Nicebdsmer/tribe-v2-api

fix: comprehensive CPU patches for ALL extractors (video + audio + text)

add version tag to index endpoint for deploy tracking

bump rebuild label to force Docker cache invalidation

fix: skip audio extraction on CPU (audio extractor needs CUDA)

fix: use proper Video events with filepath + ExtractAudioFromVideo transform

fix: use official TribeModel.get_events_dataframe() pipeline

Add /predict_raw endpoint - full prediction tensor output

Make video inference CPU-safe on Spaces

Make image inference CPU-safe

Fix image endpoint: use WAV audio + silent nullsrc

Fix: add timeline field to Audio events

Add /predict_image endpoint - converts image to video for brain scoring

Add packages.txt to force rebuild with new app

Add video/audio endpoints - bypass Llama tokenizer gate

Fix README with proper YAML frontmatter

Trigger rebuild to pick up HF_TOKEN secret

Update README with proper description

Upload requirements

Upload Flask API app with BFloat16 fix

Add Dockerfile for Docker-based deployment

initial commit