Fastapi refactor update
This commit is contained in:
Carlos-Mesquita
@@ -1,13 +1,13 @@
|
||||
from .aws_polly import AWSPolly
|
||||
from .heygen import Heygen
|
||||
from .openai import OpenAI
|
||||
from .whisper import OpenAIWhisper
|
||||
from .gpt_zero import GPTZero
|
||||
|
||||
__all__ = [
|
||||
"AWSPolly",
|
||||
"Heygen",
|
||||
"OpenAI",
|
||||
"OpenAIWhisper",
|
||||
"GPTZero"
|
||||
]
|
||||
from .aws_polly import AWSPolly
|
||||
from .heygen import Heygen
|
||||
from .openai import OpenAI
|
||||
from .whisper import OpenAIWhisper
|
||||
from .gpt_zero import GPTZero
|
||||
|
||||
__all__ = [
|
||||
"AWSPolly",
|
||||
"Heygen",
|
||||
"OpenAI",
|
||||
"OpenAIWhisper",
|
||||
"GPTZero"
|
||||
]
|
||||
|
||||
@@ -1,87 +1,87 @@
|
||||
import random
|
||||
from typing import Union
|
||||
|
||||
import aiofiles
|
||||
from aiobotocore.client import BaseClient
|
||||
|
||||
from app.services.abc import ITextToSpeechService
|
||||
from app.configs.constants import NeuralVoices
|
||||
|
||||
|
||||
class AWSPolly(ITextToSpeechService):
|
||||
|
||||
def __init__(self, client: BaseClient):
|
||||
self._client = client
|
||||
|
||||
async def synthesize_speech(self, text: str, voice: str, engine: str = "neural", output_format: str = "mp3"):
|
||||
tts_response = await self._client.synthesize_speech(
|
||||
Engine=engine,
|
||||
Text=text,
|
||||
OutputFormat=output_format,
|
||||
VoiceId=voice
|
||||
)
|
||||
return await tts_response['AudioStream'].read()
|
||||
|
||||
async def text_to_speech(self, text: Union[list[str], str], file_name: str):
|
||||
if isinstance(text, str):
|
||||
audio_segments = await self._text_to_speech(text)
|
||||
elif isinstance(text, list):
|
||||
audio_segments = await self._conversation_to_speech(text)
|
||||
else:
|
||||
raise ValueError("Unsupported argument for text_to_speech")
|
||||
|
||||
final_message = await self.synthesize_speech(
|
||||
"This audio recording, for the listening exercise, has finished.",
|
||||
"Stephen"
|
||||
)
|
||||
|
||||
# Add finish message
|
||||
audio_segments.append(final_message)
|
||||
|
||||
# Combine the audio segments into a single audio file
|
||||
combined_audio = b"".join(audio_segments)
|
||||
# Save the combined audio to a single file
|
||||
async with aiofiles.open(file_name, "wb") as f:
|
||||
await f.write(combined_audio)
|
||||
|
||||
print("Speech segments saved to " + file_name)
|
||||
|
||||
async def _text_to_speech(self, text: str):
|
||||
voice = random.choice(NeuralVoices.ALL_NEURAL_VOICES)['Id']
|
||||
# Initialize an empty list to store audio segments
|
||||
audio_segments = []
|
||||
for part in self._divide_text(text):
|
||||
audio_segments.append(await self.synthesize_speech(part, voice))
|
||||
|
||||
return audio_segments
|
||||
|
||||
async def _conversation_to_speech(self, conversation: list):
|
||||
# Initialize an empty list to store audio segments
|
||||
audio_segments = []
|
||||
# Iterate through the text segments, convert to audio segments, and store them
|
||||
for segment in conversation:
|
||||
audio_segments.append(await self.synthesize_speech(segment["text"], segment["voice"]))
|
||||
|
||||
return audio_segments
|
||||
|
||||
@staticmethod
|
||||
def _divide_text(text, max_length=3000):
|
||||
if len(text) <= max_length:
|
||||
return [text]
|
||||
|
||||
divisions = []
|
||||
current_position = 0
|
||||
|
||||
