Fix indentation errors in retry and partial load paths to resolve runtime IndentationError e1b81f0 Sidoineko commited on 9 days ago
Direct load: force explicit cache_dir to HF_HOME (/home/user/.cache/huggingface) to avoid /data permissions 9b0e41a Sidoineko commited on 9 days ago
Add multiple loading approaches: direct HF Hub, /app/cache, /tmp/hf_home, /tmp/model_repo - to handle permission issues 7d8defe Sidoineko commited on 9 days ago
Force local-only model loading: always snapshot to /tmp/model_repo; remove direct HF cache loads; redirect HF cache to /tmp; disable TRANSFORMERS_CACHE a754003 Sidoineko commited on 9 days ago
Fix infinite loops and multiple processes: add rate limiting, cooldowns, and proper cleanup to prevent system overload 4c6302b Sidoineko commited on 9 days ago
Add complete automated recovery system: auto-detect partial loads, stuck states, and recover automatically without manual intervention 5d18e68 Sidoineko commited on 9 days ago
Add partial load recovery method and endpoint to handle interrupted model loading a02cc5c Sidoineko commited on 9 days ago
Add robust multiprocessing startup and persistent state management to handle Hugging Face Spaces interruptions d841385 Sidoineko commited on 9 days ago
Fix model ID to google/gemma-3n-E4B-it and optimize loading strategy: try cache first, then snapshot as fallback cf8e6a2 Sidoineko commited on 9 days ago
Add persistent model loader and retry mechanism to handle Hugging Face Spaces interruptions 9534306 Sidoineko commited on 9 days ago
Fix indentation issues and add robust middleware for model loading to prevent CancelledError a84214d Sidoineko commited on 9 days ago
Fix startup warmup: use threading instead of asyncio to prevent CancelledError during model loading 1c9e2cf Sidoineko commited on 9 days ago
API hardening: auto-fallback HF_HOME to /tmp when unwritable; disable hf_transfer if missing eda4143 Sidoineko commited on 10 days ago
Fix model loading with better error handling and memory management 86532db Sidoineko commited on 10 days ago
API: snapshot_download + local_files_only + warmup; robust retries for model load 6e9e3e0 Sidoineko commited on 11 days ago
Fix: Optimize FastAPI for Hugging Face Spaces - Force CPU mode, add logging, fix Dockerfile 5c88d83 Sidoineko commited on 23 days ago
Fix: multimodal inference with apply_chat_template + image preprocessing; align with Streamlit app 151f468 Sidoineko commited on 24 days ago
Update FastAPI to work with Streamlit model cache - API now detects when Streamlit has loaded the model and can share model information a5debd6 Sidoineko commited on 24 days ago
Add supervisord + Dockerfile for Streamlit+FastAPI; ignore .env d7dbf7e Sidoineko commited on 24 days ago