UnityAILab/If-Only-I-Had-A-Brain
4
0
Fork 0
Code Issues Pull requests Projects Releases 39 Packages Wiki Activity Actions
39 releases 43 tags
  • donor-v0.3.6 5900fdc7d0

    donor-v0.3.6
    All checks were successful
    deploy / deploy (push) Successful in 4s
    Details
    donor-release / build-and-release (push) Successful in 5m59s
    Details
    Stable

    Sponge released this 2026-07-05 06:41:03 +00:00 | 3222 commits to main since this release

    donor-v0.3.6 — idle-watchdog fix + light theme + headless GUI. The native donor no longer strangles a healthy link during heavy teach: IDLE_TIMEOUT 45s→150s, so a coordinator that queues its pong past 45s mid-teach no longer gets reconnect-cycled (which dropped the brain replica and forced a full matrix re-upload every ~60-90s). Also ships the OS light/white high-contrast theme (readable, no more grey-on-white) and a headless GUI launch on Windows (no console window behind the app; the pure-headless CLI build keeps its console). Linux + Windows x86-64 (GUI+CUDA; CUDA dynamic-loaded with automatic WebGPU fallback on AMD/Intel/Apple). Same brain, same results. Browser-tab donors are unaffected — this fix is native-binary only.

    Downloads
  • donor-v0.3.5 fc482cb45d

    donor-v0.3.5
    All checks were successful
    deploy / deploy (push) Successful in 3s
    Details
    Stable

    Sponge released this 2026-06-30 19:46:25 +00:00 | 3254 commits to main since this release

    donor-v0.3.5 — work-stealing 'mining' donor model. High-RTT / Starlink donors now carry real compute batches: server-side work-eligibility floor + completion-driven work-stealing dispatch + replica-sync pacing so a slow uplink isn't flooded. Adds a self-measured downlink hint (linkDownMbps) so the brain paces sync to your link. Linux + Windows x86-64 (GUI+CUDA; CUDA with automatic WebGPU fallback on AMD/Intel/Apple). Same brain, same results.

    Downloads
  • donor-v0.3.4 ba3c0c8f93

    donor-v0.3.4 — donated-capacity auto-scale
    All checks were successful
    deploy / deploy (push) Successful in 4s
    Details
    Pre-release

    Sponge released this 2026-06-28 22:48:43 +00:00 | 3276 commits to main since this release

    Native GPU donor v0.3.4 — auto-scale now counts DONATED capacity, not full card VRAM.

    Builds on v0.3.3 (flap-resistant keepalive + platform telemetry) and adds:

    • The donor now reports its donation in gpu_register: utilizationPct (duty-cycle) + donatedMB (explicit --memory VRAM cap). The brain weights community-compute auto-scale by EFFECTIVE donated capacity, so e.g. two 15GB cards set to 60% count as 18GB, not 30GB — no false tier-up.
    • Honors the data-parallel caveat: utilization is a throughput duty-cycle, not VRAM held; the admin /autoscale + dashboard now expose minDonorMB (the smallest donor's committed VRAM = the real brain-SIZE bound).

    Downloads: Windows (.exe) + Linux (x86-64). NVIDIA = CUDA; auto-falls-back to WebGPU on AMD/Intel/Apple. WIP.

    Downloads
  • donor-v0.3.3 d467d10f52

    donor-v0.3.3 — flap-resistant keepalive + platform telemetry
    All checks were successful
    deploy / deploy (push) Successful in 4s
    Details
    Pre-release

    Sponge released this 2026-06-28 21:43:52 +00:00 | 3285 commits to main since this release

    Native GPU donor v0.3.3 — connection-flap resistance + platform telemetry.

    Highlights:

    • Client-initiated WS keepalive (15s) so the donor link never goes idle long enough for CGNAT / a reverse proxy to reap it — the fix for donors that 'keep reconnecting' / show red / 0 Gn/s on a flaky (e.g. Starlink) link.
    • Fast dead-link detection (45s) -> proactive reconnect instead of waiting minutes for the RST.
    • Jittered per-install reconnect backoff (no lockstep rejoins).
    • LOUD CUDA PTX/kernel load-failure logging (no more silent-0).
    • Reports osPlatform / engineBackend (cuda/vulkan/dx12) / driverVersion / computeCapability — visible in the admin dashboard Clients table.

