Fix: Load CPU-safe model for HF Space

app.py

@@ -2,20 +2,20 @@ import streamlit as st
 import torch
 from transformers import AutoTokenizer, AutoModelForCausalLM
 import re
+from peft import PeftModel
 from pydub import AudioSegment
 import speech_recognition as sr
 import io
 
 # Load model and tokenizer from local fine-tuned directory
-MODEL_PATH = "Tufan1/BioMedLM-Cardio-Fold4-CPU"
-tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
+# Define base and adapter model paths
+BASE_MODEL = "stanfordai/BioMedLM"  # or the path you used originally
+ADAPTER_PATH = "Tufan1/BioMedLM-Cardio-Fold4-CPU"
+tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
 
 # Force CPU-safe model loading
-model = AutoModelForCausalLM.from_pretrained(
-    MODEL_PATH,
-    torch_dtype=torch.float32,
-    low_cpu_mem_usage=True
-).to("cpu")
+base_model = AutoModelForCausalLM.from_pretrained(BASE_MODEL, torch_dtype=torch.float32)
+model = PeftModel.from_pretrained(base_model, ADAPTER_PATH, device_map=None).to("cpu")
 
 
 # Dictionaries to decode user inputs