mirror of
https://github.com/aljazceru/chatgpt-telegram-bot.git
synced 2025-12-21 06:35:02 +01:00
added stream support
This commit is contained in:
gilcu3
@@ -384,52 +384,177 @@ class OpenAIHelper:
|
|||||||
logging.exception(e)
|
logging.exception(e)
|
||||||
raise Exception(f"⚠️ _{localized_text('error', self.config['bot_language'])}._ ⚠️\n{str(e)}") from e
|
raise Exception(f"⚠️ _{localized_text('error', self.config['bot_language'])}._ ⚠️\n{str(e)}") from e
|
||||||
|
|
||||||
async def interpret_image(self, chat_id, fileobj, prompt=None):
|
@retry(
|
||||||
|
reraise=True,
|
||||||
|
retry=retry_if_exception_type(openai.RateLimitError),
|
||||||
|
wait=wait_fixed(20),
|
||||||
|
stop=stop_after_attempt(3)
|
||||||
|
)
|
||||||
|
async def __common_get_chat_response_vision(self, chat_id: int, content: list, stream=False):
|
||||||
"""
|
"""
|
||||||
Interprets a given PNG image file using the Vision model.
|
Request a response from the GPT model.
|
||||||
|
:param chat_id: The chat ID
|
||||||
|
:param query: The query to send to the model
|
||||||
|
:return: The answer from the model and the number of tokens used
|
||||||
"""
|
"""
|
||||||
|
bot_language = self.config['bot_language']
|
||||||
try:
|
try:
|
||||||
image = encode_image(fileobj)
|
|
||||||
prompt = self.config['vision_prompt'] if prompt is None else prompt
|
|
||||||
|
|
||||||
# for now I am not adding the image itself to the history
|
|
||||||
|
|
||||||
if chat_id not in self.conversations or self.__max_age_reached(chat_id):
|
if chat_id not in self.conversations or self.__max_age_reached(chat_id):
|
||||||
self.reset_chat_history(chat_id)
|
self.reset_chat_history(chat_id)
|
||||||
|
|
||||||
self.last_updated[chat_id] = datetime.datetime.now()
|
self.last_updated[chat_id] = datetime.datetime.now()
|
||||||
|
|
||||||
|
for message in content:
|
||||||
|
if message['type'] == 'text':
|
||||||
|
query = message['text']
|
||||||
|
break
|
||||||
|
|
||||||
|
self.__add_to_history(chat_id, role="user", content=query)
|
||||||
|
|
||||||
|
# Summarize the chat history if it's too long to avoid excessive token usage
|
||||||
|
token_count = self.__count_tokens(self.conversations[chat_id])
|
||||||
|
exceeded_max_tokens = token_count + self.config['max_tokens'] > self.__max_model_tokens()
|
||||||
|
exceeded_max_history_size = len(self.conversations[chat_id]) > self.config['max_history_size']
|
||||||
|
|
||||||
|
if exceeded_max_tokens or exceeded_max_history_size:
|
||||||
|
logging.info(f'Chat history for chat ID {chat_id} is too long. Summarising...')
|
||||||
|
try:
|
||||||
|
summary = await self.__summarise(self.conversations[chat_id][:-1])
|
||||||
|
logging.debug(f'Summary: {summary}')
|
||||||
|
self.reset_chat_history(chat_id, self.conversations[chat_id][0]['content'])
|
||||||
|
self.__add_to_history(chat_id, role="assistant", content=summary)
|
||||||
|
self.__add_to_history(chat_id, role="user", content=query)
|
||||||
|
except Exception as e:
|
||||||
|
logging.warning(f'Error while summarising chat history: {str(e)}. Popping elements instead...')
