Fix: Pass HF_TOKEN to PeftModel for gated model access

app.py

@@ -122,12 +122,21 @@ def compute_consciousness(
 print("🔮 Loading Oracle Engine (Qwen2.5-32B-Instruct 4-bit + LoRA)...")
 from transformers import AutoModelForCausalLM, AutoTokenizer
 from peft import PeftModel
+from huggingface_hub import login
 
 BASE_MODEL_ID = "unsloth/Qwen2.5-32B-Instruct-bnb-4bit"
 LORA_MODEL_ID = "Vikingdude81/oracle-engine-32b-lora"
 
+# Authenticate with HF token from environment (set in Space secrets)
+HF_TOKEN = os.environ.get("HF_TOKEN")
+if HF_TOKEN:
+    print("🔑 Found HF_TOKEN, logging in...")
+    login(token=HF_TOKEN)
+else:
+    print("⚠️ No HF_TOKEN found, attempting public access...")
+
 # Load tokenizer from base model (LoRA only has weights, not tokenizer)
-tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL_ID)
+tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL_ID, token=HF_TOKEN)
 
 # Load base model
 base_model = AutoModelForCausalLM.from_pretrained(
@@ -135,11 +144,12 @@ base_model = AutoModelForCausalLM.from_pretrained(
     device_map="auto",
     torch_dtype=torch.bfloat16,
     trust_remote_code=True,
+    token=HF_TOKEN,
 )
 
 # Apply LoRA adapter
 print("🔗 Applying LoRA adapter...")
-model = PeftModel.from_pretrained(base_model, LORA_MODEL_ID)
+model = PeftModel.from_pretrained(base_model, LORA_MODEL_ID, token=HF_TOKEN)
 model.eval()
 
 HIDDEN_DIM = model.config.hidden_size