while current_position < len(text):
|
||||
next_position = min(current_position + max_length, len(text))
|
||||
next_period_position = text.rfind('.', current_position, next_position)
|
||||
|
||||
if next_period_position != -1 and next_period_position > current_position:
|
||||
divisions.append(text[current_position:next_period_position + 1])
|
||||
current_position = next_period_position + 1
|
||||
else:
|
||||
# If no '.' found in the next chunk, split at max_length
|
||||
divisions.append(text[current_position:next_position])
|
||||
current_position = next_position
|
||||
|
||||
return divisions
|
||||
import random
|
||||
from typing import Union
|
||||
|
||||
import aiofiles
|
||||
from aiobotocore.client import BaseClient
|
||||
|
||||
from app.services.abc import ITextToSpeechService
|
||||
from app.configs.constants import NeuralVoices
|
||||
|
||||
|
||||
class AWSPolly(ITextToSpeechService):
|
||||
|
||||
def __init__(self, client: BaseClient):
|
||||
self._client = client
|
||||
|
||||
async def synthesize_speech(self, text: str, voice: str, engine: str = "neural", output_format: str = "mp3"):
|
||||
tts_response = await self._client.synthesize_speech(
|
||||
Engine=engine,
|
||||
Text=text,
|
||||
OutputFormat=output_format,
|
||||
VoiceId=voice
|
||||
)
|
||||
return await tts_response['AudioStream'].read()
|
||||
|
||||
async def text_to_speech(self, text: Union[list[str], str], file_name: str):
|
||||
if isinstance(text, str):
|
||||
audio_segments = await self._text_to_speech(text)
|
||||
elif isinstance(text, list):
|
||||
audio_segments = await self._conversation_to_speech(text)
|
||||
else:
|
||||
raise ValueError("Unsupported argument for text_to_speech")
|
||||
|
||||
final_message = await self.synthesize_speech(
|
||||
"This audio recording, for the listening exercise, has finished.",
|
||||
"Stephen"
|
||||
)
|
||||
|
||||
# Add finish message
|
||||
audio_segments.append(final_message)
|
||||
|
||||
# Combine the audio segments into a single audio file
|
||||
combined_audio = b"".join(audio_segments)
|
||||
# Save the combined audio to a single file
|
||||
async with aiofiles.open(file_name, "wb") as f:
|
||||
await f.write(combined_audio)
|
||||
|
||||
print("Speech segments saved to " + file_name)
|
||||
|
||||
async def _text_to_speech(self, text: str):
|
||||
voice = random.choice(NeuralVoices.ALL_NEURAL_VOICES)['Id']
|
||||
# Initialize an empty list to store audio segments
|
||||
audio_segments = []
|
||||
for part in self._divide_text(text):
|
||||
audio_segments.append(await self.synthesize_speech(part, voice))
|
||||
|
||||
return audio_segments
|
||||
|
||||
async def _conversation_to_speech(self, conversation: list):
|
||||
# Initialize an empty list to store audio segments
|
||||
audio_segments = []
|
||||
# Iterate through the text segments, convert to audio segments, and store them
|
||||
for segment in conversation:
|
||||
audio_segments.append(await self.synthesize_speech(segment["text"], segment["voice"]))
|
||||
|
||||
return audio_segments
|
||||
|
||||
@staticmethod
|
||||
def _divide_text(text, max_length=3000):
|
||||
if len(text) <= max_length:
|
||||
return [text]
|
||||
|
||||
divisions = []
|
||||
current_position = 0
|
||||
|
||||
while current_position < len(text):
|
||||
next_position = min(current_position + max_length, len(text))
|
||||
next_period_position = text.rfind('.', current_position, next_position)
