Update app.py

app.py

@@ -32,7 +32,7 @@ from transformers import AutoTokenizer, AutoModelForCausalLM
 ###############################################################################
 # Configuration                                                                #
 ###############################################################################
-MODEL_NAME           = "meta-llama/Llama-3.1-8B-Instruct"  # full-precision model
+MODEL_NAME           = "meta-llama/Llama-3.1-8B-Instruct"  # gated, requires HF_READ_TOKEN
 PROMPTS_PATH         = "full_prompts.json"                 # 100 full prompts
 STATE_PATH           = "current_state.json"                # persistent Oracle state
 DATA_PATH            = "data.json"                         # JSONL log of guesses
@@ -78,13 +78,17 @@ def load_prompts() -> list[str]:
 ###############################################################################
 # Model loading                                                                #
 ###############################################################################
-log.info("Loading Llama-3.1-8B-Instruct (FP32 CPU-only)… this can take a while.")
+log.info("Loading Llama-3.1-8B-Instruct (FP32 CPU-only) using secret token…")
 
-tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
+tokenizer = AutoTokenizer.from_pretrained(
+    MODEL_NAME,
+    use_auth_token=os.environ.get("HF_READ_TOKEN")
+)
 model = AutoModelForCausalLM.from_pretrained(
     MODEL_NAME,
     torch_dtype=torch.float32,
-    device_map={"": "cpu"},  # force CPU placement
+    device_map={"": "cpu"},
+    use_auth_token=os.environ.get("HF_READ_TOKEN")
 )
 model.eval()
 log.info("Model ready – Oracle awakened.")