main.py

from fastapi import FastAPI, HTTPException
from pydantic import BaseModel
from transformers import pipeline
import torch
from fastapi.middleware.cors import CORSMiddleware

app = FastAPI(title="Model Inference API")

# Allow CORS for external frontend
app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_methods=["*"],
    allow_headers=["*"],
)

MODEL_MAP = {
    "tinny-llama": "Lyon28/Tinny-Llama",
    "pythia": "Lyon28/Pythia",
    "bert-tinny": "Lyon28/Bert-Tinny",
    "albert-base-v2": "Lyon28/Albert-Base-V2",
    "t5-small": "Lyon28/T5-Small",
    "gpt-2": "Lyon28/GPT-2",
    "gpt-neo": "Lyon28/GPT-Neo",
    "distilbert-base-uncased": "Lyon28/Distilbert-Base-Uncased",
    "distil-gpt-2": "Lyon28/Distil_GPT-2",
    "gpt-2-tinny": "Lyon28/GPT-2-Tinny",
    "electra-small": "Lyon28/Electra-Small"
}

TASK_MAP = {
    "text-generation": ["gpt-2", "gpt-neo", "distil-gpt-2", "gpt-2-tinny", "tinny-llama", "pythia"],
    "text-classification": ["bert-tinny", "albert-base-v2", "distilbert-base-uncased", "electra-small"],
    "text2text-generation": ["t5-small"]
}

class InferenceRequest(BaseModel):
    text: str
    max_length: int = 100
    temperature: float = 0.9

def get_task(model_id: str):
    for task, models in TASK_MAP.items():
        if model_id in models:
            return task
    return "text-generation"

@app.on_event("startup")
async def load_models():
    # Initialize models (optional: pre-load critical models)
    app.state.pipelines = {}
    print("Models initialized in memory")

@app.post("/inference/{model_id}")
async def model_inference(model_id: str, request: InferenceRequest):
    try:
        if model_id not in MODEL_MAP:
            raise HTTPException(status_code=404, detail="Model not found")

        task = get_task(model_id)
        
        # Load pipeline with caching
        if model_id not in app.state.pipelines:
            app.state.pipelines[model_id] = pipeline(
                task=task,
                model=MODEL_MAP[model_id],
                device_map="auto",
                torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32
            )

        pipe = app.state.pipelines[model_id]

        # Process based on task
        if task == "text-generation":
            result = pipe(
                request.text,
                max_length=request.max_length,
                temperature=request.temperature
            )[0]['generated_text']
        
        elif task == "text-classification":
            output = pipe(request.text)[0]
            result = {
                "label": output['label'],
                "confidence": round(output['score'], 4)
            }
        
        elif task == "text2text-generation":
            result = pipe(request.text)[0]['generated_text']

        return {"result": result}
    
    except Exception as e:
        raise HTTPException(status_code=500, detail=str(e))

@app.get("/models")
async def list_models():
    return {"available_models": list(MODEL_MAP.keys())}

@app.get("/health")
async def health_check():
    return {"status": "healthy"}