|
||||
|
||||
if next_period_position != -1 and next_period_position > current_position:
|
||||
divisions.append(text[current_position:next_period_position + 1])
|
||||
current_position = next_period_position + 1
|
||||
else:
|
||||
# If no '.' found in the next chunk, split at max_length
|
||||
divisions.append(text[current_position:next_position])
|
||||
current_position = next_position
|
||||
|
||||
return divisions
|
||||
|
||||
@@ -1,52 +1,52 @@
|
||||
from logging import getLogger
|
||||
from typing import Dict, Optional
|
||||
|
||||
from httpx import AsyncClient
|
||||
|
||||
from app.services.abc.third_parties.ai_detector import IAIDetectorService
|
||||
|
||||
|
||||
class GPTZero(IAIDetectorService):
|
||||
|
||||
_GPT_ZERO_ENDPOINT = 'https://api.gptzero.me/v2/predict/text'
|
||||
|
||||
def __init__(self, client: AsyncClient, gpt_zero_key: str):
|
||||
self._header = {
|
||||
'x-api-key': gpt_zero_key
|
||||
}
|
||||
self._http_client = client
|
||||
self._logger = getLogger(__name__)
|
||||
|
||||
async def run_detection(self, text: str):
|
||||
data = {
|
||||
'document': text,
|
||||
'version': '',
|
||||
'multilingual': False
|
||||
}
|
||||
|
||||
response = await self._http_client.post(self._GPT_ZERO_ENDPOINT, headers=self._header, json=data)
|
||||
if response.status_code != 200:
|
||||
return None
|
||||
return self._parse_detection(response.json())
|
||||
|
||||
def _parse_detection(self, response: Dict) -> Optional[Dict]:
|
||||
try:
|
||||
text_scan = response["documents"][0]
|
||||
|
||||
filtered_sentences = [
|
||||
{
|
||||
"sentence": item["sentence"],
|
||||
"highlight_sentence_for_ai": item["highlight_sentence_for_ai"]
|
||||
}
|
||||
for item in text_scan["sentences"]
|
||||
]
|
||||
|
||||
return {
|
||||
"class_probabilities": text_scan["class_probabilities"],
|
||||
"confidence_category": text_scan["confidence_category"],
|
||||
"predicted_class": text_scan["predicted_class"],
|
||||
"sentences": filtered_sentences
|
||||
}
|
||||
except Exception as e:
|
||||
self._logger.error(f'Failed to parse GPT\'s Zero response: {str(e)}')
|
||||
return None
|
||||
from logging import getLogger
|
||||
from typing import Dict, Optional
|
||||
|
||||
from httpx import AsyncClient
|
||||
|
||||
from app.services.abc.third_parties.ai_detector import IAIDetectorService
|
||||
|
||||
|
||||
class GPTZero(IAIDetectorService):
|
||||
|
||||
_GPT_ZERO_ENDPOINT = 'https://api.gptzero.me/v2/predict/text'
|
||||
|
||||
def __init__(self, client: AsyncClient, gpt_zero_key: str):
|
||||
self._header = {
|
||||
'x-api-key': gpt_zero_key
|
||||
}
|
||||
self._http_client = client
|
||||
self._logger = getLogger(__name__)
|
||||
|
||||
async def run_detection(self, text: str):
|
||||
data = {
|
||||
'document': text,
|
||||
'version': '',
|
||||
'multilingual': False
|
||||
}
|
||||
|
||||
response = await self._http_client.post(self._GPT_ZERO_ENDPOINT, headers=self._header, json=data)
|
||||
if response.status_code != 200:
|
||||
return None
|
||||
return self._parse_detection(response.json())
|
||||
|
||||
def _parse_detection(self, response: Dict) -> Optional[Dict]:
|
||||
try:
|
||||
text_scan = response["documents"][0]
|
||||
|
||||
filtered_sentences = [
|
||||
{
|
||||
"sentence": item["sentence"],
|
||||
"highlight_sentence_for_ai": item["highlight_sentence_for_ai"]
|
||||
}
|
||||
