from transformers import AutoModelForCausalLM, AutoTokenizer, LoraFromPretrained
import torch

# Load the base model and tokenizer
base_model_name = "armaniii/mistral-argument-classification/"
tokenizer = AutoTokenizer.from_pretrained(base_model_name)
base_model = AutoModelForCausalLM.from_pretrained(base_model_name)

# Load the LoRA adapter
lora_model_name = "mistral_lora"
lora_weights = LoraFromPretrained(lora_model_name).to(base_model.device)

# Merge the LoRA adapter with the base model
merged_model = base_model.merge_lora(lora_weights)

# Define your API endpoint
@app.post("/generate")
def generate(request_body):
    input_text = request_body["input_text"]
    ...
    # Use the merged model to generate output
    output = merged_model.generate(...)
    return {"output": output}