mnt bucket sync p1

.gitignore

@@ -1,4 +1,5 @@
 .DS_Store
+mnt/
 .venv/
 venv/
 __pycache__/

app.py

@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import base64
 import html
 import json
 import os
@@ -12,14 +13,56 @@ from pathlib import Path
 
 import gradio as gr
 
-IMAGE_CDN_BASE = (
-    "https://huggingface.co/DTanzillo/one-haut-encoded/resolve/main/data/images"
-)
+# Default: HF dataset bucket with resized demo images. Override for local testing:
+#   export ONEHAUT_IMAGE_CDN_BASE=https://huggingface.co/datasets/USER/REPO/resolve/main/images
+def _image_cdn_base() -> str:
+    return os.environ.get(
+        "ONEHAUT_IMAGE_CDN_BASE",
+        "https://huggingface.co/datasets/alexoh2020/onehautapp-storage/resolve/main/images",
+    ).strip().rstrip("/")
+
+
+def _mnt_data_roots() -> list[Path]:
+    """HF Space persistent volume is /mnt/data; repo clone uses onehautapp/mnt/data."""
+    extra = os.environ.get("ONEHAUT_MNT_DATA", "").strip()
+    roots: list[Path] = []
+    if extra:
+        roots.append(Path(extra))
+    roots.append(Path("/mnt/data"))
+    roots.append(Path(__file__).resolve().parent / "mnt/data")
+    return roots
+
+
+def _local_image_file(article_id: str) -> Path | None:
+    rel = Path("images") / article_id[:3] / f"{article_id}.jpg"
+    for root in _mnt_data_roots():
+        p = (root / rel).resolve()
+        if p.is_file():
+            return p
+    return None
+
+
+# Max bytes to inline as data: URL in HTML (avoid huge payloads)
+_MAX_INLINE_IMAGE_BYTES = 512 * 1024
 
 MODEL_API = {"popularity": "popularity", "knn": "knn", "ncf": "ncf"}
 MODEL_STATIC_KEY = {"popularity": "popularity", "knn": "knn", "ncf": "ncf_meta"}
 
-_DATA_PATH = Path(__file__).resolve().parent / "data/processed/precomputed_recommendations.json"
+
+def _precomputed_path() -> Path:
+    """Prefer Space /mnt/metadata, then app bundle mnt/metadata, then data/processed."""
+    candidates = [
+        Path("/mnt/metadata/precomputed_recommendations.json"),
+        Path(__file__).resolve().parent / "mnt/metadata/precomputed_recommendations.json",
+        Path(__file__).resolve().parent / "data/processed/precomputed_recommendations.json",
+    ]
+    for p in candidates:
+        if p.is_file():
+            return p
+    return candidates[1]
+
+
+_DATA_PATH = _precomputed_path()
 with open(_DATA_PATH, encoding="utf-8") as f:
     _RAW_USERS: list[dict] = json.load(f)
 _USER_BY_ID: dict[str, dict] = {u["customer_id"]: u for u in _RAW_USERS}
@@ -30,8 +73,21 @@ def get_api_base() -> str:
 
 
 def image_url(article_id: object) -> str:
+    """Prefer JPEG under /mnt/data (Space) or onehautapp/mnt/data; else HF CDN.
+
+    Local files are inlined as data: URLs so gr.HTML img tags work without extra routes.
+    """
     aid = str(article_id)
-    return f"{IMAGE_CDN_BASE}/{aid[:3]}/{aid}.jpg"
+    local = _local_image_file(aid)
+    if local is not None:
+        try:
+            raw = local.read_bytes()
+            if len(raw) <= _MAX_INLINE_IMAGE_BYTES:
+                b64 = base64.b64encode(raw).decode("ascii")
+                return f"data:image/jpeg;base64,{b64}"
+        except OSError:
+            pass
+    return f"{_image_cdn_base()}/{aid[:3]}/{aid}.jpg"
 
 
 def normalize_rec_response(data: object) -> list[dict]:
@@ -209,4 +265,5 @@ def build_ui():
 demo = build_ui()
 
 if __name__ == "__main__":
-    demo.queue().launch()
+    # Allow Gradio to resolve file paths under /mnt if needed (e.g. future file= URLs)
+    demo.queue().launch(allowed_paths=["/mnt"])