|
||||||
|
self.conversations[chat_id] = self.conversations[chat_id][-self.config['max_history_size']:]
|
||||||
|
|
||||||
|
message = {'role':'user', 'content':content}
|
||||||
|
|
||||||
message = {'role':'user', 'content':[{'type':'text', 'text':prompt}, {'type':'image_url', \
|
|
||||||
'image_url': {'url':f'data:image/jpeg;base64,{image}', 'detail':self.config['vision_detail'] } }]}
|
|
||||||
common_args = {
|
common_args = {
|
||||||
'model': 'gpt-4-vision-preview', # the only one that currently makes sense here
|
'model': 'gpt-4-vision-preview', # the only one that currently makes sense here
|
||||||
'messages': self.conversations[chat_id] + [message],
|
'messages': self.conversations[chat_id][:-1] + [message],
|
||||||
'temperature': self.config['temperature'],
|
'temperature': self.config['temperature'],
|
||||||
'n': 1, # several choices is not implemented yet
|
'n': 1, # several choices is not implemented yet
|
||||||
'max_tokens': self.config['vision_max_tokens'],
|
'max_tokens': self.config['vision_max_tokens'],
|
||||||
'presence_penalty': self.config['presence_penalty'],
|
'presence_penalty': self.config['presence_penalty'],
|
||||||
'frequency_penalty': self.config['frequency_penalty'],
|
'frequency_penalty': self.config['frequency_penalty'],
|
||||||
'stream': False # We need to refactor this class to make this feasible without too much repetition
|
'stream': stream
|
||||||
}
|
}
|
||||||
self.__add_to_history(chat_id, role="user", content=prompt)
|
|
||||||
|
|
||||||
response = await self.client.chat.completions.create(**common_args)
|
|
||||||
|
|
||||||
|
|
||||||
content = response.choices[0].message.content
|
# vision model does not yet support functions
|
||||||
self.__add_to_history(chat_id, role="assistant", content=content)
|
|
||||||
|
|
||||||
|
# if self.config['enable_functions']:
|
||||||
|
# functions = self.plugin_manager.get_functions_specs()
|
||||||
|
# if len(functions) > 0:
|
||||||
|
# common_args['functions'] = self.plugin_manager.get_functions_specs()
|
||||||
|
# common_args['function_call'] = 'auto'
|
||||||
|
|
||||||
return content, response.usage.total_tokens
|
return await self.client.chat.completions.create(**common_args)
|
||||||
|
|
||||||
except openai.RateLimitError as e:
|
except openai.RateLimitError as e:
|
||||||
raise e
|
raise e
|
||||||
|
|
||||||
except openai.BadRequestError as e:
|
except openai.BadRequestError as e:
|
||||||
raise Exception(f"⚠️ _{localized_text('openai_invalid', self.config['bot_language'])}._ ⚠️\n{str(e)}") from e
|
raise Exception(f"⚠️ _{localized_text('openai_invalid', bot_language)}._ ⚠️\n{str(e)}") from e
|
||||||
|
|
||||||
except Exception as e:
|
except Exception as e:
|
||||||
logging.exception(e)
|
raise Exception(f"⚠️ _{localized_text('error', bot_language)}._ ⚠️\n{str(e)}") from e
|
||||||
raise Exception(f"⚠️ _{localized_text('error', self.config['bot_language'])}._ ⚠️\n{str(e)}") from e
|
|
||||||
|
|
||||||
|
async def interpret_image(self, chat_id, fileobj, prompt=None):
|
||||||
|
"""
|
||||||
|
Interprets a given PNG image file using the Vision model.
