Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,285 +1,30 @@
|
|
| 1 |
-
from fastapi import FastAPI
|
| 2 |
-
from
|
| 3 |
-
from
|
| 4 |
-
import torch
|
| 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 |
-
image: str # Base64 encoded image
|
| 32 |
-
model_name: str = "oasis500m" # Default to oasis model
|
| 33 |
-
|
| 34 |
-
class InferenceResponse(BaseModel):
|
| 35 |
-
predictions: List[Dict[str, Any]]
|
| 36 |
-
model_used: str
|
| 37 |
-
confidence_scores: List[float]
|
| 38 |
-
|
| 39 |
-
def load_models():
|
| 40 |
-
"""Load both models from local files"""
|
| 41 |
-
global oasis_model, oasis_processor, vit_model, vit_processor
|
| 42 |
-
|
| 43 |
-
try:
|
| 44 |
-
logger.info("Loading Oasis 500M model from local files...")
|
| 45 |
-
# Load Oasis model from local files
|
| 46 |
-
oasis_processor = AutoImageProcessor.from_pretrained("microsoft/oasis-500m")
|
| 47 |
-
oasis_model = AutoModelForImageClassification.from_pretrained(
|
| 48 |
-
"microsoft/oasis-500m",
|
| 49 |
-
local_files_only=False # Will download config but use local weights
|
| 50 |
-
)
|
| 51 |
-
|
| 52 |
-
# Load local weights if available
|
| 53 |
-
oasis_model_path = "/app/models/oasis500m.safetensors"
|
| 54 |
-
if os.path.exists(oasis_model_path):
|
| 55 |
-
logger.info("Loading Oasis weights from local file...")
|
| 56 |
-
from safetensors.torch import load_file
|
| 57 |
-
state_dict = load_file(oasis_model_path)
|
| 58 |
-
oasis_model.load_state_dict(state_dict, strict=False)
|
| 59 |
-
|
| 60 |
-
oasis_model.eval()
|
| 61 |
-
|
| 62 |
-
logger.info("Loading ViT-L-20 model from local files...")
|
| 63 |
-
# Load ViT model from local files
|
| 64 |
-
vit_processor = AutoImageProcessor.from_pretrained("google/vit-large-patch16-224")
|
| 65 |
-
vit_model = AutoModelForImageClassification.from_pretrained(
|
| 66 |
-
"google/vit-large-patch16-224",
|
| 67 |
-
local_files_only=False # Will download config but use local weights
|
| 68 |
-
)
|
| 69 |
-
|
| 70 |
-
# Load local weights if available
|
| 71 |
-
vit_model_path = "/app/models/vit-l-20.safetensors"
|
| 72 |
-
if os.path.exists(vit_model_path):
|
| 73 |
-
logger.info("Loading ViT weights from local file...")
|
| 74 |
-
from safetensors.torch import load_file
|
| 75 |
-
state_dict = load_file(vit_model_path)
|
| 76 |
-
vit_model.load_state_dict(state_dict, strict=False)
|
| 77 |
-
|
| 78 |
-
vit_model.eval()
|
| 79 |
-
|
| 80 |
-
logger.info("All models loaded successfully!")
