f

app.py

@@ -13,10 +13,20 @@ print("PEFT Base Model:", peft_config.base_model_name_or_path)
 
 # 2. Load the tokenizer & base model
 tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL, trust_remote_code=True)
-base_model = AutoModelForCausalLM.from_pretrained(BASE_MODEL, revision="4831ee1375be5b4ff5a4abf7984e13628db44e35", ignore_mismatched_sizes=True, trust_remote_code=True)
+base_model = AutoModelForCausalLM.from_pretrained(
+    BASE_MODEL,
+    revision="4831ee1375be5b4ff5a4abf7984e13628db44e35",
+    ignore_mismatched_sizes=True,
+    trust_remote_code=True,
+    device_map="auto",
+)
 
 # 3. Load your LoRA adapter weights onto the base model
-model = PeftModel.from_pretrained(base_model, ADAPTER_REPO)
+model = PeftModel.from_pretrained(
+    base_model,
+    ADAPTER_REPO,
+    ignore_mismatched_sizes=True,  # Add this parameter
+)
 
 def classify_text(text):
     """