for item in text_scan["sentences"]
|
||||
]
|
||||
|
||||
return {
|
||||
"class_probabilities": text_scan["class_probabilities"],
|
||||
"confidence_category": text_scan["confidence_category"],
|
||||
"predicted_class": text_scan["predicted_class"],
|
||||
"sentences": filtered_sentences
|
||||
}
|
||||
except Exception as e:
|
||||
self._logger.error(f'Failed to parse GPT\'s Zero response: {str(e)}')
|
||||
return None
|
||||
|
||||
@@ -1,90 +1,90 @@
|
||||
import asyncio
|
||||
import os
|
||||
import logging
|
||||
import aiofiles
|
||||
|
||||
from httpx import AsyncClient
|
||||
|
||||
from app.services.abc import IVideoGeneratorService
|
||||
|
||||
|
||||
class Heygen(IVideoGeneratorService):
|
||||
|
||||
# TODO: Not used, remove if not necessary
|
||||
# CREATE_VIDEO_URL = 'https://api.heygen.com/v1/template.generate'
|
||||
|
||||
_GET_VIDEO_URL = 'https://api.heygen.com/v1/video_status.get'
|
||||
|
||||
def __init__(self, client: AsyncClient, heygen_token: str):
|
||||
self._get_header = {
|
||||
'X-Api-Key': heygen_token
|
||||
}
|
||||
self._post_header = {
|
||||
'X-Api-Key': heygen_token,
|
||||
'Content-Type': 'application/json'
|
||||
}
|
||||
self._http_client = client
|
||||
self._logger = logging.getLogger(__name__)
|
||||
|
||||
async def create_video(self, text: str, avatar: str):
|
||||
# POST TO CREATE VIDEO
|
||||
create_video_url = 'https://api.heygen.com/v2/template/' + avatar + '/generate'
|
||||
data = {
|
||||
"test": False,
|
||||
"caption": False,
|
||||
"title": "video_title",
|
||||
"variables": {
|
||||
"script_here": {
|
||||
"name": "script_here",
|
||||
"type": "text",
|
||||
"properties": {
|
||||
"content": text
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
response = await self._http_client.post(create_video_url, headers=self._post_header, json=data)
|
||||
self._logger.info(response.status_code)
|
||||
self._logger.info(response.json())
|
||||
|
||||
# GET TO CHECK STATUS AND GET VIDEO WHEN READY
|
||||
video_id = response.json()["data"]["video_id"]
|
||||
params = {
|
||||
'video_id': response.json()["data"]["video_id"]
|
||||
}
|
||||
response = {}
|
||||
status = "processing"
|
||||
error = None
|
||||
|
||||
while status != "completed" and error is None:
|
||||
response = await self._http_client.get(self._GET_VIDEO_URL, headers=self._get_header, params=params)
|
||||
response_data = response.json()
|
||||
|
||||
status = response_data["data"]["status"]
|
||||
error = response_data["data"]["error"]
|
||||
|
||||
if status != "completed" and error is None:
|
||||
self._logger.info(f"Status: {status}")
|
||||
await asyncio.sleep(10) # Wait for 10 second before the next request
|
||||
|
||||
self._logger.info(response.status_code)
|
||||
self._logger.info(response.json())
|
||||
|
||||
# DOWNLOAD VIDEO
|
||||
download_url = response.json()['data']['video_url']
|
||||
output_directory = 'download-video/'
|
||||
output_filename = video_id + '.mp4'
|
||||
|
||||
response = await self._http_client.get(download_url)
|
||||
|
||||
if response.status_code == 200:
|
||||
os.makedirs(output_directory, exist_ok=True) # Create the directory if it doesn't exist
|
||||
output_path = os.path.join(output_directory, output_filename)
|
||||
async with aiofiles.open(output_path, 'wb') as f:
|
||||
await f.write(response.content)
|
||||
self._logger.info(f"File '{output_filename}' downloaded successfully.")
