From fda27336008bd7920bd2673e245c4e87045bc1bd Mon Sep 17 00:00:00 2001
From: Migatu <mtuszowski@gmail.com>
Date: Fri, 20 Oct 2023 18:45:03 +0200
Subject: [PATCH] Token amount change

Last (hopefully) fix

Fix+1

Fix

Fix

Fix

Test

Fix tokens

Fix
---
 bot.py | 56 ++++++++++++++++++++++++++++++++++++--------------------
 1 file changed, 36 insertions(+), 20 deletions(-)

diff --git a/bot.py b/bot.py
index eee7b35..b63bf46 100644
--- a/bot.py
+++ b/bot.py
@@ -335,16 +335,27 @@ async def on_message(message):
 
 # *=========================================== Define Functions
 
-def num_tokens_from_string(string, model):
+def num_tokens_from_string(message, model):
     """
-    The function `num_tokens_from_string` returns the number of tokens in a given string.
+    The function takes a string message and a model as input and returns the number of tokens in the
+    message according to the given model.
 
-    :param string: The `string` parameter is a string of text that you want to count the number of
-    tokens for
-    :return: the number of tokens in the given string.
+    :param message: A string containing the message or text from which you want to count the number of
+    tokens
+    :param model: The model parameter refers to a language model or tokenizer that can be used to
+    tokenize the input string. It could be a pre-trained model or a custom tokenizer
     """
+    tokens_per_message = 3
+    tokens_per_name = 1
     chat_gpt_encoding = tiktoken.encoding_for_model(model)
-    num_tokens = len(chat_gpt_encoding.encode(string))
+
+    num_tokens = 0
+    num_tokens += tokens_per_message
+    for keys, values in message.items():
+        num_tokens += len(chat_gpt_encoding.encode(values))
+        if keys == "role":
+            num_tokens += tokens_per_name
+    num_tokens += 3  # every reply is primed with <|start|>assistant<|message|>
     return num_tokens
 
 
@@ -480,7 +491,6 @@ async def play(ctx, zamawial=None, arg=None):
 async def check():
     """Funkcja sprawdzająca czy grać następny kawałek."""
     while True:
-        # TODO: Tu dołożyć skanowanie rozmiaru pliku logowania oraz plików pamięci i przenoszenie ich na większy dysk.
         # trunk-ignore(codespell/misspelled)
         # TODO: dlaczego sie tu wypierdala
         if client.voice_clients:
@@ -548,6 +558,7 @@ async def check():
         await asyncio.sleep(60)
 
 
+# trunk-ignore(pylint/R0914)
 # trunk-ignore(pylint/R0913)
 # trunk-ignore(pylint/R0915)
 async def handle_response(prompt, vykidailo, bartender, history, username, music):
@@ -593,43 +604,48 @@ async def handle_response(prompt, vykidailo, bartender, history, username, music
             json.dump(file_data, file_memory, indent=4)
     history = []
     history.append(GPT_SETTINGS[0])
-    tokens = num_tokens_from_string(GPT_SETTINGS[0]["content"], "gpt-4")
+    chat_gpt_config_request_size = num_tokens_from_string(GPT_SETTINGS[0], "gpt-4")
+
     for (slowo, reakcja) in word_reactions.items():
         if not reakcja[3]:
             content = "Kiedy słyszysz " + slowo + " to reagujesz lub dzieje się to " + reakcja[0]
             temp = {"role": "system", "content": content}
-            tokens += num_tokens_from_string(content, "gpt-4")
+            chat_gpt_config_request_size += num_tokens_from_string(temp, "gpt-4")
             history.append(temp)
 
     final_prompt = username + ":" + prompt
-    logger.error("Liczba tokenów przed dodaniem historii %s", tokens)
+    logger.info("Rozmiar zapytania przed dodaniem historii %s", chat_gpt_config_request_size)
     if music:
         algorithm = "gpt-3.5-turbo-16k"
         table = message_table_muzyka
-        token_amount = 16000
+        token_amount = 15700
     else:
         table = MESSAGE_TABLE
         algorithm = "gpt-4"
-        token_amount = 80000
+        token_amount = 7700
 
-    prompt_tokens = num_tokens_from_string(final_prompt, "gpt-4")
+    prompt_gpt_request_size = num_tokens_from_string({"role": "user", "content": final_prompt}, "gpt-4")
+    temptable = []
     for i in reversed(table):
-        temp_token = num_tokens_from_string(i["content"], "gpt-4")
-        if tokens < token_amount + prompt_tokens + temp_token:
-            history.append(i)
-            tokens += temp_token
-
+        temp_token = num_tokens_from_string(i, "gpt-4")
+        logger.info("Rozmiar zapytania %s prompt %s temp %s", chat_gpt_config_request_size, prompt_gpt_request_size, temp_token)
+        if chat_gpt_config_request_size < token_amount + prompt_gpt_request_size + temp_token:
+            temptable.insert(1, i)
+            chat_gpt_config_request_size += temp_token
+    history.extend(temptable)
     temp = {"role": "user", "content": final_prompt}
     history.append(temp)
+    logger.info("Rozmiar zapytania po wyslaniu %s", chat_gpt_config_request_size)
+
     response = await openai.ChatCompletion.acreate(
             model=algorithm, messages=history
         )
-    logger.error("Liczba tokenów przed dodaniem historii %s", tokens)
     logger.info("Historia wysłana:")
     logger.info(history)
-    await asyncio.sleep(10)
+    await asyncio.sleep(15)
     result = ""
     logger.debug("Odpowiedzi")
+    logger.info(response)
     logger.debug(response.choices)
     for choice in response.choices:
         result += choice.message.content