handler.py

# handler.py
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel

# Base model that your LoRA was trained on (must match training)
BASE_MODEL = "deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B"   # change if you trained on a different DeepSeek variant
ADAPTER_PATH = "GilbertAkham/deepseek-R1-multitask-lora"

class EndpointHandler:
    def __init__(self, path=""):
        print("🚀 Loading base model...")
        self.tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL, trust_remote_code=True)

        # Load base model
        base_model = AutoModelForCausalLM.from_pretrained(
            BASE_MODEL,
            torch_dtype=torch.float16,
            device_map="auto",
            trust_remote_code=True
        )

        print(f"🔗 Attaching LoRA adapter from {ADAPTER_PATH}...")
        # Load the LoRA adapter properly
        self.model = PeftModel.from_pretrained(base_model, ADAPTER_PATH)
        self.model.eval()

        print("✅ Model + LoRA adapter loaded successfully.")

    def __call__(self, data):
        prompt = data.get("inputs", "")
        inputs = self.tokenizer(prompt, return_tensors="pt").to(self.model.device)

        with torch.no_grad():
            outputs = self.model.generate(
                **inputs,
                max_new_tokens=512,
                temperature=0.7,
                top_p=0.9,
                do_sample=True,
                pad_token_id=self.tokenizer.eos_token_id,
                eos_token_id=self.tokenizer.eos_token_id,
            )

        text = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
        return {"generated_text": text}