in tools/aoai.py [0:0]
def get_completion(self, prompt, image_base64=None, max_tokens=800, retry_after=True):
"""
Generates a completion for the given prompt using the Azure OpenAI service.
Args:
prompt (str): The input prompt for the model.
image_base64 (str, optional): Base64 encoded image to be included with the prompt. Defaults to None.
max_tokens (int, optional): The maximum number of tokens to generate. Defaults to 800.
retry_after (bool, optional): Flag to determine if the method should retry after rate limiting. Defaults to True.
Returns:
str: The generated completion.
"""
one_liner_prompt = prompt.replace('\n', ' ')
logging.debug(f"[aoai]{self.document_filename} Getting completion for prompt: {one_liner_prompt[:100]}")
# Truncate prompt if needed
prompt = self._truncate_input(prompt, self.max_gpt_model_input_tokens)
try:
input_messages = [
{"role": "system", "content": "You are a helpful assistant."},
]
if not image_base64:
input_messages.append({"role": "user", "content": prompt})
else:
input_messages.append({"role": "user", "content": [
{
"type": "text",
"text": prompt
},
{
"type": "image_url",
"image_url": {
"url":f"data:image/jpeg;base64,{image_base64}"
}
}
]})
response = self.client.chat.completions.create(
messages=input_messages,
model=self.openai_gpt_deployment,
temperature=0.7,
top_p=0.95,
max_tokens=max_tokens
)
completion = response.choices[0].message.content
logging.debug(f"[aoai]{self.document_filename} Completion received successfully.")
return completion
except RateLimitError as e:
if not retry_after:
logging.error(f"[aoai]{self.document_filename} get_completion: Rate limit error occurred after retries: {e}")
raise
retry_after_ms = e.response.headers.get('retry-after-ms')
if retry_after_ms:
retry_after_ms = int(retry_after_ms)
logging.info(f"[aoai]{self.document_filename} get_completion: Reached rate limit, retrying after {retry_after_ms} ms")
time.sleep(retry_after_ms / 1000)
return self.get_completion(prompt, max_tokens=max_tokens, retry_after=False)
else:
logging.error(f"[aoai]{self.document_filename} get_completion: Rate limit error occurred, no 'retry-after-ms' provided: {e}")
raise
except ClientAuthenticationError as e:
logging.error(f"[aoai]{self.document_filename} get_completion: Authentication failed: {e}")
raise
except Exception as e:
logging.error(f"[aoai]{self.document_filename} get_completion: An unexpected error occurred: {e}")
raise