All models
Observed · not hosted

Gemma 2 27B Instruction-Tuned

A signed record of the public facts about Google DeepMind's Gemma 2 27B Instruction-Tuned, as observed on 2026-06-25. LeanLogix does not host, train, or evaluate this model — this passport makes the cited facts tamper-evident, not a benchmark claim.

Signed public facts

DeveloperGoogle DeepMind
Parameters27.2B
Base architecturedecoder-only transformer (interleaved local/global attention, logit soft-capping, GQA)
Context window8,192 tokens
Modalitytext
Released2024-06
LicenseGemma Terms of Use
SourceGoogle Gemma 2 model card + technical report (arXiv 2408.00118)
Observed byLeanLogix · 2026-06-25
Dispositionobserved — public facts only, no benchmark signed

The seal — Ed25519, re-checkable offline

Signed with the same key that signs our hosted model passports — one provenance authority across both tiers. Verify it here, or paste the receipt into the public verifier and re-derive the proof in your own browser.

Signing key fpa97122c37f5c…97fac1e1
Body sha25683336b3ff864…7f185000

Public-fact observation of a model LeanLogix does not host, train, or evaluate. No benchmark or quality claim is made or signed — only the cited public facts on the observed date. Signed with the LeanLogix provenance key so the record is tamper-evident and offline-verifiable.