Upload app.py with huggingface_hub

app.py

@@ -21,21 +21,21 @@ if HF_TOKEN:
 
 # ─── Configuration ───
 MODEL_CONFIGS = {
-    "Qwen-0.5B (Recommended)": {
+    "Gemma-270M (Fastest)": {
+        "base": "google/gemma-3-270m-it",
+        "adapter": "tarun5986/MicroGuard-Gemma-270M",
+    },
+    "Qwen-0.5B (Balanced)": {
         "base": "Qwen/Qwen2.5-0.5B-Instruct",
         "adapter": "tarun5986/MicroGuard-Qwen-0.5B",
     },
-    "SmolLM-135M (Fastest, 135M params)": {
-        "base": "HuggingFaceTB/SmolLM2-135M-Instruct",
-        "adapter": "tarun5986/MicroGuard-SmolLM-135M",
-    },
-    "TinyLlama-1.1B (Largest)": {
-        "base": "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
-        "adapter": "tarun5986/MicroGuard-TinyLlama-1.1B",
+    "Gemma-1B (Best Accuracy)": {
+        "base": "google/gemma-3-1b-it",
+        "adapter": "tarun5986/MicroGuard-Gemma-1B",
     },
 }
 
-DEFAULT_MODEL = "Qwen-0.5B (Recommended)"
+DEFAULT_MODEL = "Qwen-0.5B (Balanced)"
 
 current_model = None
 current_tokenizer = None
@@ -63,14 +63,14 @@ def load_model(model_choice):
     config = MODEL_CONFIGS[model_choice]
 
     try:
-        tokenizer = AutoTokenizer.from_pretrained(config["base"], trust_remote_code=True)
+        tokenizer = AutoTokenizer.from_pretrained(config["base"], trust_remote_code=True, token=HF_TOKEN)
         if tokenizer.pad_token is None:
             tokenizer.pad_token = tokenizer.eos_token
 
         base_model = AutoModelForCausalLM.from_pretrained(
-            config["base"], torch_dtype=DTYPE, trust_remote_code=True
+            config["base"], torch_dtype=DTYPE, trust_remote_code=True, token=HF_TOKEN
         )
-        model = PeftModel.from_pretrained(base_model, config["adapter"])
+        model = PeftModel.from_pretrained(base_model, config["adapter"], token=HF_TOKEN)
         model = model.to(DEVICE)
         model.eval()