|
| 81 |
-
|
| 82 |
-
except Exception as e:
|
| 83 |
-
logger.error(f"Error loading models: {e}")
|
| 84 |
-
raise e
|
| 85 |
-
|
| 86 |
-
@app.on_event("startup")
|
| 87 |
-
async def startup_event():
|
| 88 |
-
"""Load models when the application starts"""
|
| 89 |
-
load_models()
|
| 90 |
-
|
| 91 |
-
@app.get("/")
|
| 92 |
-
async def root():
|
| 93 |
-
"""Root endpoint with API information"""
|
| 94 |
-
return {
|
| 95 |
-
"message": "ChatGPT Oasis Model Inference API",
|
| 96 |
-
"version": "1.0.0",
|
| 97 |
-
"deployed_on": "Hugging Face Spaces (Docker)",
|
| 98 |
-
"available_models": ["oasis500m", "vit-l-20"],
|
| 99 |
-
"endpoints": {
|
| 100 |
-
"health": "/health",
|
| 101 |
-
"inference": "/inference",
|
| 102 |
-
"upload_inference": "/upload_inference",
|
| 103 |
-
"predict": "/predict"
|
| 104 |
-
},
|
| 105 |
-
"usage": {
|
| 106 |
-
"base64_inference": "POST /inference with JSON body containing 'image' (base64) and 'model_name'",
|
| 107 |
-
"file_upload": "POST /upload_inference with multipart form containing 'file' and optional 'model_name'",
|
| 108 |
-
"simple_predict": "POST /predict with file upload for quick inference"
|
| 109 |
-
}
|
| 110 |
-
}
|
| 111 |
-
|
| 112 |
-
@app.get("/health")
|
| 113 |
-
async def health_check():
|
| 114 |
-
"""Health check endpoint"""
|
| 115 |
-
models_status = {
|
| 116 |
-
"oasis500m": oasis_model is not None,
|
| 117 |
-
"vit-l-20": vit_model is not None
|
| 118 |
-
}
|
| 119 |
-
|
| 120 |
-
# Check if model files exist
|
| 121 |
-
model_files = {
|
| 122 |
-
"oasis500m": os.path.exists("/app/models/oasis500m.safetensors"),
|
| 123 |
-
"vit-l-20": os.path.exists("/app/models/vit-l-20.safetensors")
|
| 124 |
-
}
|
| 125 |
-
|
| 126 |
-
return {
|
| 127 |
-
"status": "healthy",
|
| 128 |
-
"models_loaded": models_status,
|
| 129 |
-
"model_files_present": model_files,
|
| 130 |
-
"deployment": "huggingface-spaces-docker"
|
| 131 |
-
}
|
| 132 |
-
|
| 133 |
-
def process_image_with_model(image: Image.Image, model_name: str):
|
| 134 |
-
"""Process image with the specified model"""
|
| 135 |
-
if model_name == "oasis500m":
|
| 136 |
-
if oasis_model is None or oasis_processor is None:
|
| 137 |
-
raise HTTPException(status_code=500, detail="Oasis model not loaded")
|
| 138 |
-
|
| 139 |
-
inputs = oasis_processor(images=image, return_tensors="pt")
|
| 140 |
-
with torch.no_grad():
|
| 141 |
-
outputs = oasis_model(**inputs)
|
| 142 |
-
logits = outputs.logits
|
| 143 |
-
probabilities = F.softmax(logits, dim=-1)
|
| 144 |
-
|
| 145 |
-
# Get top predictions
|
| 146 |
-
top_probs, top_indices = torch.topk(probabilities, 5)
|
| 147 |
-
|
| 148 |
-
predictions = []
|
| 149 |
-
for i in range(top_indices.shape[1]):
|
| 150 |
-
pred = {
|
| 151 |
-
"label": oasis_model.config.id2label[top_indices[0][i].item()],
|
| 152 |
-
"confidence": top_probs[0][i].item()
|
| 153 |
-
}
|
| 154 |
-
predictions.append(pred)
|
| 155 |
-
|
| 156 |
-
return predictions
|
| 157 |
-
|
| 158 |
-
elif model_name == "vit-l-20":
|
| 159 |
-
if vit_model is None or vit_processor is None:
|
