Upload folder using huggingface_hub

Dockerfile

@@ -2,12 +2,15 @@ FROM python:3.12-slim
 
 # Non-root user (HF Spaces convention; writable HOME for the HF cache the model
 # is pulled into at boot).
+# GLINER_MOCK=0: the deployed Space runs live inference (the local dev kit
+# defaults to mock; the env var flips only this image).
 RUN useradd -m -u 1000 user
 USER user
 ENV HOME=/home/user \
     PATH=/home/user/.local/bin:$PATH \
     HF_HOME=/home/user/.cache/huggingface \
-    PYTHONUNBUFFERED=1
+    PYTHONUNBUFFERED=1 \
+    GLINER_MOCK=0
 
 WORKDIR /home/user/app
 
@@ -20,16 +23,14 @@ USER user
 # model; the mock path works on flask alone.
 RUN pip install --no-cache-dir --user torch --index-url https://download.pytorch.org/whl/cpu
 RUN pip install --no-cache-dir --user flask gunicorn protobuf
-# GLiNER2 from testg2 branch (required for boundary architecture support)
-RUN pip install --no-cache-dir --user "gliner2[local] @ git+https://github.com/fastino-ai/GLiNER2.git@testg2"
+# GLiNER2 from main branch (required for boundary architecture support)
+RUN pip install --no-cache-dir --user "gliner2[local] @ git+https://github.com/fastino-ai/GLiNER2.git@main"
 
 COPY --chown=user . .
 
-# Pre-download model at build time to speed up cold starts (requires HF_TOKEN
-# as a Space secret for the private model repo).
-ARG HF_TOKEN
-ENV HF_TOKEN=${HF_TOKEN}
-RUN python -c "import os; os.environ.setdefault('HF_TOKEN', os.environ.get('HF_TOKEN', '')); from huggingface_hub import snapshot_download; snapshot_download('fastino/gliner2.5-multi-v1')" || echo "model pre-download skipped (no HF_TOKEN)"
+# Pre-download model at build time to speed up cold starts. The model repo is
+# public on the HF hub, so no token is needed.
+RUN python -c "from huggingface_hub import snapshot_download; snapshot_download('fastino/gliner2.5-multi-v1')" || echo "model pre-download skipped"
 
 EXPOSE 7860