feat: add startup script for model downloads and upload script for Hugging Face Hub

Dockerfile

@@ -37,10 +37,15 @@ COPY requirements.txt .
 
 # PyTorch CPU versiyonu + diğer paketler
 RUN pip install --no-cache-dir torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu \
-    && pip install --no-cache-dir -r requirements.txt
+    && pip install --no-cache-dir -r requirements.txt \
+    && pip install --no-cache-dir "huggingface-hub>=0.20.0"
 
 # Uygulama kodu
 COPY app ./app
+COPY startup.py .
+
+# Models dizini — startup.py çalışınca doldurulacak
+RUN mkdir -p /app/models
 
 # Hugging Face Spaces için non-root kullanıcı (güvenlik)
 RUN useradd -m -u 1000 user \
@@ -48,14 +53,16 @@ RUN useradd -m -u 1000 user \
 USER user
 
 ENV HOME=/home/user \
-    PATH=/home/user/.local/bin:$PATH
+    PATH=/home/user/.local/bin:$PATH \
+    AURIS_MODELS_REPO=Rtur2003/auris-models \
+    MODELS_DIR=/app/models
 
 # Hugging Face Spaces varsayılan port: 7860
 EXPOSE 7860
 
-# Health check
-HEALTHCHECK --interval=30s --timeout=10s --start-period=5s --retries=3 \
+# Health check — model download ~2-3 dk sürebilir, start-period uzun
+HEALTHCHECK --interval=30s --timeout=10s --start-period=180s --retries=5 \
     CMD curl -f http://localhost:7860/api/health || exit 1
 
-# Başlat
-CMD ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "7860"]
+# Başlat: önce modelleri indir, sonra API'yi ayağa kaldır
+CMD ["sh", "-c", "python startup.py && uvicorn app.main:app --host 0.0.0.0 --port 7860"]

startup.py

@@ -0,0 +1,141 @@
+"""
+AURIS startup script — downloads model artifacts from HuggingFace Hub
+before the API server starts.
+
+Usage (in Dockerfile CMD):
+    python startup.py && uvicorn app.main:app --host 0.0.0.0 --port 7860
+
+Environment variables:
+    HF_TOKEN          — HuggingFace access token (optional for public repos)
+    AURIS_MODELS_REPO — HF repo ID, default: Rtur2003/auris-models
+    MODELS_DIR        — Local destination, default: /app/models
+    SKIP_MODEL_DOWNLOAD — Set to "1" to skip (for local dev)
+"""
+
+from __future__ import annotations
+
+import os
+import sys
+import time
+from pathlib import Path
+
+REPO_ID   = os.getenv("AURIS_MODELS_REPO", "Rtur2003/auris-models")
+MODELS_DIR = Path(os.getenv("MODELS_DIR", "/app/models"))
+HF_TOKEN  = os.getenv("HF_TOKEN") or os.getenv("HUGGING_FACE_HUB_TOKEN")
+SKIP      = os.getenv("SKIP_MODEL_DOWNLOAD", "0") == "1"
+
+# Files that must exist for the API to work
+REQUIRED_FILES = [
+    "auris_classifier_v1.pkl",
+    "feature_scaler_v1.pkl",
+    "feature_columns_v1.json",
+    "feature_stats_v1.json",
+    "training_results.json",
+    "deep_learning_results.json",
+    "model_lightgbm.pkl",
+    "model_xgboost.pkl",
+    "model_random_forest.pkl",
+    "model_gradient_boosting.pkl",
+    "model_svm_rbf.pkl",
+    "model_mlp_neural_network.pkl",
+    "model_logistic_regression.pkl",
+    "model_dl_deep_mlp_512_256_128_64.pkl",
+    "model_dl_1d_cnn.pkl",
+    "model_dl_residual_mlp_3_blocks.pkl",
+    "model_dl_attention_mlp.pkl",
+    "wav2vec2_auris_v1.pt",
+]
+
+# Large files that are optional (wav2vec2 tower works without wav2vec2)
+OPTIONAL_FILES = {
+    "wav2vec2_auris_v1.pt",
+}
+
+
+def _already_downloaded() -> bool:
+    """Return True if all required non-optional files already exist."""
+    missing = [
+        f for f in REQUIRED_FILES
+        if f not in OPTIONAL_FILES and not (MODELS_DIR / f).exists()
+    ]
+    if missing:
+        print(f"[startup] Missing files: {missing}")
+        return False
+    return True
+
+
+def download_models() -> None:
+    if SKIP:
+        print("[startup] SKIP_MODEL_DOWNLOAD=1 — skipping download.")
+        return
+
+    MODELS_DIR.mkdir(parents=True, exist_ok=True)
+
+    if _already_downloaded():
+        print("[startup] All required model files already present — skipping download.")
+        return
+
+    print(f"[startup] Downloading models from {REPO_ID} → {MODELS_DIR}")
+    t0 = time.time()
+
+    try:
+        from huggingface_hub import hf_hub_download, list_repo_files
+    except ImportError:
+        print("[startup] huggingface_hub not installed — pip install huggingface-hub")
+        sys.exit(1)
+
+    kwargs = {"repo_id": REPO_ID, "repo_type": "model"}
+    if HF_TOKEN:
+        kwargs["token"] = HF_TOKEN
+
+    # Get list of files in the repo
+    try:
+        repo_files = list(list_repo_files(**{k: v for k, v in kwargs.items() if k != "token"},
+                                          token=HF_TOKEN))
+    except Exception as e:
+        print(f"[startup] Cannot list repo files: {e}")
+        print("[startup] Trying to download known files directly...")
+        repo_files = REQUIRED_FILES
+
+    errors: list[str] = []
+    for filename in REQUIRED_FILES:
+        dest = MODELS_DIR / filename
+        if dest.exists():
+            print(f"[startup]   skip  {filename} (exists)")
+            continue
+
+        if filename not in repo_files and filename not in REQUIRED_FILES:
+            if filename in OPTIONAL_FILES:
+                print(f"[startup]   skip  {filename} (not in repo, optional)")
+                continue
+
+        is_optional = filename in OPTIONAL_FILES
+        try:
+            print(f"[startup]   dl    {filename} ...", end=" ", flush=True)
+            path = hf_hub_download(
+                filename=filename,
+                local_dir=str(MODELS_DIR),
+                **kwargs,
+            )
+            print(f"OK ({Path(path).stat().st_size / 1024 / 1024:.1f} MB)")
+        except Exception as e:
+            if is_optional:
+                print(f"SKIP (optional: {e})")
+            else:
+                print(f"ERROR: {e}")
+                errors.append(f"{filename}: {e}")
+
+    elapsed = time.time() - t0
+    print(f"[startup] Download complete in {elapsed:.1f}s")
+
+    if errors:
+        print(f"[startup] FATAL — {len(errors)} required file(s) failed:")
+        for err in errors:
+            print(f"  - {err}")
+        print("[startup] Set AURIS_MODELS_REPO and HF_TOKEN env vars if repo is private.")
+        sys.exit(1)
+
+
+if __name__ == "__main__":
+    download_models()
+    print("[startup] Ready.")