|
||||||
|
"""
|
||||||
|
image = encode_image(fileobj)
|
||||||
|
prompt = self.config['vision_prompt'] if prompt is None else prompt
|
||||||
|
|
||||||
|
content = [{'type':'text', 'text':prompt}, {'type':'image_url', \
|
||||||
|
'image_url': {'url':f'data:image/jpeg;base64,{image}', 'detail':self.config['vision_detail'] } }]
|
||||||
|
|
||||||
|
response = await self.__common_get_chat_response_vision(chat_id, content)
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
# functions are not available for this model
|
||||||
|
|
||||||
|
# if self.config['enable_functions']:
|
||||||
|
# response, plugins_used = await self.__handle_function_call(chat_id, response)
|
||||||
|
# if is_direct_result(response):
|
||||||
|
# return response, '0'
|
||||||
|
|
||||||
|
answer = ''
|
||||||
|
|
||||||
|
if len(response.choices) > 1 and self.config['n_choices'] > 1:
|
||||||
|
for index, choice in enumerate(response.choices):
|
||||||
|
content = choice.message.content.strip()
|
||||||
|
if index == 0:
|
||||||
|
self.__add_to_history(chat_id, role="assistant", content=content)
|
||||||
|
answer += f'{index + 1}\u20e3\n'
|
||||||
|
answer += content
|
||||||
|
answer += '\n\n'
|
||||||
|
else:
|
||||||
|
answer = response.choices[0].message.content.strip()
|
||||||
|
self.__add_to_history(chat_id, role="assistant", content=answer)
|
||||||
|
|
||||||
|
bot_language = self.config['bot_language']
|
||||||
|
# Plugins are not enabled either
|
||||||
|
# show_plugins_used = len(plugins_used) > 0 and self.config['show_plugins_used']
|
||||||
|
# plugin_names = tuple(self.plugin_manager.get_plugin_source_name(plugin) for plugin in plugins_used)
|
||||||
|
if self.config['show_usage']:
|
||||||
|
answer += "\n\n---\n" \
|
||||||
|
f"💰 {str(response.usage.total_tokens)} {localized_text('stats_tokens', bot_language)}" \
|
||||||
|
f" ({str(response.usage.prompt_tokens)} {localized_text('prompt', bot_language)}," \
|
||||||
|
f" {str(response.usage.completion_tokens)} {localized_text('completion', bot_language)})"
|
||||||
|
# if show_plugins_used:
|
||||||
|
# answer += f"\n🔌 {', '.join(plugin_names)}"
|
||||||
|
# elif show_plugins_used:
|
||||||
|
# answer += f"\n\n---\n🔌 {', '.join(plugin_names)}"
|
||||||
|
|
||||||
|
return answer, response.usage.total_tokens
|
||||||
|
|
||||||
|
async def interpret_image_stream(self, chat_id, fileobj, prompt=None):
|
||||||
|
"""
|
||||||
|
Interprets a given PNG image file using the Vision model.
|
||||||
|
"""
|
||||||
|
image = encode_image(fileobj)
|
||||||
|
prompt = self.config['vision_prompt'] if prompt is None else prompt
|
||||||
|
|
||||||
|
content = [{'type':'text', 'text':prompt}, {'type':'image_url', \
|
||||||
|
'image_url': {'url':f'data:image/jpeg;base64,{image}', 'detail':self.config['vision_detail'] } }]
|
||||||
|
|
||||||
|
response = await self.__common_get_chat_response_vision(chat_id, content, stream=True)
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
|
# if self.config['enable_functions']:
|
||||||
|
# response, plugins_used = await self.__handle_function_call(chat_id, response, stream=True)
|
||||||
|
# if is_direct_result(response):
|
||||||
|
# yield response, '0'
|
||||||
|
# return
|
||||||
|
|
||||||
|
answer = ''
|
||||||
|
async for chunk in response:
|
||||||
|
if len(chunk.choices) == 0:
|
||||||
|
continue
|
||||||
|
delta = chunk.choices[0].delta
|
||||||
|
if delta.content:
|
||||||
|
answer += delta.content
|
||||||
|
yield answer, 'not_finished'
|
||||||
|
answer = answer.strip()
|
||||||
|
self.__add_to_history(chat_id, role="assistant", content=answer)
|
||||||
|
tokens_used = str(self.__count_tokens(self.conversations[chat_id]))
|
||||||
|
|
||||||
|