| 160 |
-
raise HTTPException(status_code=500, detail="ViT model not loaded")
|
| 161 |
-
|
| 162 |
-
inputs = vit_processor(images=image, return_tensors="pt")
|
| 163 |
-
with torch.no_grad():
|
| 164 |
-
outputs = vit_model(**inputs)
|
| 165 |
-
logits = outputs.logits
|
| 166 |
-
probabilities = F.softmax(logits, dim=-1)
|
| 167 |
-
|
| 168 |
-
# Get top predictions
|
| 169 |
-
top_probs, top_indices = torch.topk(probabilities, 5)
|
| 170 |
-
|
| 171 |
-
predictions = []
|
| 172 |
-
for i in range(top_indices.shape[1]):
|
| 173 |
-
pred = {
|
| 174 |
-
"label": vit_model.config.id2label[top_indices[0][i].item()],
|
| 175 |
-
"confidence": top_probs[0][i].item()
|
| 176 |
-
}
|
| 177 |
-
predictions.append(pred)
|
| 178 |
-
|
| 179 |
-
return predictions
|
| 180 |
-
|
| 181 |
-
else:
|
| 182 |
-
raise HTTPException(status_code=400, detail=f"Unknown model: {model_name}")
|
| 183 |
-
|
| 184 |
-
@app.post("/inference", response_model=InferenceResponse)
|
| 185 |
-
async def inference(request: InferenceRequest):
|
| 186 |
-
"""Inference endpoint using base64 encoded image"""
|
| 187 |
-
try:
|
| 188 |
-
import base64
|
| 189 |
-
|
| 190 |
-
# Decode base64 image
|
| 191 |
-
image_data = base64.b64decode(request.image)
|
| 192 |
-
image = Image.open(io.BytesIO(image_data)).convert('RGB')
|
| 193 |
-
|
| 194 |
-
# Process with model
|
| 195 |
-
predictions = process_image_with_model(image, request.model_name)
|
| 196 |
-
|
| 197 |
-
# Extract confidence scores
|
| 198 |
-
confidence_scores = [pred["confidence"] for pred in predictions]
|
| 199 |
-
|
| 200 |
-
return InferenceResponse(
|
| 201 |
-
predictions=predictions,
|
| 202 |
-
model_used=request.model_name,
|
| 203 |
-
confidence_scores=confidence_scores
|
| 204 |
-
)
|
| 205 |
-
|
| 206 |
-
except Exception as e:
|
| 207 |
-
logger.error(f"Inference error: {e}")
|
| 208 |
-
raise HTTPException(status_code=500, detail=str(e))
|
| 209 |
-
|
| 210 |
-
@app.post("/upload_inference", response_model=InferenceResponse)
|
| 211 |
-
async def upload_inference(
|
| 212 |
-
file: UploadFile = File(...),
|
| 213 |
-
model_name: str = "oasis500m"
|
| 214 |
-
):
|
| 215 |
-
"""Inference endpoint using file upload"""
|
| 216 |
-
try:
|
| 217 |
-
# Validate file type
|
| 218 |
-
if not file.content_type.startswith('image/'):
|
| 219 |
-
raise HTTPException(status_code=400, detail="File must be an image")
|
| 220 |
-
|
| 221 |
-
# Read and process image
|
| 222 |
-
image_data = await file.read()
|
| 223 |
-
image = Image.open(io.BytesIO(image_data)).convert('RGB')
|
| 224 |
-
|
| 225 |
-
# Process with model
|
| 226 |
-
predictions = process_image_with_model(image, model_name)
|
| 227 |
-
|
| 228 |
-
# Extract confidence scores
|
| 229 |
-
confidence_scores = [pred["confidence"] for pred in predictions]
|
| 230 |
-
|
| 231 |
-
return InferenceResponse(
|
| 232 |
-
predictions=predictions,
|
| 233 |
-
model_used=model_name,
|
| 234 |
-
confidence_scores=confidence_scores
|
| 235 |
-
)
|
| 236 |
-
|
| 237 |
-
except Exception as e:
|
| 238 |
-
logger.error(f"Upload inference error: {e}")
