Update app.py

app.py

@@ -455,13 +455,14 @@ from peft import get_peft_model, LoraConfig, TaskType
 # ✅ Force CPU execution for Streamlit Cloud
 device = torch.device("cpu")
 
-# 🔹 Load IBM Granite Model (CPU-Compatible)
+# 🔹 Load IBM Granite Model (No Shard Checkpoints)
 MODEL_NAME = "ibm-granite/granite-3.1-2b-instruct"
 
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_NAME,
-    device_map="cpu",  # Force CPU execution
-    torch_dtype=torch.float32  # Use float32 since Streamlit runs on CPU
+    device_map="cpu",
+    torch_dtype=torch.float32,
+    ignore_mismatched_sizes=True  # 🚀 Fixes sharded checkpoint issues
 )
 
 tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
@@ -580,7 +581,6 @@ if __name__ == '__main__':
 
 
 
-
 # import streamlit as st
 # from PyPDF2 import PdfReader