#show_plugins_used = len(plugins_used) > 0 and self.config['show_plugins_used']
|
||||||
|
#plugin_names = tuple(self.plugin_manager.get_plugin_source_name(plugin) for plugin in plugins_used)
|
||||||
|
if self.config['show_usage']:
|
||||||
|
answer += f"\n\n---\n💰 {tokens_used} {localized_text('stats_tokens', self.config['bot_language'])}"
|
||||||
|
# if show_plugins_used:
|
||||||
|
# answer += f"\n🔌 {', '.join(plugin_names)}"
|
||||||
|
# elif show_plugins_used:
|
||||||
|
# answer += f"\n\n---\n🔌 {', '.join(plugin_names)}"
|
||||||
|
|
||||||
|
yield answer, tokens_used
|
||||||
|
|
||||||
def reset_chat_history(self, chat_id, content=''):
|
def reset_chat_history(self, chat_id, content=''):
|
||||||
"""
|
"""
|
||||||
|
|||||||
@@ -518,46 +518,128 @@ class ChatGPTTelegramBot:
|
|||||||
if user_id not in self.usage:
|
if user_id not in self.usage:
|
||||||
self.usage[user_id] = UsageTracker(user_id, update.message.from_user.name)
|
self.usage[user_id] = UsageTracker(user_id, update.message.from_user.name)
|
||||||
|
|
||||||
try:
|
if self.config['stream']:
|
||||||
interpretation, tokens = await self.openai.interpret_image(chat_id, temp_file_png, prompt=prompt)
|
|
||||||
|
|
||||||
vision_token_price = self.config['vision_token_price']
|
stream_response = self.openai.interpret_image_stream(chat_id=chat_id, fileobj=temp_file_png, prompt=prompt)
|
||||||
self.usage[user_id].add_vision_tokens(tokens, vision_token_price)
|
i = 0
|
||||||
|
prev = ''
|
||||||
|
sent_message = None
|
||||||
|
backoff = 0
|
||||||
|
stream_chunk = 0
|
||||||
|
|
||||||
allowed_user_ids = self.config['allowed_user_ids'].split(',')
|
async for content, tokens in stream_response:
|
||||||
if str(user_id) not in allowed_user_ids and 'guests' in self.usage:
|
if is_direct_result(content):
|
||||||
self.usage["guests"].add_vision_tokens(tokens, vision_token_price)
|
return await handle_direct_result(self.config, update, content)
|
||||||
|
|
||||||
|
if len(content.strip()) == 0:
|
||||||
|
continue
|
||||||
|
|
||||||
|
stream_chunks = split_into_chunks(content)
|
||||||
|
if len(stream_chunks) > 1:
|
||||||
|
content = stream_chunks[-1]
|
||||||
|
if stream_chunk != len(stream_chunks) - 1:
|
||||||
|
stream_chunk += 1
|
||||||
|
try:
|
||||||
|
await edit_message_with_retry(context, chat_id, str(sent_message.message_id),
|
||||||
|
stream_chunks[-2])
|
||||||
|
except:
|
||||||
|
pass
|
||||||
|
try:
|
||||||
|
sent_message = await update.effective_message.reply_text(
|
||||||
|
message_thread_id=get_thread_id(update),
|
||||||
|
text=content if len(content) > 0 else "..."
|
||||||
|
)
|
||||||
|
except:
|
||||||
|
pass
|
||||||
|
continue
|
||||||
|
|
||||||
|
cutoff = get_stream_cutoff_values(update, content)
|
||||||
|
cutoff += backoff
|
||||||
|
|
||||||
|
if i == 0:
|
||||||
|
try:
|
||||||
|
if sent_message is not None:
|
||||||
|
await context.bot.delete_message(chat_id=sent_message.chat_id,
|
||||||
|
message_id=sent_message.message_id)
|
||||||
|
sent_message = await update.effective_message.reply_text(
|
||||||
|
message_thread_id=get_thread_id(update),
|
||||||
|
reply_to_message_id=get_reply_to_message_id(self.config, update),
|
||||||
|
text=content,
|
||||||
|
)
|
||||||
|
except:
|
||||||
|
continue
|
||||||
|
|
||||||
|
elif abs(len(content) - len(prev)) > cutoff or tokens != 'not_finished':
|
||||||
|
prev = content
|
||||||
|
|
||||||
|
try:
|
||||||
|
use_markdown = tokens != 'not_finished'
|
||||||
|
await edit_message_with_retry(context, chat_id, str(sent_message.message_id),
|
||||||
|
text=content, markdown=use_markdown)
|
||||||
|
|
||||||
|
except RetryAfter as e:
|
||||||
|
backoff += 5
|
||||||
|
await asyncio.sleep(e.retry_after)
|
||||||