|
| 239 |
-
raise HTTPException(status_code=500, detail=str(e))
|
| 240 |
-
|
| 241 |
-
@app.get("/models")
|
| 242 |
-
async def list_models():
|
| 243 |
-
"""List available models and their status"""
|
| 244 |
-
return {
|
| 245 |
-
"available_models": [
|
| 246 |
-
{
|
| 247 |
-
"name": "oasis500m",
|
| 248 |
-
"description": "Oasis 500M vision model",
|
| 249 |
-
"loaded": oasis_model is not None,
|
| 250 |
-
"file_present": os.path.exists("/app/models/oasis500m.safetensors")
|
| 251 |
-
},
|
| 252 |
-
{
|
| 253 |
-
"name": "vit-l-20",
|
| 254 |
-
"description": "Vision Transformer Large model",
|
| 255 |
-
"loaded": vit_model is not None,
|
| 256 |
-
"file_present": os.path.exists("/app/models/vit-l-20.safetensors")
|
| 257 |
-
}
|
| 258 |
-
]
|
| 259 |
-
}
|
| 260 |
-
|
| 261 |
-
# Hugging Face Spaces specific endpoint for Gradio compatibility
|
| 262 |
-
@app.post("/predict")
|
| 263 |
-
async def predict(file: UploadFile = File(...)):
|
| 264 |
-
"""Simple prediction endpoint for Hugging Face Spaces integration"""
|
| 265 |
-
try:
|
| 266 |
-
# Validate file type
|
| 267 |
-
if not file.content_type.startswith('image/'):
|
| 268 |
-
raise HTTPException(status_code=400, detail="File must be an image")
|
| 269 |
-
|
| 270 |
-
# Read and process image
|
| 271 |
-
image_data = await file.read()
|
| 272 |
-
image = Image.open(io.BytesIO(image_data)).convert('RGB')
|
| 273 |
-
|
| 274 |
-
# Process with default model (oasis500m)
|
| 275 |
-
predictions = process_image_with_model(image, "oasis500m")
|
| 276 |
-
|
| 277 |
-
# Return simplified format for Gradio
|
| 278 |
-
return {
|
| 279 |
-
"predictions": predictions[:3], # Top 3 predictions
|
| 280 |
-
"model_used": "oasis500m"
|
| 281 |
-
}
|
| 282 |
-
|
| 283 |
-
except Exception as e:
|
| 284 |
-
logger.error(f"Predict error: {e}")
|
| 285 |
-
raise HTTPException(status_code=500, detail=str(e))
|
|
|
|
| 1 |
+
from fastapi import FastAPI
|
| 2 |
+
from pydantic import BaseModel
|
| 3 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 4 |
+
import torch
|
| 5 |
+
|
| 6 |
+
app = FastAPI()
|
| 7 |
+
|
| 8 |
+
# Load model & tokenizer
|
| 9 |
+
MODEL_PATH = "./" # since it's inside the same repo
|
| 10 |
+
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
|
| 11 |
+
model = AutoModelForCausalLM.from_pretrained(
|
| 12 |
+
MODEL_PATH,
|
| 13 |
+
torch_dtype=torch.float16,
|
| 14 |
+
device_map="auto"
|
| 15 |
+
)
|
| 16 |
+
|
| 17 |
+
class RequestBody(BaseModel):
|
| 18 |
+
prompt: str
|
| 19 |
+
max_length: int = 100
|
| 20 |
+
|
| 21 |
+
@app.post("/generate")
|
| 22 |
+
def generate_text(req: RequestBody):
|
| 23 |
+
inputs = tokenizer(req.prompt, return_tensors="pt").to(model.device)
|
| 24 |
+
outputs = model.generate(**inputs, max_length=req.max_length)
|
| 25 |
+
text = tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 26 |
+
return {"generated_text": text}
|
| 27 |
+
|
| 28 |
+
@app.get("/")
|
| 29 |
+
def root():
|
| 30 |
+
return {"message": "FastAPI Hugging Face Space is running!"}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|