ENPT

garum

Last updated:

Open—$0/M input

Quadrant scores

See the full quadrant

Scored with the public quadrant rubric (maturity × capability; bubble = adoption). Revise as evidence lands.

  • Maturity4.5/10

    Apache-2.0 LoRA for google/gemma-4-31B-it plus a Bayesian last-layer head, loaded through the repo's own Python class. GitHub mirror holds the code. Independent author; no hosted API, no server or integrations, no SLA.

  • Capability5.4/10

    Decision Index 0.4.1 (22 Oct) Full 60.27 (#39), just above Jev 1.13 (59.22). Choice and yes/no only (no score), probabilities from a calibrated Bayesian head, no ECE published. No latency figures.

  • Adoption1/100

    About 32 downloads and no likes (20 Oct). No integrations or production users found.

Vendor claims

publicado (OK do Fabricio 20/10 BRT). Scores are calculated from scratch.

What it is

Independent LoRA plus Bayesian head for Gemma 4 31B that answers choice and yes/no questions with a probability per option. Decision Index 0.4 Full 61.69 (#31). Source: huggingface.co/plantwaterer/garum.

Board result

Board Row Result
Decision Index 0.4 (20 Oct 2026) garum Full 61.69 (#31); Jev 1.13 61.26 (#34), Jev 1.14 62.22 (#28)

Not checked against JevBench or Artificial Analysis for this page; the boards are never combined.

Vendor claims (UnverifiedClaim)

  • Calibrated Bayesian last-layer head: garum’s probabilities are the head’s output, not the option-letter logits (source)

Scores (from scratch, 20 Oct 2026)

  • Capability 5.4 = types × 0.30 + calibration × 0.25 + latency × 0.25 + accuracy × 0.20 = 6 × 0.30 + 4 × 0.25 + 4 × 0.25 + 8 × 0.20 = 5.40. Accuracy: 8 × 61.69 ÷ 61.26 = 8.06 → 8.
  • Maturity 4.5 = availability × 0.30 + docs × 0.25 + integrations × 0.25 + support × 0.20 = 7 × 0.30 + 6 × 0.25 + 2 × 0.25 + 2 × 0.20 = 4.50.
  • Adoption 1 = engagement × 0.35 + ecosystem × 0.35 + production × 0.30 = 1 × 0.35 + 1 × 0.35 + 0 × 0.30 = 0.70 → 1.

Board audit (22 Oct 2026, DI 0.4.1)

publicado (OK do Fabricio 22/10 BRT).

Decision Index 0.4.1 (22 Oct; reference Jev 1.13 Full 59.22, was 61.26 on 0.4) lists this page’s row at Full 60.27 (#39). Accuracy rule: 8 × 60.27 ÷ 59.22 = 8.14 → 8 (was 8). Capability unchanged. Maturity and adoption unchanged.

Same author: oenogarum and allec on Decision Index 0.4.1 (29 Oct 2026)

publicado (OK do Fabricio 30/10 BRT).

The 29 Oct batch of the Decision Index added two newer models by plantwaterer, both Apache-2.0 fine-tunes of google/gemma-4-31B-it with the same kind of calibrated last-layer head:

Board Row Result
Decision Index 0.4.1 (data 29 Oct 14:04 UTC) oenogarum (repo 25 Oct) Full 68.10 (#11 of 247), public 62.80; ECE 0.040; median 259 ms
same allec, oavg engine Full 66.96 (#21), public 62.03; ECE 0.047
same garum Full 60.27, unchanged

oenogarum’s card describes option-order averaging for short requests, temperature T = 1.080 on choice questions and a Bayesian last-layer correction on yes/no. It needs the repo’s own attention function (full-attention layers bidirectional). About 199 downloads, 0 likes on 30 Oct.

Scores unchanged (4.5 / 5.4 / 1): this page is scored on garum. oenogarum now has its own page (4.8 / 6.7 / 1; its Full is 8 points above garum’s).