Update Dockerfile

Fix attention IO capture for keyword-only Llama modules

Set wave tank training to 750 steps per layer

Fix dtype mismatch during wave tank replacement training

Expose workload traceback in Space health endpoint

Use provided HF token explicitly for download and push

Fix Space health timeout by serving port 7860 during long run

Fix HF cache permission: use /app/.cache instead of /.cache

Update Dockerfile

Update app.py

Update app.py

Update Dockerfile

Update wave_tank_surgeon.py

Update app.py

Increase training steps to 1000 (script converges ~200-300 per layer, was only 100)

Trigger rebuild with clean Dockerfile (no wget)

Fix: use python3.10 (available in Ubuntu 22.04)

Add wave tank surgeon script

Add Dockerfile for auto-run wave tank surgeon

Upload requirements.txt with huggingface_hub

Upload app.py with huggingface_hub

initial commit