Upload 14 files

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+models/resnet18.onnx.data filter=lfs diff=lfs merge=lfs -text

Dockerfile

@@ -0,0 +1,23 @@
+FROM python:3.11-slim
+
+# Create a user to avoid running as root (Required for Hugging Face Spaces)
+RUN useradd -m -u 1000 user
+USER user
+ENV PATH="/home/user/.local/bin:${PATH}"
+
+WORKDIR /app
+
+# Copy requirements and install dependencies
+COPY --chown=user requirements.txt .
+RUN pip install --no-cache-dir --user -r requirements.txt
+
+# Copy the application code and models
+COPY --chown=user app/ ./app/
+COPY --chown=user models/ ./models/
+
+# Use port 7860 (Standard for Hugging Face Spaces)
+EXPOSE 7860
+
+# Run with multiple workers for higher throughput
+# Note: On Linux/Docker, uvicorn workers work perfectly
+CMD ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "7860", "--workers", "4", "--log-level", "warning"]

app/main.py

@@ -0,0 +1,35 @@
+from fastapi import FastAPI, File, UploadFile, HTTPException
+from concurrent.futures import ProcessPoolExecutor
+import asyncio
+from app.model import run_inference
+from app.schemas import PredictionResponse
+
+app = FastAPI(title="ResNet-18 Image Classifier", version="1.0.0")
+
+# ใช้ 6 Workers ตามจำนวน Physical Cores ของ Ryzen 7500F
+executor = ProcessPoolExecutor(max_workers=6)
+
+MAX_FILE_SIZE = 10 * 1024 * 1024  # 10 MB
+ALLOWED_CONTENT_TYPES = {"image/jpeg", "image/png", "image/webp", "image/gif"}
+
+@app.get("/health")
+async def health():
+    return {"status": "ok"}
+
+@app.post("/predict", response_model=PredictionResponse)
+async def predict(file: UploadFile = File(...)):
+    if file.content_type not in ALLOWED_CONTENT_TYPES:
+        raise HTTPException(status_code=415, detail="Unsupported media type")
+
+    image_bytes = await file.read()
+    
+    if len(image_bytes) > MAX_FILE_SIZE:
+        raise HTTPException(status_code=413, detail="File too large")
+
+    # รัน Inference ใน ProcessPoolExecutor เพื่อกระจายโหลดลง 6 Cores
+    loop = asyncio.get_event_loop()
+    try:
+        result = await loop.run_in_executor(executor, run_inference, image_bytes)
+        return result
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Inference error: {str(e)}")

app/model.py

@@ -0,0 +1,44 @@
+import onnxruntime as ort
+import numpy as np
+from PIL import Image
+import io
+import time
+from transformers import AutoImageProcessor, ResNetForImageClassification
+
+# Load feature extractor
+processor = AutoImageProcessor.from_pretrained("microsoft/resnet-18")
+
+# Optimize session for multi-process environment
+sess_options = ort.SessionOptions()
+sess_options.intra_op_num_threads = 1 # One thread per process worker
+sess_options.inter_op_num_threads = 1
+sess_options.graph_optimization_level = ort.GraphOptimizationLevel.ORT_ENABLE_ALL
+
+# Load ONNX session
+session = ort.InferenceSession(
+    "models/resnet18_quantized.onnx",
+    sess_options=sess_options,
+    providers=["CPUExecutionProvider"]
+)
+
+# Load label mapping
+cfg = ResNetForImageClassification.from_pretrained("microsoft/resnet-18").config
+
+def run_inference(image_bytes: bytes) -> dict:
+    img = Image.open(io.BytesIO(image_bytes)).convert("RGB")
+    inputs = processor(images=img, return_tensors="np")
+    pixel_values = inputs["pixel_values"].astype(np.float32)
+
+    t0 = time.perf_counter()
+    outputs = session.run(["logits"], {"pixel_values": pixel_values})
+    elapsed = (time.perf_counter() - t0) * 1000
+
+    logits = outputs[0][0]
+    predicted_class_id = int(np.argmax(logits))
+    
+    return {
+        "label": cfg.id2label[predicted_class_id],
+        "score": float(np.exp(logits[predicted_class_id]) / np.sum(np.exp(logits))),
+        "label_id": predicted_class_id,
+        "inference_time_ms": elapsed
+    }

app/schemas.py

@@ -0,0 +1,11 @@
+from pydantic import BaseModel
+
+class PredictionResponse(BaseModel):
+    label: str
+    score: float
+    label_id: int
+    inference_time_ms: float
+
+class ErrorResponse(BaseModel):
+    detail: str
+    error_code: str

models/resnet18.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67045d45797b2be759e40bb17417045699bab836398694cdf7a2b5ff144d7ab6
+size 180730

models/resnet18.onnx.data

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:30b1aa4bdeba9751b30d868ef749ccb8c7fcb1be491e854eeca3f52369a00267
+size 46792704

models/resnet18_quantized.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5dc64fbf9cf470ed1db8c391577040069500cce3a1b7139a9194a07a0c98d63d
+size 11828572

models/resnet18_temp.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:108be9f23944c1ee1a0c9e97d97ac27550c06eb468a9cf1265cdc4a8a686be43
+size 46916000

requirements.txt

@@ -0,0 +1,13 @@
+fastapi
+uvicorn[standard]
+python-multipart
+onnxruntime
+numpy
+Pillow
+transformers
+torch
+torchvision
+pydantic
+pytest
+httpx
+onnx