v1.3.0: GPU inference, live score range, 25 contributors

About 60 pull requests from 25 contributors since 1.2.0. Thank you all.

Highlights

• GPU inference. Every model now runs on CUDA when torch can see a GPU. On an RTX 5070 Ti, 12 texts take 5.3 s instead of 319 s, with identical scores (#169).
• Live score range. Cheap engines run first, and the report shows where the score can still land and when it is final (#142, #143, #144, #147).
• Daily quota for public instances. SLOPTOTAL_DIRECT_DAILY_LIMIT caps API calls per IP from outside your own site. Off by default (#162).
• Calibrated quick score. /api/quick-score and /api/paragraph-score, used by the Chrome extension, are now fitted on SlopBench like the full report: 1.4% of human texts called AI, down from 12.5% (#101).
• READMEs in Chinese, Russian and Spanish, plus Node.js and Ruby example clients.
• Fixes: reports stuck at "Scanning..." after a failed database write (#152), DOCX order (#153), Windows encoding (#176), focus trapping in the queue dialog (#185).
• CI: Docker build on pull requests, CodeQL and OpenSSF Scorecard.

Full list in the changelog.

Docker

docker run -p 8000:8000 -v sloptotal-models:/app/models ghcr.io/pablocaeg/sloptotal:1.3.0

Contributors

@AK-Lmn @Bhavyashah20 @CloudCorpRecords @GreedyC @JohannGaviria @Mr-Neutr0n @Priyadharshini537 @Reviem0 @ST-SACHIN-KUM @SakarPaisen @SamipPandey @akkusnisanur250-commits @ammadakram @anmol-96 @brunnojob @dbdefault @have-you-tried-restarting @jay23606 @jotilohana21 @karthik-srivathsa-05 @kudala-bharani @markankamaransheogarat-sys @saisarans2007 @yashthakur2008 @yellow-dog-dog