FJDaz/spinoza-inference-experiment

fix: remove model bake step + pytorch CUDA base + clean auto-installer

Optimize: Full RunPod Serverless Compatibility & Input Harmonization

Fix: Force FastAPI initialization and root route

Improve: Non-blocking model loading for HF health checks

Fix: Add root route for HF Spaces to resolve 404

debug: add immediate logging and force OMP at runtime

fix: hard-force OMP_NUM_THREADS in both Dockerfile and python script

fix: set OMP_NUM_THREADS to prevent build hanging on libgomp error

perf: use snapshot_download to avoid RAM usage during build

perf: optimize memory usage for 15GB GPUs and auto-detect HF environment

fix: upgrade filelock to avoid TypeError during model download

refactor: use custom secret name 'inference_space' for HF auth

fix: make login non-fatal for debugging

debug: add token source and masking to download script

fix: handle HF login inside python script to avoid shell path issues

docs: update deployment guide with push instructions

feat: implement hybrid inference with remote fallback and fix HF build login

fix(docker): Use python module to run huggingface-cli

fix: Use build-time secret for downloading models

docs: Add Hugging Face Space README.md

fix: Configure port 7860 for Hugging Face Spaces

feat: Initial setup for multi-platform inference service