Spaces:
Running on Zero
Running on Zero
Commit History
-
demo: re-curate scenarios (first 3 per system ticket+issue, blockers demoted)
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>b93b071
-
scenarios: lead each system with iconic disruption (mirrors monorepo)
First card per system is now the locally-distinctive disruption (hurricane / sandstorm / earthquake / typhoon / polar vortex / smog) so the landing pulse cue lands on something that immediately shows the agent earning its keep. Clean baselines moved to last. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>15620b4 -
revert landing replay; add first-scenario pulse cue instead
Mirrors monorepo. Replay was fighting Gradio's tab/queue model. Replaced with a one-shot CSS pulse on the first scenario card, killed by a click delegate on first user interaction. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>9cda05e
-
landing replays for all 6 systems + render helpers
Mirrors monorepo d8176c6. data/replays/{system}.json (~600 KB total) auto-play on landing, picked by IANA timezone. app.py imports rendering helpers from render.py (new) so capture tooling can format identical traces without triggering the model load. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>d23bf95
-
flash-attn for ZeroGPU + regional scenario flavor + subtitle
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>32606be
-
init: MetroLLM-Bench kiosk demo (2B + LoRA, free CPU)
Buttons-only Gradio UI driving a Qwen3.5-2B + continker/Qwen3.5-2B-metro-v23 LoRA adapter on free CPU Basic. In-process FastAPI mock_server for the 7-tool kiosk loop; verify_data.json drives the Leaflet map.cbfaae5