L5: Beyond Text: Learning Across Modalities

REQUIREMENTS

Apple Silicon Mac (M1 or newer), macOS 14+, 16 GB RAM and 25 GB free disk.
Internet is needed for setup; demos run locally and can run offline afterward.
No account, API key or M3 server is needed. Close other memory-heavy apps and
stop any L3/L4 model helper before starting.

SET UP ONCE

Fully extract the instructor ZIP and open Terminal in that folder.
Use your normal user account, not sudo. If Homebrew is missing, install it
from https://brew.sh, follow its shell setup instructions and reopen Terminal.

Run these commands in order:

1. bash install_requirements.command
   If Firefox is newly installed, open it once and close its window.
2. bash start_lecture.command
3. bash stop_lecture.command
   These two steps initialize the lecture's Firefox profile.
4. bash lecture.deckroom/demos/install_l5_demos.command
   Accept dependency installation if prompted. Wait for "L5 setup complete".

Setup installs dependencies and downloads Qwen plus its vision projector,
Parakeet, FLUX and Magpie. Allow time for several GB of downloads.
Allow Camera for Firefox and Microphone for Lecture Subtitles when prompted.
If macOS blocks Lecture Subtitles, use System Settings > Privacy & Security >
Open Anyway, then rerun setup. Do not disable Gatekeeper.

START EACH SESSION

In the extracted release folder:
  bash start_lecture.command

In a second Terminal in the same folder:
  bash lecture.deckroom/demos/start_l5_demos.command

Wait for "All L5 helpers ready" and keep that Terminal open.
Use demo controls in Presenter view; Audience view mirrors the results.
The camera model loads on demand. Only one slide 1 demo is active at a time.

TA TESTS

- Slide 1: webcam questions, English transcription, image generation and speech
  generation. Check camera freeze/resume and enlarged images.
- Slide 5: camera selection, RGB pixel inspection and Default Image.
  Check that the image carries through to slides 6–7.
- Slide 8: microphone selection, Record, Stop, Play and Default Recording.
  Advance once to reveal the transcription and processing diagram.
- Slide 11: image editing with Cat, Car and Webcam. Check the instruction,
  result, matching aspect ratios and cancellation.
- Check Audience synchronization. Complete each model demo online once,
  restart the helpers, disconnect Wi-Fi and repeat, including image editing.

Slide 1 transcription uses the macOS default microphone (Sound > Input).
Slide 8 has its own microphone selector. Default Image and Default Recording
provide fallbacks; answering image questions still requires Qwen.

Report Mac model, RAM, macOS, pass/fail, exact errors, response delays and
any yellow/red memory pressure. Performance on a 16 GB Mac needs TA testing.

STOP / UPDATE / TROUBLESHOOT

Stop: turn off the demo, press Ctrl+C in the helper Terminal, then run
  bash stop_lecture.command
Quit Lecture Subtitles in Activity Monitor to release its model memory.

New package: stop old helpers and quit Lecture Subtitles, then run setup once
from the new package while online. Cached models are reused.

Permission issue: check System Settings > Privacy & Security.
Download or missing-model error: reconnect and rerun the L5 installer.
Qwen unsupported-model error: run brew upgrade llama.cpp and restart helpers.
Quiet audio: select the built-in microphone and check Sound > Input.
Port conflict: stop the old demo terminal; do not kill unrelated services.
Generation errors show a local log path. Share the error, not private media.

Keep model caches for offline use. Generated images/audio and logs remain in
~/.local/share/ailt_l5_generation/outputs. Camera/transcription demos do not save
captures or recordings to disk.

Custom Deckroom sessions: pass DECKROOM_LECTURE_FIREFOX_PROFILE and
DECKROOM_WEB_PORT to setup; defaults use the release's profile and port 2804.
