Strata v0.1.12

Fixed: generation stopping for good on big-VRAM cards, and no silent hangs any more.

• Fixed: generation could stop after thousands of tokens and never continue (#29, thanks @Lufen34). Seen on an RTX 4090: the GPU showed 100% but drew little power, one CPU core was busy, and no more tokens came. The CPU expert workers go to sleep when they have no work (since the fix for issue #4). With a large GPU most experts live in VRAM, so they also fell asleep in the middle of an answer. A worker waking late could then pick up work from the next batch while it was still being set up, and the engine could wait forever. Each piece of work is now tied to its batch, so a late worker can't touch the next one. Same speed as before.
• The engine can no longer hang silently. If a request stops making progress for 2 minutes, the engine stops with a line in its log saying where it was stuck. The server answers with an error, names the cause, and starts the engine again on the next request. If you ever see no progress for 120 s ... (issue #29) in strata-<model>.log, please open an issue with that line. (STRATA_WATCHDOG_S changes the 120 s; 0 turns it off.)
• Text files in the browser chat (#30, thanks @XeonG). Drop a text file (code, notes, logs, data; up to 512 KB) on the chat, or use the attach button, and it goes into your message.

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 0.1.12 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.exe (the engine), strata-vision.exe (the optional image encoder), BUILD.json.