Update app.py

app.py

@@ -61,10 +61,11 @@ presets = {
 }
 
 
-def custom_lp_logits_processor(ids, logits, lp_start, lp_decay):
-    if (len(ids) > lp_start):
+def custom_lp_logits_processor(ids, logits, lp_start, lp_decay, prompt_tok_len):
+    generated_tok_number = len(ids) - prompt_tok_len
+    if (generated_tok_number > lp_start):
         print(len(ids), lp_start, pow(lp_decay, len(ids)-lp_start)) 
-        logits[tokenizer.eos_token_id] *= pow(lp_decay, len(ids)-lp_start)
+        logits[tokenizer.eos_token_id] *= pow(lp_decay, generated_tok_number-lp_start)
     return logits
 
 def respond(
@@ -105,7 +106,7 @@ def respond(
         max_tokens=128,
         frequency_penalty=frequency_penalty,
         presence_penalty=presence_penalty,
-        logits_processor=lambda ids, logits: custom_lp_logits_processor(ids, logits, lp_start, lp_decay)
+        logits_processor=lambda ids, logits: custom_lp_logits_processor(ids, logits, lp_start, lp_decay, len(convo))
     ):
         token = message["choices"][0]["text"]