|
||||
return output_filename
|
||||
else:
|
||||
self._logger.error(f"Failed to download file. Status code: {response.status_code}")
|
||||
return None
|
||||
|
||||
import asyncio
|
||||
import os
|
||||
import logging
|
||||
import aiofiles
|
||||
|
||||
from httpx import AsyncClient
|
||||
|
||||
from app.services.abc import IVideoGeneratorService
|
||||
|
||||
|
||||
class Heygen(IVideoGeneratorService):
|
||||
|
||||
# TODO: Not used, remove if not necessary
|
||||
# CREATE_VIDEO_URL = 'https://api.heygen.com/v1/template.generate'
|
||||
|
||||
_GET_VIDEO_URL = 'https://api.heygen.com/v1/video_status.get'
|
||||
|
||||
def __init__(self, client: AsyncClient, heygen_token: str):
|
||||
self._get_header = {
|
||||
'X-Api-Key': heygen_token
|
||||
}
|
||||
self._post_header = {
|
||||
'X-Api-Key': heygen_token,
|
||||
'Content-Type': 'application/json'
|
||||
}
|
||||
self._http_client = client
|
||||
self._logger = logging.getLogger(__name__)
|
||||
|
||||
async def create_video(self, text: str, avatar: str):
|
||||
# POST TO CREATE VIDEO
|
||||
create_video_url = 'https://api.heygen.com/v2/template/' + avatar + '/generate'
|
||||
data = {
|
||||
"test": False,
|
||||
"caption": False,
|
||||
"title": "video_title",
|
||||
"variables": {
|
||||
"script_here": {
|
||||
"name": "script_here",
|
||||
"type": "text",
|
||||
"properties": {
|
||||
"content": text
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
response = await self._http_client.post(create_video_url, headers=self._post_header, json=data)
|
||||
self._logger.info(response.status_code)
|
||||
self._logger.info(response.json())
|
||||
|
||||
# GET TO CHECK STATUS AND GET VIDEO WHEN READY
|
||||
video_id = response.json()["data"]["video_id"]
|
||||
params = {
|
||||
'video_id': response.json()["data"]["video_id"]
|
||||
}
|
||||
response = {}
|
||||
status = "processing"
|
||||
error = None
|
||||
|
||||
while status != "completed" and error is None:
|
||||
response = await self._http_client.get(self._GET_VIDEO_URL, headers=self._get_header, params=params)
|
||||
response_data = response.json()
|
||||
|
||||
status = response_data["data"]["status"]
|
||||
error = response_data["data"]["error"]
|
||||
|
||||
if status != "completed" and error is None:
|
||||
self._logger.info(f"Status: {status}")
|
||||
await asyncio.sleep(10) # Wait for 10 second before the next request
|
||||
|
||||
self._logger.info(response.status_code)
|
||||
self._logger.info(response.json())
|
||||
|
||||
# DOWNLOAD VIDEO
|
||||
download_url = response.json()['data']['video_url']
|
||||
output_directory = 'download-video/'
|
||||
output_filename = video_id + '.mp4'
|
||||
|
||||
response = await self._http_client.get(download_url)
|
||||
|
||||
if response.status_code == 200:
|
||||
os.makedirs(output_directory, exist_ok=True) # Create the directory if it doesn't exist
|
||||
output_path = os.path.join(output_directory, output_filename)
|
||||
async with aiofiles.open(output_path, 'wb') as f:
|
||||
await f.write(response.content)
|
||||
self._logger.info(f"File '{output_filename}' downloaded successfully.")
