Up to date
DavidHatley Apache 2.0 out 39 days ago
69.3M parameters, text in, embeddings out.
Verified 2026-10-25 Rank #0 in Decision models · score 0 method v1 Sources 3 Downloads 30d 76 · #617 of 619 tracked
Picked by rule from 3 tracked pieces: the lab's own post, the most shared, a field report, the latest.
TypeSafe AI releases Jev, a System One Model designed for fast, structured automation decisions.
The 2 picks above, plus these 1. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
No API provider serves it, so its downloads are people running it themselves: 5 Spaces.
Weights only, an estimate: leave room for the KV cache. Method
About 0.1 GB at FP8, 0.2 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2026-09-16), amber triangles are 3 dated articles (each links to it), grey ticks are repo commits.
Everything else, folded: the full tables, every source.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported Accuracy | 86.25% percent | Final variant: Raw | model author | 2026-10-07 |
| author-reported Accuracy | 86.25% percent | Final variant: Temperature scaled | model author | 2026-10-07 |
| author-reported Accuracy | 100.00% percent | Raw per-head: task_complete | model author | 2026-10-07 |
| author-reported Accuracy | 82.65% percent | Raw per-head: should_retry | model author | 2026-10-07 |
| author-reported Accuracy | 92.95% percent | Raw per-head: failure_source | model author | 2026-10-07 |
| author-reported Accuracy | 62.80% percent | Raw per-head: next_action | model author | 2026-10-07 |
| author-reported Accuracy | 92.85% percent | Raw per-head: evidence_sufficient | model author | 2026-10-07 |
| author-reported Accuracy | 59.27% percent | Development variant: Raw | model author | 2026-10-07 |
| author-reported Accuracy | 59.27% percent | Development variant: Temperature scaled | model author | 2026-10-07 |
| author-reported Brier | 0.26 | Final variant: Raw | model author | 2026-10-07 |
| author-reported Brier | 0.39 | Final variant: Temperature scaled | model author | 2026-10-07 |
| author-reported Brier | 0.77 | Development variant: Raw | model author | 2026-10-07 |
| author-reported Brier | 0.43 | Development variant: Temperature scaled | model author | 2026-10-07 |
| author-reported ECE | 0.13 | Final variant: Raw | model author | 2026-10-07 |
| author-reported ECE | 0.27 | Final variant: Temperature scaled | model author | 2026-10-07 |
| author-reported ECE | 0 | Raw per-head: task_complete | model author | 2026-10-07 |
| author-reported ECE | 0.17 | Raw per-head: should_retry | model author | 2026-10-07 |
| author-reported ECE | 0.05 | Raw per-head: failure_source | model author | 2026-10-07 |
| author-reported ECE | 0.36 | Raw per-head: next_action | model author | 2026-10-07 |
| author-reported ECE | 0.06 | Raw per-head: evidence_sufficient | model author | 2026-10-07 |
| author-reported ECE | 0.37 | Development variant: Raw | model author | 2026-10-07 |
| author-reported ECE | 0.15 | Development variant: Temperature scaled | model author | 2026-10-07 |
| author-reported Mean confidence | 98.89% percent | Final variant: Raw | model author | 2026-10-07 |
| author-reported Mean confidence | 63.19% percent | Final variant: Temperature scaled | model author | 2026-10-07 |
| author-reported Mean confidence | 99.96% percent | Raw per-head: task_complete | model author | 2026-10-07 |
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported Mean confidence | 99.24% percent | Raw per-head: should_retry | model author | 2026-10-07 |
| author-reported Mean confidence | 97.46% percent | Raw per-head: failure_source | model author | 2026-10-07 |
| author-reported Mean confidence | 99.17% percent | Raw per-head: next_action | model author | 2026-10-07 |
| author-reported Mean confidence | 98.63% percent | Raw per-head: evidence_sufficient | model author | 2026-10-07 |
| author-reported Mean confidence | 96.58% percent | Development variant: Raw | model author | 2026-10-07 |
| author-reported Mean confidence | 62.39% percent | Development variant: Temperature scaled | model author | 2026-10-07 |
| author-reported NLL | 0.85 | Final variant: Raw | model author | 2026-10-07 |
| author-reported NLL | 0.71 | Final variant: Temperature scaled | model author | 2026-10-07 |
| author-reported NLL | 0 | Raw per-head: task_complete | model author | 2026-10-07 |
| author-reported NLL | 1.2 | Raw per-head: should_retry | model author | 2026-10-07 |
| author-reported NLL | 0.3 | Raw per-head: failure_source | model author | 2026-10-07 |
| author-reported NLL | 2.4 | Raw per-head: next_action | model author | 2026-10-07 |
| author-reported NLL | 0.34 | Raw per-head: evidence_sufficient | model author | 2026-10-07 |
| author-reported NLL | 2.44 | Development variant: Raw | model author | 2026-10-07 |
| author-reported NLL | 0.79 | Development variant: Temperature scaled | model author | 2026-10-07 |
System One Mini is a 69.3M-parameter DistilBERT encoder designed for five fixed, typed decisions over synthetic software-diagnosis summaries.[1][2] The model uses a shared encoder with five independent classification heads and is released under the Apache 2.0 license.[3][4]
Generated 2026-10-07 15:19 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
Listed in Decision models
| Pipeline tag | feature-extraction |
|---|---|
| Library | transformers |
| Architectures | SystemOneModel |
| Model type | system_one |
| Dtype | float32 |
| Params by dtype | F32: 69,328,144 |
| Size band | S |
| MoE | no |
| Languages | en |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-25 |
vLLM: not in registry · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: not in registry · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Weights-only memory for system-one-mini: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 0.2 GB · Edge (<4 GB) |
|---|---|
| FP8 (1 byte/param) | 0.1 GB · Edge (<4 GB) |
| INT4 (0.5 bytes/param) | 0.0 GB · Edge (<4 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
| How it was made | full fine-tune, five fixed heads, RTX 3060 |
|---|---|
| Lineage bucket | F - fine-tune of an open model (public/synthetic data) |
| Lineage evidence | model card |
| Base model | distilbert-base-uncased |
| Hosted price (input) | self-hosted/M |
| Notes | fixed decisions only; not general |
| Source | https://huggingface.co/DavidHatley/system-one-mini |
Base model: distilbert/distilbert-base-uncased
| Date | Commit | SHA |
|---|---|---|
| 2026-09-16 | Complete Apache-2.0 license text | b004fb1 |
| 2026-09-16 | Upload validated model release | 3df6849 |
| 2026-09-16 | Add model configuration | e2794f0 |
| 2026-09-16 | Add calibration parameters | 4736f08 |
| 2026-09-16 | Add custom Transformers model | 0b4d63e |
| 2026-09-16 | Add inference requirements | b31ce0e |
| 2026-09-16 | Add calibrated inference helper | 502d2a7 |
| 2026-09-16 | Add model card | 6240635 |
| 2026-09-16 | initial commit | ce754f0 |
Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:44 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:19 UTC · Last verified 2026-10-25
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.