File size: 5,640 Bytes
65f58ee |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 |
from fastapi import FastAPI, HTTPException
# from fastapi import FastAPI, HTTPException
from fastapi.middleware.cors import CORSMiddleware
# from fastapi.middleware.cors import CORSMiddleware
from pydantic import BaseModel
# from pydantic import BaseModel
# from pydantic import BaseModel
# from pydantic import BaseModel
# from pydantic import BaseModel
# from pydantic import BaseModel
# from pydantic import BaseModel
import librosa
# import librosa
import torch
import base64
# import base64
import io
# import io
import logging
import numpy as np
# import numpy as np
# import numpy as np
from transformers import AutoModel, AutoTokenizer
# from transformers import AutoModel, AutoTokenizer
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
app = FastAPI()
# Add CORS middleware
app.add_middleware(
CORSMiddleware,
allow_origins=["*"],
allow_credentials=True,
allow_methods=["*"],
allow_headers=["*"],
)
class AudioRequest(BaseModel):
audio_data: str
sample_rate: int
class AudioResponse(BaseModel):
audio_data: str
text: str = ""
# Model initialization status
INITIALIZATION_STATUS = {
"model_loaded": False,
"error": None
}
# Global model and tokenizer instances
class Model:
def __init__(self):
self.model = model = AutoModel.from_pretrained(
'./models/checkpoint',
trust_remote_code=True,
torch_dtype=torch.bfloat16,
attn_implementation='sdpa'
)
model = model.eval().cuda()
self.tokenizer = AutoTokenizer.from_pretrained(
'./models/checkpoint',
trust_remote_code=True
)
# Initialize TTS
model.init_tts()
model.tts.float() # Convert TTS to float32 if needed
self.model_in_sr = 16000
self.model_out_sr = 24000
self.ref_audio, _ = librosa.load('./ref_audios/female_example.wav', sr=self.model_in_sr, mono=True) # load the reference audio
self.sys_prompt = model.get_sys_prompt(ref_audio=self.ref_audio, mode='audio_assistant', language='en')
# warmup
audio_data = librosa.load('./ref_audios/male_example.wav', sr=self.model_in_sr, mono=True)[0]
_ = self.inference(audio_data, self.model_in_sr)
def inference(self, audio_np, input_audio_sr):
if input_audio_sr != self.model_in_sr:
audio_np = librosa.resample(audio_np, orig_sr=input_audio_sr, target_sr=self.model_in_sr)
user_question = {'role': 'user', 'content': [audio_np]}
# round one
msgs = [self.sys_prompt, user_question]
res = self.model.chat(
msgs=msgs,
tokenizer=self.tokenizer,
sampling=True,
max_new_tokens=128,
use_tts_template=True,
generate_audio=True,
temperature=0.3,
)
audio = res["audio_wav"].cpu().numpy()
if self.model_out_sr != input_audio_sr:
audio = librosa.resample(audio, orig_sr=self.model_out_sr, target_sr=input_audio_sr)
return audio, res["text"]
def initialize_model():
"""Initialize the MiniCPM model"""
global model, INITIALIZATION_STATUS
try:
logger.info("Initializing model...")
model = Model()
INITIALIZATION_STATUS["model_loaded"] = True
logger.info("MiniCPM model initialized successfully")
return True
except Exception as e:
INITIALIZATION_STATUS["error"] = str(e)
logger.error(f"Failed to initialize model: {e}")
return False
@app.on_event("startup")
async def startup_event():
"""Initialize model on startup"""
initialize_model()
@app.get("/api/v1/health")
def health_check():
"""Health check endpoint"""
status = {
"status": "healthy" if INITIALIZATION_STATUS["model_loaded"] else "initializing",
"model_loaded": INITIALIZATION_STATUS["model_loaded"],
"error": INITIALIZATION_STATUS["error"]
}
return status
@app.post("/api/v1/inference")
async def inference(request: AudioRequest) -> AudioResponse:
"""Run inference with MiniCPM model"""
if not INITIALIZATION_STATUS["model_loaded"]:
raise HTTPException(
status_code=503,
detail=f"Model not ready. Status: {INITIALIZATION_STATUS}"
)
try:
# Decode audio data from base64
audio_bytes = base64.b64decode(request.audio_data)
audio_np = np.load(io.BytesIO(audio_bytes)).flatten()
# Generate response
import time
start = time.time()
print(f"starting inference with audio length {audio_np.shape}")
audio_response, text_response = model.inference(audio_np, request.sample_rate)
print(f"inference took {time.time() - start} seconds")
# If we got audio, save it and encode to base64
buffer = io.BytesIO()
np.save(buffer, audio_response)
audio_b64 = base64.b64encode(buffer.getvalue()).decode()
return AudioResponse(
audio_data=audio_b64,
text=text_response
)
except Exception as e:
logger.error(f"Inference failed: {str(e)}")
raise HTTPException(
status_code=500,
detail=str(e)
)
if __name__ == "__main__":
import uvicorn
uvicorn.run(app, host="0.0.0.0", port=8000)
# 2541
# 8039
# 4517
# 2928
# 4271
# 6148
# 6843
# 1322
# 6293
# 5317
# 3270
# 2405
# 7264
# 2900
# 6266
# 6321
# 8042
# 1716
# 6329
# 8477
# 7676
# 5182
# 8343
# 7225
|