|
||||
return output_filename
|
||||
else:
|
||||
self._logger.error(f"Failed to download file. Status code: {response.status_code}")
|
||||
return None
|
||||
|
||||
|
||||
@@ -1,150 +1,150 @@
|
||||
import json
|
||||
import re
|
||||
import logging
|
||||
from typing import List, Optional, Callable, TypeVar
|
||||
from openai import AsyncOpenAI
|
||||
from openai.types.chat import ChatCompletionMessageParam
|
||||
|
||||
from app.services.abc import ILLMService
|
||||
from app.helpers import count_tokens
|
||||
from app.configs.constants import BLACKLISTED_WORDS
|
||||
from pydantic import BaseModel
|
||||
|
||||
T = TypeVar('T', bound=BaseModel)
|
||||
|
||||
|
||||
class OpenAI(ILLMService):
|
||||
|
||||
MAX_TOKENS = 4097
|
||||
TRY_LIMIT = 2
|
||||
|
||||
def __init__(self, client: AsyncOpenAI):
|
||||
self._client = client
|
||||
self._logger = logging.getLogger(__name__)
|
||||
self._default_model = "gpt-4o-2024-08-06"
|
||||
|
||||
async def prediction(
|
||||
self,
|
||||
model: str,
|
||||
messages: List[ChatCompletionMessageParam],
|
||||
fields_to_check: Optional[List[str]],
|
||||
temperature: float,
|
||||
check_blacklisted: bool = True,
|
||||
token_count: int = -1
|
||||
):
|
||||
if token_count == -1:
|
||||
token_count = self._count_total_tokens(messages)
|
||||
return await self._prediction(model, messages, token_count, fields_to_check, temperature, 0, check_blacklisted)
|
||||
|
||||
async def _prediction(
|
||||
self,
|
||||
model: str,
|
||||
messages: List[ChatCompletionMessageParam],
|
||||
token_count: int,
|
||||
fields_to_check: Optional[List[str]],
|
||||
temperature: float,
|
||||
try_count: int,
|
||||
check_blacklisted: bool,
|
||||
):
|
||||
result = await self._client.chat.completions.create(
|
||||
model=model,
|
||||
max_tokens=int(self.MAX_TOKENS - token_count - 300),
|
||||
temperature=float(temperature),
|
||||
messages=messages,
|
||||
response_format={"type": "json_object"}
|
||||
)
|
||||
result = result.choices[0].message.content
|
||||
|
||||
if check_blacklisted:
|
||||
found_blacklisted_word = self._get_found_blacklisted_words(result)
|
||||
|
||||
if found_blacklisted_word is not None and try_count < self.TRY_LIMIT:
|
||||
self._logger.warning("Result contains blacklisted words: " + str(found_blacklisted_word))
|
||||
return await self._prediction(
|
||||
model, messages, token_count, fields_to_check, temperature, (try_count + 1), check_blacklisted
|
||||
)
|
||||
elif found_blacklisted_word is not None and try_count >= self.TRY_LIMIT:
|
||||
return ""
|
||||
|
||||
if fields_to_check is None:
|
||||
return json.loads(result)
|
||||
|
||||
if not self._check_fields(result, fields_to_check) and try_count < self.TRY_LIMIT:
|
||||
return await self._prediction(
|
||||
model, messages, token_count, fields_to_check, temperature, (try_count + 1), check_blacklisted
|
||||
)
|
||||
|
||||
return json.loads(result)
|
||||
|
||||
async def prediction_override(self, **kwargs):
|
||||
return await self._client.chat.completions.create(
|
||||
**kwargs
|
||||
)
|
||||
|
||||
@staticmethod
|
||||
def _get_found_blacklisted_words(text: str):
|
||||
text_lower = text.lower()
|
||||
for word in BLACKLISTED_WORDS:
|
||||
if re.search(r'\b' + re.escape(word) + r'\b', text_lower):
|
||||
return word
|
||||
return None
|
||||
|
||||
@staticmethod
|
||||
def _count_total_tokens(messages):
|
||||
total_tokens = 0
|
||||
for message in messages:
|
||||
total_tokens += count_tokens(message["content"])["n_tokens"]
|
||||
return total_tokens
|
||||
|
||||
@staticmethod
|
||||
def _check_fields(obj, fields):
|
||||
return all(field in obj for field in fields)
|
||||
|
||||
async def pydantic_prediction(
|
||||
self,
|
||||
messages: List[ChatCompletionMessageParam],
|
||||
map_to_model: Callable,
|
||||
json_scheme: str,
|
||||
*,
|
||||
model: Optional[str] = None,
|
||||
temperature: Optional[float] = None,
|
||||
max_retries: int = 3
|
||||
) -> List[T] | T | None:
|
||||
params = {
|
||||
"messages": messages,
|
||||
"response_format": {"type": "json_object"},
|
||||
"model": model if model else self._default_model
|
||||
}
|
||||
|
||||
if temperature:
|
||||
params["temperature"] = temperature
|
||||
|
||||
attempt = 0
|
||||
while attempt < max_retries:
|
||||
result = await self._client.chat.completions.create(**params)
|
||||
result_content = result.choices[0].message.content
|
||||
try:
|
||||
result_json = json.loads(result_content)
|
||||
return map_to_model(result_json)
|
||||
except Exception as e:
|
||||
attempt += 1
|
||||
self._logger.info(f"GPT returned malformed response: {result_content}\n {str(e)}")
|
||||
params["messages"] = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": (
|
||||
"Your previous response wasn't in the json format I've explicitly told you to output. "
|
||||
f"In your next response, you will fix it and return me just the json I've asked."