|
continue
|
||||||
|
|
||||||
|
except TimedOut:
|
||||||
|
backoff += 5
|
||||||
|
await asyncio.sleep(0.5)
|
||||||
|
continue
|
||||||
|
|
||||||
|
except Exception:
|
||||||
|
backoff += 5
|
||||||
|
continue
|
||||||
|
|
||||||
|
await asyncio.sleep(0.01)
|
||||||
|
|
||||||
|
i += 1
|
||||||
|
if tokens != 'not_finished':
|
||||||
|
total_tokens = int(tokens)
|
||||||
|
|
||||||
|
|
||||||
|
else:
|
||||||
|
|
||||||
try:
|
try:
|
||||||
await update.effective_message.reply_text(
|
interpretation, total_tokens = await self.openai.interpret_image(chat_id, temp_file_png, prompt=prompt)
|
||||||
message_thread_id=get_thread_id(update),
|
|
||||||
reply_to_message_id=get_reply_to_message_id(self.config, update),
|
|
||||||
text=interpretation,
|
|
||||||
parse_mode=constants.ParseMode.MARKDOWN
|
|
||||||
)
|
|
||||||
except BadRequest:
|
|
||||||
try:
|
try:
|
||||||
await update.effective_message.reply_text(
|
await update.effective_message.reply_text(
|
||||||
message_thread_id=get_thread_id(update),
|
message_thread_id=get_thread_id(update),
|
||||||
reply_to_message_id=get_reply_to_message_id(self.config, update),
|
reply_to_message_id=get_reply_to_message_id(self.config, update),
|
||||||
text=interpretation
|
text=interpretation,
|
||||||
)
|
|
||||||
except Exception as e:
|
|
||||||
logging.exception(e)
|
|
||||||
await update.effective_message.reply_text(
|
|
||||||
message_thread_id=get_thread_id(update),
|
|
||||||
reply_to_message_id=get_reply_to_message_id(self.config, update),
|
|
||||||
text=f"{localized_text('vision_fail', bot_language)}: {str(e)}",
|
|
||||||
parse_mode=constants.ParseMode.MARKDOWN
|
parse_mode=constants.ParseMode.MARKDOWN
|
||||||
)
|
)
|
||||||
except Exception as e:
|
except BadRequest:
|
||||||
logging.exception(e)
|
try:
|
||||||
await update.effective_message.reply_text(
|
await update.effective_message.reply_text(
|
||||||
message_thread_id=get_thread_id(update),
|
message_thread_id=get_thread_id(update),
|
||||||
reply_to_message_id=get_reply_to_message_id(self.config, update),
|
reply_to_message_id=get_reply_to_message_id(self.config, update),
|
||||||
text=f"{localized_text('vision_fail', bot_language)}: {str(e)}",
|
text=interpretation
|
||||||
parse_mode=constants.ParseMode.MARKDOWN
|
)
|
||||||
)
|
except Exception as e:
|
||||||
|
logging.exception(e)
|
||||||
|
await update.effective_message.reply_text(
|
||||||
|
message_thread_id=get_thread_id(update),
|
||||||
|
reply_to_message_id=get_reply_to_message_id(self.config, update),
|
||||||
|
text=f"{localized_text('vision_fail', bot_language)}: {str(e)}",
|
||||||
|
parse_mode=constants.ParseMode.MARKDOWN
|
||||||
|
)
|
||||||
|
except Exception as e:
|
||||||
|
logging.exception(e)
|
||||||
|
await update.effective_message.reply_text(
|
||||||
|
message_thread_id=get_thread_id(update),
|
||||||
|
reply_to_message_id=get_reply_to_message_id(self.config, update),
|
||||||
|
text=f"{localized_text('vision_fail', bot_language)}: {str(e)}",
|
||||||
|
parse_mode=constants.ParseMode.MARKDOWN
|
||||||
|
)
|
||||||
|
vision_token_price = self.config['vision_token_price']
|
||||||
|
self.usage[user_id].add_vision_tokens(total_tokens, vision_token_price)
|
||||||
|
|
||||||
|
allowed_user_ids = self.config['allowed_user_ids'].split(',')
|
||||||
|
if str(user_id) not in allowed_user_ids and 'guests' in self.usage:
|
||||||
|
self.usage["guests"].add_vision_tokens(total_tokens, vision_token_price)
|
||||||
|
|
||||||
await wrap_with_indicator(update, context, _execute, constants.ChatAction.TYPING)
|
await wrap_with_indicator(update, context, _execute, constants.ChatAction.TYPING)
|
||||||
|
|
||||||
|
|||||||
Reference in New Issue
Block a user