Fix: skip VQ-VAE if checkpoint exists, reduce epochs to 3

Fix: efficient log reading (last 5000 chars only)

Updated: scaled pipeline with real data (10K images, 5K LLM samples, checkpoint/resume support)

Updated: scaled pipeline with real data (10K images, 5K LLM samples, checkpoint/resume support)

Fix: try multiple COCO datasets with fallbacks (HuggingFaceM4/COCO no longer works)

Fix README YAML

Update requirements.txt: full training pipeline with real datasets

Update train_full_pipeline.py: full training pipeline with real datasets

Update app.py: full training pipeline with real datasets

Add constrained decoding: force visual tokens after <video_start>

Rewrite: generation-only app, preload models, no auto-training

Update requirements.txt: add video generation tab

Update app.py: add video generation tab

Add VQ-VAE checkpoint for video generation

Fix: remove every param from demo.load

Update train_on_hf_spaces.py: auto-start training, file-based logging

Update app.py: auto-start training, file-based logging

Update train_on_hf_spaces.py: better error handling, show_error=True

Update app.py: better error handling, show_error=True

Update Gradio SDK to 5.31.0, Python 3.11

Fix requirements.txt

Fix README.md

Fix README short_description

Upload requirements.txt

Upload tokenized_dataset.json

Upload train_on_hf_spaces.py

Upload app.py

initial commit