fix: use base model for demo (remove LoRA adapter dependency)

requirements.txt

@@ -1,6 +1,5 @@
 torch>=2.0.0
 transformers>=4.30.0
-peft>=0.4.0
 jsonschema>=4.0.0
 datasets>=2.0.0
 gradio>=5.0.0

test_constrained_model.py

@@ -9,13 +9,13 @@ import torch
 import json
 import jsonschema
 from transformers import AutoTokenizer, AutoModelForCausalLM
-from peft import PeftModel
+# from peft import PeftModel  # Not needed for base model demo
 from typing import Dict, List
 import time
 
 def load_trained_model():
     """Load our intensively trained model."""
-    print("🔄 Loading intensively trained SmolLM3-3B...")
+    print("🔄 Loading SmolLM3-3B (base model for demo)...")
     
     # Load base model
     base_model_name = "HuggingFaceTB/SmolLM3-3B"
@@ -29,10 +29,9 @@ def load_trained_model():
         device_map="mps" if torch.backends.mps.is_available() else "auto"
     )
     
-    # Load LoRA weights
-    print("🔧 Loading LoRA adapter...")
-    model = PeftModel.from_pretrained(model, "./smollm3_robust")
-    model = model.merge_and_unload()  # Merge for faster inference
+    # Note: Using base model for demo (LoRA adapter not included to keep repo size small)
+    print("🔧 Using base model (LoRA adapter excluded for size constraints)...")
+    # For production deployment, upload LoRA adapter to HF Hub and load from there
     
     print("✅ Trained model loaded successfully")
     return model, tokenizer