The story so far
Weights, every dated article, and repo commits on one rail.
9 repo commits since the weights landed
Teal is the weights (2026-09-21), grey ticks are repo commits.
sokudan-ja-310m
GeneLab Apache 2.0 out 34 days ago
315M parameters, text in, labels out.
Verified 2026-10-26 07:27 UTC Rank #0 in Decision models · score 0 method v1 Sources 3 Downloads 30d 268 · #611 of 619 tracked
Do people use it?268downloads, 30 days#611 of 619 we track · below the median+0 likes, downloads −12% this week0 likesHugging Face2026-10-26
Where it's used
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: 1 Spaces.
- As a service0no API provider listed
- On own hardware0quantized builds
- Built on00 fine-tunes · 0 adapters · 0 merges
- In apps1Hugging Face Spaces
Will it fit?
Weights only, an estimate: leave room for the KV cache. Method
Fits a 16 GB card at INT4
About 0.2 GB at INT4, 0.4 GB at FP8, 0.8 GB at BF16: weights only, so leave room for the KV cache.
Estimate params × bytes per param × 1.2 · how we estimate
Reference
Everything else, folded: the full tables, every source.
Benchmarks
29 rows
Benchmarks 29 rows
measured third-party leaderboard rowauthor-reported number from the model card or eval files (29)No confidence intervals unless the source gave them; settings differ between rows, so compare within a benchmark, not across. Date = the benchmark's own date when it gives one, else when we read it.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported M1m（held-out 属性の AUROC の平均） | 0.86 | v0.2 | model author | 2026-10-07 |
| author-reported M1（held-out bool AUROC、まとめて） | 0.91 | v0.2 | model author | 2026-10-07 |
| author-reported M2（state 400〜799 トークン帯の AUROC） | 0.86 | v0.2 | model author | 2026-10-07 |
| author-reported M5（`implies_declining` の AUROC） | 0.83 | v0.2 | model author | 2026-10-07 |
| author-reported `ends_with_question` | 0.78 | v0.2 | model author | 2026-10-07 |
| author-reported `implies_declining` | 0.83 | v0.2 | model author | 2026-10-07 |
| author-reported `implies_escalation` | 0.84 | v0.2 | model author | 2026-10-07 |
| author-reported `implies_running_out_of_patience` | 0.87 | v0.2 | model author | 2026-10-07 |
| author-reported `requests_owner_change` | 1 | v0.2 | model author | 2026-10-07 |
| author-reported bench_en | 0.87 | v0.2 | model author | 2026-10-07 |
| author-reported bench_en | 0.11 | v0.2 | model author | 2026-10-07 |
| author-reported bench_en | 0.65 | v0.2 | model author | 2026-10-07 |
| author-reported bench_en | 0.69 | v0.2 | model author | 2026-10-07 |
| author-reported bench_en | 0.62 | v0.2 | model author | 2026-10-07 |
| author-reported bench_ja | 0.88 | v0.2 | model author | 2026-10-07 |
| author-reported bench_ja | 0.1 | v0.2 | model author | 2026-10-07 |
| author-reported bench_ja | 0.07 | v0.2 | model author | 2026-10-07 |
| author-reported bench_ja | 0.82 | v0.2 | model author | 2026-10-07 |
| author-reported bench_ja | 0.21 | v0.2 | model author | 2026-10-07 |
| author-reported bench_ja | 0.78 | v0.2 | model author | 2026-10-07 |
| author-reported bench_ja | 0.18 | v0.2 | model author | 2026-10-07 |
| author-reported bench_ja | 0.84 | v0.2 | model author | 2026-10-07 |
| author-reported bench_ja | 0.13 | v0.2 | model author | 2026-10-07 |
| author-reported held-out bool ECE↓ | 0.12 | v0.2 | model author | 2026-10-07 |
| author-reported val bool ECE↓（10 分割） | 0.05 | v0.2 | model author | 2026-10-07 |
Show all 29 rows (4 more)
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported val bool acc | 0.93 | v0.2 | model author | 2026-10-07 |
| author-reported val choice acc | 0.72 | v0.2 | model author | 2026-10-07 |
| author-reported val score RPS↓ | 0.16 | v0.2 | model author | 2026-10-07 |
| author-reported val score acc | 0.55 | v0.2 | model author | 2026-10-07 |
Use-case evidence
2 use cases
What it is good at 2 use cases
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.