upload_models_to_hub.py

@@ -0,0 +1,97 @@
+"""
+One-time script: upload all AURIS model artifacts to HuggingFace Hub.
+
+Usage:
+    python upload_models_to_hub.py
+
+Requires:
+    pip install huggingface-hub
+    huggingface-cli login   (or set HF_TOKEN env var)
+
+Creates / updates: Rtur2003/auris-models  (model repo, public)
+"""
+
+from __future__ import annotations
+
+import os
+import sys
+from pathlib import Path
+
+REPO_ID    = "Rtur2003/auris-models"
+MODELS_DIR = Path(__file__).parent / "models"
+HF_TOKEN   = os.getenv("HF_TOKEN") or os.getenv("HUGGING_FACE_HUB_TOKEN")
+
+FILES_TO_UPLOAD = [
+    # Core classifier
+    "auris_classifier_v1.pkl",
+    "feature_scaler_v1.pkl",
+    "feature_columns_v1.json",
+    "feature_stats_v1.json",
+    "training_results.json",
+    "deep_learning_results.json",
+    # ML models
+    "model_logistic_regression.pkl",
+    "model_random_forest.pkl",
+    "model_gradient_boosting.pkl",
+    "model_svm_rbf.pkl",
+    "model_mlp_neural_network.pkl",
+    "model_xgboost.pkl",
+    "model_lightgbm.pkl",
+    # DL models
+    "model_dl_deep_mlp_512_256_128_64.pkl",
+    "model_dl_1d_cnn.pkl",
+    "model_dl_residual_mlp_3_blocks.pkl",
+    "model_dl_attention_mlp.pkl",
+    # wav2vec2 transformer
+    "wav2vec2_auris_v1.pt",
+]
+
+
+def main() -> None:
+    try:
+        from huggingface_hub import HfApi, create_repo
+    except ImportError:
+        print("ERROR: pip install huggingface-hub")
+        sys.exit(1)
+
+    api = HfApi(token=HF_TOKEN)
+
+    # Create repo if it doesn't exist
+    try:
+        create_repo(REPO_ID, repo_type="model", exist_ok=True, token=HF_TOKEN)
+        print(f"Repo ready: https://huggingface.co/{REPO_ID}")
+    except Exception as e:
+        print(f"WARNING: could not create repo: {e}")
+
+    errors: list[str] = []
+    for filename in FILES_TO_UPLOAD:
+        src = MODELS_DIR / filename
+        if not src.exists():
+            print(f"  SKIP  {filename} (not found locally)")
+            continue
+
+        size_mb = src.stat().st_size / 1024 / 1024
+        print(f"  UP    {filename} ({size_mb:.1f} MB) ...", end=" ", flush=True)
+        try:
+            api.upload_file(
+                path_or_fileobj=str(src),
+                path_in_repo=filename,
+                repo_id=REPO_ID,
+                repo_type="model",
+            )
+            print("OK")
+        except Exception as e:
+            print(f"ERROR: {e}")
+            errors.append(f"{filename}: {e}")
+
+    if errors:
+        print(f"\n{len(errors)} upload(s) failed:")
+        for e in errors:
+            print(f"  - {e}")
+        sys.exit(1)
+    else:
+        print(f"\nAll files uploaded to https://huggingface.co/{REPO_ID}")
+
+
+if __name__ == "__main__":
+    main()