It was and not or 🙂

detect-pretrain-code-contamination/src/run.py

@@ -87,7 +87,7 @@ def sample_generation(sentence, model, tokenizer, args,data_name):
     input_ids = input_ids.to(model.device)
 
     output = None
-    if data_name != "cais/mmlu" or data_name != "gsm8k":
+    if data_name != "cais/mmlu" and data_name != "gsm8k":
         output = model.generate(input_ids, max_new_tokens=len(sentence.split())-half_sentence_index, min_new_tokens=1, num_return_sequences=args['num_z'], pad_token_id=tokenizer.eos_token_id, **args['generate_args'])
     else:
         output = model.generate(input_ids, max_new_tokens=(len(sentence.split())-half_sentence_index)/2, min_new_tokens=1, num_return_sequences=int(args['num_z']/2), pad_token_id=tokenizer.eos_token_id, **args['generate_args'])