer.eos_token_id |
|
) |
|
|
|
response = tokenizer.decode(outputs[0], skip_special_tokens=True) |
|
return {"response": response} |
|
|
