update generate

utils.py

@@ -182,6 +182,8 @@ def moderate(chat, model, tokenizer, UNSAFE_TOKEN_ID):
         pad_token_id=tokenizer.eos_token_id,
         output_logits=True,  # get logits
         do_sample=False,
+        temperature=None,
+        top_p=None
     )
     ######
     # Get generated text
@@ -192,7 +194,7 @@ def moderate(chat, model, tokenizer, UNSAFE_TOKEN_ID):
     # Ignore the tokens from the input to get the tokens generated by the model
     generated_token_ids = outputs.sequences[:, input_length:].cpu()
     generated_text = tokenizer.decode(generated_token_ids[0], skip_special_tokens=True)
-
+    logger.info(generated_text)
     ######
     # Get Probability of "unsafe" token
     ######
@@ -206,9 +208,9 @@ def moderate(chat, model, tokenizer, UNSAFE_TOKEN_ID):
     unsafe_probability = first_token_probs[0, UNSAFE_TOKEN_ID]
     unsafe_probability = unsafe_probability.item()
 
-    ######
+    ########
     # Result
-    ######
+    ########
     return {
         "unsafe_score": unsafe_probability,
         "generated_text": generated_text