    Downloads: Windows (.exe) + Linux (x86-64). NVIDIA = CUDA (full VRAM, best perf); auto-falls-back to WebGPU on AMD/Intel/Apple. WIP.

    Downloads
  • donor-v0.3.2 37c57b25da

    donor-v0.3.2
    All checks were successful
    deploy / deploy (push) Successful in 7s
    Details
    Stable

    Sponge released this 2026-06-28 18:24:18 +00:00 | 3330 commits to main since this release

    Donor app v0.3.2 — Light / Dark / System theme toggle in Settings (default follows the OS; manually settable; readable in both). All settings now persist on this machine (theme, name, server, per-GPU enable/%, auto-reconnect) — load up and donate, no reconfiguration. Windows .exe (no console) + Linux x86-64.

    Downloads
  • donor-v0.3.1 9b5d455ce1

    donor-v0.3.1
    All checks were successful
    deploy / deploy (push) Successful in 7s
    Details
    Stable

    Sponge released this 2026-06-28 18:14:42 +00:00 | 3333 commits to main since this release

    Donor app v0.3.1 — light/white high-contrast GUI theme (PD.2) + no console window on Windows (PD.3, GUI subsystem). Rebuilt from PUBDASH source. Downloads: Windows .exe + Linux x86-64. Auto-reconnect default ON.

    Downloads
  • donor-v0.3.0 68705eb0ab

    donor-v0.3.0 Pre-release

    Sponge released this 2026-06-25 05:46:59 +00:00 | 3390 commits to main since this release

    Unity Brain Donor v0.3.0 — native GPU compute donor.

    GUI overhaul (DA.1-DA.13): tabbed 'Unity Brain Donor' app — Donate / Settings / Dashboard / Leaderboard / About. 1280x720 resizable, app identity + brain+GPU icon, Live/Local/Custom server radio, verbose status, green/red Start-Stop, dark organic-tech theme. Headless --name flag unchanged (tags compute under a name; defaults anonymous).

    Assets:

    • unity-donor-windows-x86_64.exe (Windows x86-64)
    • unity-donor-linux-x86_64 (Linux x86-64)

    Built from develop @ donor-app v0.3.0; both targets compile-verified (cargo build --release native + x86_64-pc-windows-gnu).

    Downloads
  • donor-v0.2.0 6256842a82

    Native GPU Donor v0.2.0
    All checks were successful
    deploy / deploy (push) Successful in 6s
    Details
    Pre-release

    Sponge released this 2026-06-24 08:40:20 +00:00 | 3411 commits to main since this release

    Native GPU compute donor for the Unity brain — v0.2.0.

    NEW in v0.2.0: auto-reconnect supervisor (a dropped/closed connection now rejoins on its own with backoff — no more going dark until you press Start), a GUI/CLI toggle (--no-reconnect to opt out), and a legend link.

    Carries everything from v0.1.0: CUDA backend for NVIDIA (no 2 GB binding cap) with automatic WebGPU fallback on AMD/Intel/Apple; multi-GPU-as-one-host; per-GPU utilization; optional leaderboard name (same name across devices aggregates your compute). Connects to production by default; --local for a local brain.

    GPU support: NVIDIA → CUDA; everything else → WebGPU. ⚠ WIP — functional but may have rough edges.

    Downloads
  • donor-v0.1.0 132b2bc238

    Native GPU Donor v0.1.0 (WIP)
    All checks were successful
    deploy / deploy (push) Successful in 7s
    Details
    Pre-release

    Sponge released this 2026-06-23 05:53:45 +00:00 | 3441 commits to main since this release

    Native GPU compute donor for the Unity brain — a compiled, cross-platform app that donates GPU compute (like the in-browser compute.html, but native and more powerful).

    ⚠️ WORK IN PROGRESS — functional but may have rough edges. Please report issues.

    GPU support: built for NVIDIA (CUDA backend — no 2 GB binding cap, full VRAM, best performance). Automatically falls back to WebGPU on AMD / Intel / Apple GPUs — same brain, same results.

    Default: connects to the production brain out of the box. Use --local for a local brain, or --server to override. GUI: run it, pick GPUs, press Start. Headless: --headless --autostart --gpus 0,1 --utilization 10.

    Tip: avoid heavy compute on your primary display GPU (can lag the desktop) — the GUI defaults to a non-display card.

    Downloads