Strata v0.1.10

No more stuck servers after an engine crash, a clear warning when RAM is tight, and a new front page.

• Fixed: answers stopping early and every later request failing (#27, thanks @ast321). When the engine process died mid-answer - on Linux this is the system's out-of-memory killer, when the model's experts leave too little RAM for everything else - the server recorded the reply as "Max tokens", cut the stream, and then failed on every request after it. Now:
  • the reply ends with a clear error ("the engine stopped unexpectedly"), and the Monitor shows it as an error;
  • the next request starts the engine again by itself (a minute or two), instead of failing forever;
  • the server window explains the likely cause (RAM) and how to check it on Linux (sudo dmesg | grep -i -E 'killed process|out of memory');
  • at start, the server warns when the model's experts leave less than ~6 GB of RAM for the system and your other programs.
• A new front page: a video of Strata's model at work, the speed and "which model?" first (with the size of shard 1, the part that has to fit in your RAM), and a simpler "Something went wrong?" section, including what to do if the PC freezes at the first start.

Updating: get the latest files (git pull or download the repo again), then run START-HERE.bat with the model window closed. It updates the engine to 0.1.10 by itself. On Linux, ./setup.sh builds the engine from source as before.

The ready-made Strata engine for Windows (RTX 30 / 40 / 50: sm_86, sm_89, sm_120 + PTX), CUDA 13.0.

You don't need to download this yourself: START-HERE.bat fetches it (and NVIDIA's cuBLAS from pip), so no compiler or CUDA Toolkit is needed, only an NVIDIA driver 580 or newer.

Contents: strata-windows-x64.zip - strata.exe (the engine), strata-vision.exe (the optional image encoder), BUILD.json. Pagoda.mp4 is the front page's demo video (49 s).