scaffold test script, partway there for actual prediction evaluation

test.py

@@ -1,16 +1,30 @@
 import datasets
 import transformers
+import torch
 
 
 dataset = datasets.load_dataset("syntaxgym", "mvrr_mod")
 metric = datasets.load_metric("syntaxgym", "mvrr_mod")
 
+tokenizer = transformers.AutoTokenizer.from_pretrained("gpt2")
+# DEV
+tokenizer.pad_token = tokenizer.eos_token
+
 model = transformers.AutoModelForCausalLM.from_pretrained("gpt2")
+model.eval()
 
 
 for item in dataset["test"]:
-    # TODO evaluate surprisals
+    # TODO full preprocessing setup
+    tokenized = tokenizer(item["conditions"]["content"], return_tensors="pt",
+                          padding=True)
 
     print(item)
+    print(tokenized)
+
+    with torch.no_grad():
+        print(model(tokenized["input_ids"]))
 
+    # TODO eval
+    
     break