|
||||
)
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": (
|
||||
f"Previous response: {result_content}\n"
|
||||
f"JSON format: {json_scheme}"
|
||||
)
|
||||
}
|
||||
]
|
||||
if attempt >= max_retries:
|
||||
self._logger.error(f"Max retries exceeded!")
|
||||
return None
|
||||
import json
|
||||
import re
|
||||
import logging
|
||||
from typing import List, Optional, Callable, TypeVar
|
||||
from openai import AsyncOpenAI
|
||||
from openai.types.chat import ChatCompletionMessageParam
|
||||
|
||||
from app.services.abc import ILLMService
|
||||
from app.helpers import count_tokens
|
||||
from app.configs.constants import BLACKLISTED_WORDS
|
||||
from pydantic import BaseModel
|
||||
|
||||
T = TypeVar('T', bound=BaseModel)
|
||||
|
||||
|
||||
class OpenAI(ILLMService):
|
||||
|
||||
MAX_TOKENS = 4097
|
||||
TRY_LIMIT = 2
|
||||
|
||||
def __init__(self, client: AsyncOpenAI):
|
||||
self._client = client
|
||||
self._logger = logging.getLogger(__name__)
|
||||
self._default_model = "gpt-4o-2024-08-06"
|
||||
|
||||
async def prediction(
|
||||
self,
|
||||
model: str,
|
||||
messages: List[ChatCompletionMessageParam],
|
||||
fields_to_check: Optional[List[str]],
|
||||
temperature: float,
|
||||
check_blacklisted: bool = True,
|
||||
token_count: int = -1
|
||||
):
|
||||
if token_count == -1:
|
||||
token_count = self._count_total_tokens(messages)
|
||||
return await self._prediction(model, messages, token_count, fields_to_check, temperature, 0, check_blacklisted)
|
||||
|
||||
async def _prediction(
|
||||
self,
|
||||
model: str,
|
||||
messages: List[ChatCompletionMessageParam],
|
||||
token_count: int,
|
||||
fields_to_check: Optional[List[str]],
|
||||
temperature: float,
|
||||
try_count: int,
|
||||
check_blacklisted: bool,
|
||||
):
|
||||
result = await self._client.chat.completions.create(
|
||||
model=model,
|
||||
max_tokens=int(self.MAX_TOKENS - token_count - 300),
|
||||
temperature=float(temperature),
|
||||
messages=messages,
|
||||
response_format={"type": "json_object"}
|
||||
)
|
||||
result = result.choices[0].message.content
|
||||
|
||||
if check_blacklisted:
|
||||
found_blacklisted_word = self._get_found_blacklisted_words(result)
|
||||
|
||||
if found_blacklisted_word is not None and try_count < self.TRY_LIMIT:
|
||||
self._logger.warning("Result contains blacklisted words: " + str(found_blacklisted_word))
|
||||
return await self._prediction(
|
||||
model, messages, token_count, fields_to_check, temperature, (try_count + 1), check_blacklisted
|
||||
)
|
||||
elif found_blacklisted_word is not None and try_count >= self.TRY_LIMIT:
|
||||
return ""
|
||||
|
||||
if fields_to_check is None:
|
||||
return json.loads(result)
|
||||
|
||||
if not self._check_fields(result, fields_to_check) and try_count < self.TRY_LIMIT:
|
||||
return await self._prediction(
|
||||
model, messages, token_count, fields_to_check, temperature, (try_count + 1), check_blacklisted
|
||||
)
|
||||
|
||||
return json.loads(result)
|
||||
|
||||
async def prediction_override(self, **kwargs):
|
||||
return await self._client.chat.completions.create(
|
||||
**kwargs
|
||||
)
|
||||
|
||||
@staticmethod
|
||||
def _get_found_blacklisted_words(text: str):
|
||||
text_lower = text.lower()
|
||||
for word in BLACKLISTED_WORDS:
|
||||
if re.search(r'\b' + re.escape(word) + r'\b', text_lower):
|
||||
return word
