#6623 [GUI] Update engine smoke-test passes --engram; engine 0.1.65 rejects it ("unknown argument"), so every auto-update reports failure
open · @xglem2-eng · 0 comments · View on GitHub
Setup & installServer & APINVIDIA / CUDAModels & quantsWindows
Description
### Area [GUI] (the desktop app) ### Desktop app, Terminal or source? Desktop app ### What happened Two automatic Strata updates today (0.2.18 → 0.2.19, engine 0.1.64 → 0.1.65) each ended with: ``` [host] update · update-engine failed: The new engine 0.1.65 did not start correctly on this computer, so it was not installed. Your current engine is unchanged. The engine started but stopped before it answered a test question. Run `strata update --force` to try again; if it fails again, report it with the log: C:\Users\xglem\AppData\Local\Strata\components\.updates\logs\test-start-0.1.65.log ``` The named log shows why: the updater launches the engine with `--engram`, which this engine build does not know — its own usage text lists only `--ple-gguf` for the second shard. The engine prints usage and exits, so the smoke-test always fails: ``` [strata] 2026-11-22 16:30:55 engine started: C:\Users\xglem\AppData\Local\Strata\components\.s\4ce1\engine\0.1.65-windows-x86_64-cuda-cuda12\strata.exe --serve --native C:\Users\xglem\Documents\Strata\models\hy4-flash\Hy4-Flash-SQ-64.strata --engram C:\Users\xglem\Documents\Strata\models\hy4-flash\Hy4-Flash-Engram-Q8.strata --vram-experts-gb 0.57 --ram-experts-gb 55.47 --max-context 4096 unknown argument: --engram strata generate --pack DIR --tokens "1,2,3" [options] ``` The message "Your current engine is unchanged. Run `strata update --force`" is also wrong in practice: after these "failed" updates the engine 0.1.65-cuda12 IS installed and is the live process right now (`components\engine\0.1.65-windows-x86_64-cuda-cuda12\strata.exe --serve --pack ...\swift-iq3_s ...` is running on the V100, 14.8/16 GB VRAM, all fine). So every auto-update looks broken because of the test, while the engine itself works. Related observation with the same Hy4-Flash model: started from the panel it does not run on the V100 with this cuda12 engine, but runs on the RTX 5070. The `--engram` flag in the updater's test command may be the same mismatch seen from the other side. ### Smallest steps or command that reproduce it 1. Strata 0.2.18 + engine 0.1.64 on Windows; an engine update appears in the feed (happened twice today, 16:08 and 16:22 local). 2. Let the auto-update run (or `strata update`). 3. The update reports "update-engine failed ... did not start correctly ... see test-start-<version>.log"; that log always shows `unknown argument: --engram`. Retries repeat identically. ### Reproduction status Reproduced locally ### Strata version Strata 0.2.19 · engine 0.1.65-windows-x86_64-cuda-cuda12 ### Graphics card, memory and system Tesla V100-SXM2-16GB (TCC, adapter board with blower) + RTX 5070 12 GB · 80 GB DDR4 · Windows 11 Insider 26300; driver 581.80 GeForce installed over full 580.88 Data Center (nvidia-smi sees both) ### Model, size and settings The live model is Qwen3.8-Flash-Next IQ3_S 54.5 GB · 256K context · vision on · kv int8. The failed smoke-test model was Hy4-Flash-SQ-64 + Hy4-Flash-Engram-Q8 (the updater's own test model). ### Logs Lines 1-3 of `C:\Users\xglem\AppData\Local\Strata\components\.updates\logs\test-start-0.1.65.log` are quoted verbatim above (launch command, `unknown argument: --engram`, usage header). Full desktop log available on request. Extra (optional, same panel, tracked in #6318): the model tab still shows the stale banner "Tesla V100-SXM2-16GB cannot run Strata" while the engine runs on that card, and its gauges read "Video memory: free 0 GB of 119 GB" and "RAM: 18.7 GB of 769 GB" — wrong numbers on a healthy system (real: V100 14.8/16 GB used, system RAM 80 GB).
Related on this site
Quick links to related pages.