import argparse import uvicorn import sys import os import io from transformers import M2M100Tokenizer, M2M100ForConditionalGeneration from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline import time import json from typing import List import torch import logging import string import random import base64 import re import requests from utils.enver import enver import shutil import tempfile import numpy as np from fastapi import FastAPI, Response, File, UploadFile, Form from fastapi.encoders import jsonable_encoder from fastapi.responses import JSONResponse, StreamingResponse from pydantic import BaseModel, Field from sse_starlette.sse import EventSourceResponse from utils.logger import logger from networks.message_streamer import MessageStreamer from messagers.message_composer import MessageComposer from googletrans import Translator from io import BytesIO from gtts import gTTS from fastapi.middleware.cors import CORSMiddleware from pathlib import Path from tempfile import NamedTemporaryFile class ChatAPIApp: def __init__(self): self.app = FastAPI( docs_url="/", title="HuggingFace LLM API", swagger_ui_parameters={"defaultModelsExpandDepth": -1}, version="1.0", ) self.setup_routes() def get_available_langs(self): f = open('apis/lang_name.json', "r") self.available_models = json.loads(f.read()) return self.available_models class TranslateCompletionsPostItem(BaseModel): from_language: str = Field( default="en", description="(str) `Detect`", ) to_language: str = Field( default="fa", description="(str) `en`", ) input_text: str = Field( default="Hello", description="(str) `Text for translate`", ) def translate_completions(self, item: TranslateCompletionsPostItem): translator = Translator() f = open('apis/lang_name.json', "r") available_langs = json.loads(f.read()) from_lang = 'en' to_lang = 'en' for lang_item in available_langs: if item.to_language == lang_item['code']: to_lang = item.to_language break translated = translator.translate(item.input_text, dest=to_lang) item_response = { "from_language": translated.src, "to_language": translated.dest, "text": item.input_text, "translate": translated.text } json_compatible_item_data = jsonable_encoder(item_response) return JSONResponse(content=json_compatible_item_data) def translate_ai_completions(self, item: TranslateCompletionsPostItem): translator = Translator() #print(os.getcwd()) f = open('apis/lang_name.json', "r") available_langs = json.loads(f.read()) from_lang = 'en' to_lang = 'en' for lang_item in available_langs: if item.to_language == lang_item['code']: to_lang = item.to_language if item.from_language == lang_item['code']: from_lang = item.from_language if to_lang == 'auto': to_lang = 'en' if from_lang == 'auto': from_lang = translator.detect(item.input_text).lang if torch.cuda.is_available(): device = torch.device("cuda:0") else: device = torch.device("cpu") logging.warning("GPU not found, using CPU, translation will be very slow.") time_start = time.time() #TRANSFORMERS_CACHE pretrained_model = "facebook/m2m100_1.2B" cache_dir = "models/" tokenizer = M2M100Tokenizer.from_pretrained(pretrained_model, cache_dir=cache_dir) model = M2M100ForConditionalGeneration.from_pretrained( pretrained_model, cache_dir=cache_dir ).to(device) model.eval() tokenizer.src_lang = from_lang with torch.no_grad(): encoded_input = tokenizer(item.input_text, return_tensors="pt").to(device) generated_tokens = model.generate( **encoded_input, forced_bos_token_id=tokenizer.get_lang_id(to_lang) ) translated_text = tokenizer.batch_decode( generated_tokens, skip_special_tokens=True )[0] time_end = time.time() translated = translated_text item_response = { "from_language": from_lang, "to_language": to_lang, "text": item.input_text, "translate": translated, "start": str(time_start), "end": str(time_end) } json_compatible_item_data = jsonable_encoder(item_response) return JSONResponse(content=json_compatible_item_data) class TranslateAiPostItem(BaseModel): model: str = Field( default="t5-base", description="(str) `Model Name`", ) from_language: str = Field( default="en", description="(str) `translate from`", ) to_language: str = Field( default="fa", description="(str) `translate to`", ) input_text: str = Field( default="Hello", description="(str) `Text for translate`", ) def ai_translate(self, item:TranslateAiPostItem): MODEL_MAP = { "t5-base": "t5-base", "t5-small": "t5-small", "t5-large": "t5-large", "t5-3b": "t5-3b", "mbart-large-50-many-to-many-mmt": "facebook/mbart-large-50-many-to-many-mmt", "nllb-200-distilled-600M": "facebook/nllb-200-distilled-600M", "madlad400-3b-mt": "jbochi/madlad400-3b-mt", "default": "t5-base", } if item.model in MODEL_MAP.keys(): target_model = item.model else: target_model = "default" real_name = MODEL_MAP[target_model] read_model = AutoModelForSeq2SeqLM.from_pretrained(real_name) tokenizer = AutoTokenizer.from_pretrained(real_name) #translator = pipeline("translation", model=read_model, tokenizer=tokenizer, src_lang=item.from_language, tgt_lang=item.to_language) translate_query = ( f"translation_{item.from_language}_to_{item.to_language}" ) translator = pipeline(translate_query) result = translator(item.input_text) item_response = { "statue": 200, "result": result, } json_compatible_item_data = jsonable_encoder(item_response) return JSONResponse(content=json_compatible_item_data) class DetectLanguagePostItem(BaseModel): input_text: str = Field( default="Hello, how are you?", description="(str) `Text for detection`", ) def detect_language(self, item: DetectLanguagePostItem): translator = Translator() detected = translator.detect(item.input_text) item_response = { "lang": detected.lang, "confidence": detected.confidence, } json_compatible_item_data = jsonable_encoder(item_response) return JSONResponse(content=json_compatible_item_data) class TTSPostItem(BaseModel): input_text: str = Field( default="Hello", description="(str) `Text for TTS`", ) from_language: str = Field( default="en", description="(str) `TTS language`", ) def text_to_speech(self, item: TTSPostItem): try: audioobj = gTTS(text = item.input_text, lang = item.from_language, slow = False) fileName = ''.join(random.SystemRandom().choice(string.ascii_uppercase + string.digits) for _ in range(10)); fileName = fileName + ".mp3"; mp3_fp = BytesIO() #audioobj.save(fileName) #audioobj.write_to_fp(mp3_fp) #buffer = bytearray(mp3_fp.read()) #base64EncodedStr = base64.encodebytes(buffer) #mp3_fp.read() #return Response(content=mp3_fp.tell(), media_type="audio/mpeg") return StreamingResponse(audioobj.stream()) except: item_response = { "status": 400 } json_compatible_item_data = jsonable_encoder(item_response) return JSONResponse(content=json_compatible_item_data) def setup_routes(self): for prefix in ["", "/v1"]: self.app.get( prefix + "/langs", summary="Get available languages", )(self.get_available_langs) self.app.post( prefix + "/translate", summary="translate text", )(self.translate_completions) self.app.post( prefix + "/translate/ai", summary="translate text with ai", )(self.translate_ai_completions) self.app.post( prefix + "/detect", summary="detect language", )(self.detect_language) self.app.post( prefix + "/tts", summary="text to speech", )(self.text_to_speech) class ArgParser(argparse.ArgumentParser): def __init__(self, *args, **kwargs): super(ArgParser, self).__init__(*args, **kwargs) self.add_argument( "-s", "--server", type=str, default="0.0.0.0", help="Server IP for HF LLM Chat API", ) self.add_argument( "-p", "--port", type=int, default=23333, help="Server Port for HF LLM Chat API", ) self.add_argument( "-d", "--dev", default=False, action="store_true", help="Run in dev mode", ) self.args = self.parse_args(sys.argv[1:]) app = ChatAPIApp().app app.add_middleware( CORSMiddleware, allow_origins=["*"], allow_credentials=True, allow_methods=["*"], allow_headers=["*"], ) @app.post("/transcribe") async def whisper_transcribe( audio_file: UploadFile = File(description="Audio file for transcribe"), language: str = Form(), model: str = Form(), ): MODEL_MAP = { "whisper-small": "openai/whisper-small", "whisper-medium": "openai/whisper-medium", "whisper-large": "openai/whisper-large", "default": "openai/whisper-small", } AUDIO_MAP = { "audio/wav": "audio/wav", "audio/mpeg": "audio/mpeg", "audio/x-flac": "audio/x-flac", } item_response = { "statue": 200, "result": "", "start": 0, "end": 0 } if audio_file.content_type in AUDIO_MAP.keys(): if model in MODEL_MAP.keys(): target_model = model else: target_model = "default" real_name = MODEL_MAP[target_model] device = 0 if torch.cuda.is_available() else "cpu" pipe = pipeline( task="automatic-speech-recognition", model=real_name, chunk_length_s=30, device=device, ) time_start = time.time() pipe.model.config.forced_decoder_ids = pipe.tokenizer.get_decoder_prompt_ids(language=language, task="transcribe") try: suffix = Path(audio_file.filename).suffix with NamedTemporaryFile(delete=False, suffix=suffix) as tmp: shutil.copyfileobj(audio_file.file, tmp) tmp_path = Path(tmp.name) finally: audio_file.file.close() #file_data = await audio_file.read() # rv = data.encode('utf-8') #rv = base64.b64encode(file_data).decode() #print(rv, "rvrvrvrvr") audio_data = np.fromfile(tmp_path) text = pipe(audio_data)["text"] time_end = time.time() item_response["status"] = 200 item_response["result"] = text item_response["start"] = time_start item_response["end"] = time_end else: item_response["status"] = 400 item_response["result"] = 'Acceptable files: audio/wav,audio/mpeg,audio/x-flac' return item_response if __name__ == "__main__": args = ArgParser().args if args.dev: uvicorn.run("__main__:app", host=args.server, port=args.port, reload=True) else: uvicorn.run("__main__:app", host=args.server, port=args.port, reload=False) # python -m apis.chat_api # [Docker] on product mode # python -m apis.chat_api -d # [Dev] on develop mode