Add null output mode to LingBot replay demo

[jarcherNV]

No description provided.

[greptile-apps]

Greptile Summary

The PR adds null-output replay support and GPU runtime validation for the LingBot demo while consolidating shared demo, keyboard-input, and WebRTC infrastructure.

• Adds null and MP4 LingBot replay coverage in a dedicated GPU workflow.
• Streams safetensors checkpoints directly into model parameters to reduce host-memory and disk requirements.
• Moves transport-neutral keyboard state into the runtime package and model-specific control processing into the integrations.
• Refactors shared demo bootstrap, timing, WebRTC services, and integration adapters.

Confidence Score: 5/5

The PR appears safe to merge because no blocking failure remains within the eligible follow-up review scope.

No blocking failure remains.

Important Files Changed

Filename	Overview
.github/workflows/lingbot-demo-runtime.yml	Adds end-to-end GPU validation for LingBot null and MP4 replay modes, including checkpoint-streaming and artifact checks.
flashdreams/flashdreams/core/checkpoint/load.py	Adds bounded safetensors loading directly into model state and parallel shard resolution.
flashdreams/flashdreams/runtime/demo/timing.py	Replaces keyboard-specific realtime timing with a transport-neutral event timeline and resampler.
flashdreams/flashdreams/runtime/keyboard.py	Centralizes keyboard normalization, state resolution, and sparse input request containers.
flashdreams/flashdreams/serving/webrtc/manager.py	Adapts shared WebRTC session management to the refactored realtime input and service contracts.
integrations/lingbot/lingbot/controls.py	Introduces LingBot-specific realtime control sampling and input-window interpretation.
integrations/lingbot/lingbot/demo/app.py	Exposes the new null-output replay mode through the LingBot demo application.
integrations/omnidreams/omnidreams/demo/controls.py	Moves OmniDreams-specific pose integration and keyboard resampling behind its integration boundary.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  Replay["LingBot replay command"] --> Runtime["Unified inference runtime"]
  Runtime --> Checkpoint["Stream checkpoint into model"]
  Runtime --> Inputs["Replay or realtime input provider"]
  Runtime --> Output{"Output mode"}
  Output --> Null["Null sink"]
  Output --> MP4["MP4 sink"]
  Output --> WebRTC["WebRTC service"]

Loading

_{Reviews (8): Last reviewed commit: "Advance OmniDreams Ludus replay keyboard..." | Re-trigger Greptile}