PSA Legal Benchmark geeks & SLM fans: smaller models get better at basic legal work

PSA Legal Benchmark geeks & SLM fans: smaller models get better at basic legal work

Google Gemma 4 26B with 4B active parameters is correct - the 31B version is not.

Check the S3 Evals framework on GitHub 👉🏽 https://lnkd.in/eRkVxQJe

Want to generate similar tests in your own jurisdiction or language? DM me for access to the tools