Facts
17 fields
Listed in Decision models
Facts
- sokudan-ja-310m has 315M parameters. (safetensors metadata, observed 2026-10-26)
- License: Apache 2.0 (friction: none). Permissive; patent grant. (license file, observed 2026-10-26)
- Created 2026-09-21, last modified 2026-09-29. (HF API, observed 2026-10-26)
| Pipeline tag | text-classification |
|---|---|
| Library | transformers |
| Architectures | SokudanJointModel |
| Model type | sokudan |
| Params by dtype | F32: 314,614,274 |
| Size band | S |
| MoE | no |
| Languages | ja |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-26 |
Servability
vLLM · SGLang
Servability
vLLM: not in registry · 4b1c04e76fe9ac3ca8b7e541dd423e7d5f38dc60SGLang: not in registry · 60d941293e3af27a12c6eb64e2c3e2e1a17f100f
vLLM registry (4b1c04e76fe9ac3ca8b7e541dd423e7d5f38dc60) · SGLang registry (60d941293e3af27a12c6eb64e2c3e2e1a17f100f) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Hardware fit
3 estimates
Hardware fit ESTIMATE
Weights-only memory for sokudan-ja-310m: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 0.8 GB · Edge (<4 GB) |
|---|---|
| FP8 (1 byte/param) | 0.4 GB · Edge (<4 GB) |
| INT4 (0.5 bytes/param) | 0.2 GB · Edge (<4 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra.
Decision facts
catalog
Decision-model catalog curated
| How it was made | Japanese encoder fine-tune |
|---|---|
| Lineage bucket | F - fine-tune of an open model (public/synthetic data) |
| Lineage evidence | HF base_model |
| Base model | sbintuitions/modernbert-ja-310m |
| Hosted price (input) | self-hosted/M |
| Source | https://huggingface.co/GeneLab/sokudan-ja-310m |
All variants
lineage
Lineage
Base model: sbintuitions/modernbert-ja-310m
Repository changelog
9 commits
Repository changelog 9 commits
| Date | Commit | SHA |
|---|---|---|
| 2026-09-29 | card: sokudan 0.3.0 (MLX, Python 3.11-3.13, catch-all limits, bench license wording) | 3e3409b |
| 2026-09-28 | Model card: pip install sokudan (PyPI 0.2.1), git+ as the development version, Colab notebook link | 5f91a0d |
| 2026-09-28 | v0.2.1: bool calibration (calibration.json, bool/2 only) and model card; weights unchanged | bb09c20 |
| 2026-09-26 | v0.2: model soup of v0.1 seeds 0-7 (v0.1 stays as revision v0.1) | 6cfe939 |
| 2026-09-24 | Model card: condition F, English reproduction, link to laya#259; absolute links in Limits | 10d80cd |
| 2026-09-22 | sokudan-ja-310m v0.1 (3 seeds) | b16df2b |
| 2026-09-21 | sokudan-ja-310m v0.1 (3 seeds) | efe7177 |
| 2026-09-21 | sokudan-ja-310m v0.1 (3 seeds) | 3426824 |
| 2026-09-21 | initial commit | 1c8ed34 |
Similar models
3 models
Similar models
- laya-multilingual · 322M same size band (S), same task (text-classification), permissive license
- laya · 421M same size band (S), same task (text-classification), permissive license
- laya-typed-decisions · 421M same size band (S), same task (text-classification), permissive license
Sources
3 sources
Sources
Hub snapshot: 2026-10-26 · HF API: 2026-10-26 07:26 UTC · Providers: 2026-10-26 07:26 UTC · Articles: 2026-10-26 07:27 UTC · Summary: 2026-10-07 15:19 UTC · Last verified 2026-10-26 07:27 UTC
Follow sokudan-ja-310m
Watch this model
- /feeds/models/GeneLab--sokudan-ja-310m.xml(RSS 2.0) · every change to sokudan-ja-310m
- /feeds/families/GeneLab--sokudan.xml(RSS 2.0) · every change across the sokudan family
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.