|
||||
return None
|
||||
|
||||
@staticmethod
|
||||
def _count_total_tokens(messages):
|
||||
total_tokens = 0
|
||||
for message in messages:
|
||||
total_tokens += count_tokens(message["content"])["n_tokens"]
|
||||
return total_tokens
|
||||
|
||||
@staticmethod
|
||||
def _check_fields(obj, fields):
|
||||
return all(field in obj for field in fields)
|
||||
|
||||
async def pydantic_prediction(
|
||||
self,
|
||||
messages: List[ChatCompletionMessageParam],
|
||||
map_to_model: Callable,
|
||||
json_scheme: str,
|
||||
*,
|
||||
model: Optional[str] = None,
|
||||
temperature: Optional[float] = None,
|
||||
max_retries: int = 3
|
||||
) -> List[T] | T | None:
|
||||
params = {
|
||||
"messages": messages,
|
||||
"response_format": {"type": "json_object"},
|
||||
"model": model if model else self._default_model
|
||||
}
|
||||
|
||||
if temperature:
|
||||
params["temperature"] = temperature
|
||||
|
||||
attempt = 0
|
||||
while attempt < max_retries:
|
||||
result = await self._client.chat.completions.create(**params)
|
||||
result_content = result.choices[0].message.content
|
||||
try:
|
||||
result_json = json.loads(result_content)
|
||||
return map_to_model(result_json)
|
||||
except Exception as e:
|
||||
attempt += 1
|
||||
self._logger.info(f"GPT returned malformed response: {result_content}\n {str(e)}")
|
||||
params["messages"] = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": (
|
||||
"Your previous response wasn't in the json format I've explicitly told you to output. "
|
||||
f"In your next response, you will fix it and return me just the json I've asked."
|
||||
)
|
||||
},
|
||||
{
|
||||
"role": "user",
|
||||
"content": (
|
||||
f"Previous response: {result_content}\n"
|
||||
f"JSON format: {json_scheme}"
|
||||
)
|
||||
}
|
||||
]
|
||||
if attempt >= max_retries:
|
||||
self._logger.error(f"Max retries exceeded!")
|
||||
return None
|
||||
|
||||
@@ -1,22 +1,22 @@
|
||||
import os
|
||||
|
||||
from fastapi.concurrency import run_in_threadpool
|
||||
|
||||
from whisper import Whisper
|
||||
from app.services.abc import ISpeechToTextService
|
||||
|
||||
|
||||
class OpenAIWhisper(ISpeechToTextService):
|
||||
|
||||
def __init__(self, model: Whisper):
|
||||
self._model = model
|
||||
|
||||
async def speech_to_text(self, file_path):
|
||||
if os.path.exists(file_path):
|
||||
result = await run_in_threadpool(
|
||||
self._model.transcribe, file_path, fp16=False, language='English', verbose=False
|
||||
)
|
||||
return result["text"]
|
||||
else:
|
||||
print("File not found:", file_path)
|
||||
raise Exception("File " + file_path + " not found.")
|
||||
import os
|
||||
|
||||
from fastapi.concurrency import run_in_threadpool
|
||||
|
||||
from whisper import Whisper
|
||||
from app.services.abc import ISpeechToTextService
|
||||
|
||||
|
||||
class OpenAIWhisper(ISpeechToTextService):
|
||||
|
||||
def __init__(self, model: Whisper):
|
||||
self._model = model
|
||||
|
||||
async def speech_to_text(self, file_path):
|
||||
if os.path.exists(file_path):
|
||||
result = await run_in_threadpool(
|
||||
self._model.transcribe, file_path, fp16=False, language='English', verbose=False
|
||||
)
|
||||
return result["text"]
|
||||
else:
|
||||
print("File not found:", file_path)
|
||||
raise Exception("File " + file_path + " not found.")
|
||||
|
||||
Reference in New Issue
Block a user