The story so far
Weights, every dated article, and repo commits on one rail.
13 repo commits since the weights landed
Teal is the weights (2026-09-24), grey ticks are repo commits.
lavoir
moganai CC BY-NC 4.0 out 31 days ago
422M parameters, text in, labels out.
Verified 2026-10-25 Rank #0 in Decision models · score 0 method v1 Sources 3 Downloads 30d 146 · #614 of 619 tracked
Can I use it?CC BY-NC 4.0Non-commercial only; commercial hosting not permitted without a separate licenselicence file2026-10-24
Do people use it?146downloads, 30 days#614 of 619 we track · below the median+0 likes, downloads +7% this week0 likesHugging Face2026-10-24
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: 5 Spaces.
- As a service0no API provider listed
- On own hardware0quantized builds
- Built on00 fine-tunes · 0 adapters · 0 merges
- In apps5Hugging Face Spaces
Top Spaces running it: multimodalart/jev-decision-index ♥ 802
Will it fit?
Weights only, an estimate: leave room for the KV cache. Method
Fits a 16 GB card at INT4
About 0.3 GB at INT4, 0.5 GB at FP8, 1.0 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 ABCD, first customer message | 0.66 | zero-shot | model author | 2026-10-07 |
| author-reported ABCD, first customer message | 0.19 | zero-shot | model author | 2026-10-07 |
| author-reported ABCD, first customer message | 49% other | zero-shot | model author | 2026-10-07 |
| author-reported ABCD, first customer message | 0.71 | zero-shot | model author | 2026-10-07 |
| author-reported AG News | 0.91 | single-turn | model author | 2026-10-07 |
| author-reported AUC, seen workflows | 0.8 | Lavoir | model author | 2026-10-07 |
| author-reported AUC, unseen workflows | 0.61 | Lavoir | model author | 2026-10-07 |
| author-reported Banking77 · Classify and make typed decisions | 0.53 | single-turn | model author | 2026-10-07 |
| author-reported Email spam | 0.99 | single-turn | model author | 2026-10-07 |
| author-reported Emotion (DAIR) | 0.57 | single-turn | model author | 2026-10-07 |
| author-reported MASSIVE-en intent | 0.81 | single-turn | model author | 2026-10-07 |
| author-reported MS MARCO · Rerank search results | 0.67 | single-turn | model author | 2026-10-07 |
| author-reported Model routing (domain) | 0.75 | single-turn | model author | 2026-10-07 |
| author-reported Phishing | 0.98 | single-turn | model author | 2026-10-07 |
| author-reported SGD, first user turn | 0.94 | zero-shot | model author | 2026-10-07 |
| author-reported SGD, first user turn | 0.04 | zero-shot | model author | 2026-10-07 |
| author-reported SGD, first user turn | 6% other | zero-shot | model author | 2026-10-07 |
| author-reported SGD, first user turn | 0.84 | zero-shot | model author | 2026-10-07 |
| author-reported Support triage | 0.38 | single-turn | model author | 2026-10-07 |
| author-reported ToxicChat · Moderate and guard content | 0.82 | single-turn | model author | 2026-10-07 |
| author-reported ToxicChat · Moderate and guard content | 0.6 | single-turn | model author | 2026-10-07 |
| author-reported VOI vs. oracle VOI | Spearman .85 | 8 training workflows | model author | 2026-10-07 |
| author-reported accuracy at ≤ 0.5 questions, seen | 0.75 | Lavoir | model author | 2026-10-07 |
| author-reported accuracy at ≤ 0.5 questions, unseen | 0.58 | Lavoir | model author | 2026-10-07 |
| author-reported ask_threshold = 0.02 + handoff_threshold = 0.3 | 0.97 | 8 training workflows | model author | 2026-10-07 |
Show all 29 rows (4 more)
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported ask_threshold = 0.05 | accuracy .610 → .885 | 8 training workflows | model author | 2026-10-07 |
| author-reported first-message accuracy | 0.51 | unseen workflows | model author | 2026-10-07 |
| author-reported first-message accuracy vs. Bayes ceiling | .648 vs .659 | 8 training workflows | model author | 2026-10-07 |
| author-reported typed-decisions | 0.77 | single-turn | model author | 2026-10-07 |
Use-case evidence
3 use cases
What it is good at 3 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
15 fields
Listed in Decision models
Facts
- lavoir has 422M parameters. (safetensors metadata, observed 2026-10-24)
- License: CC BY-NC 4.0 (friction: non-commercial). Non-commercial only; commercial hosting not permitted without a separate license. (license file, observed 2026-10-24)
- Created 2026-09-24, last modified 2026-09-25. (HF API, observed 2026-10-24)
| Pipeline tag | text-classification |
|---|---|
| Model type | lavoir |
| Params by dtype | F32: 421,562,382 |
| Size band | S |
| MoE | no |
| Languages | en |
| License | CC BY-NC 4.0 (cc-by-nc-4.0) |
| Hub snapshot | 2026-10-25 |
Servability
SGLang
Servability
SGLang: not in registry · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
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.
Hardware fit
3 estimates
Hardware fit ESTIMATE
Weights-only memory for lavoir: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 1.0 GB · Edge (<4 GB) |
|---|---|
| FP8 (1 byte/param) | 0.5 GB · Edge (<4 GB) |
| INT4 (0.5 bytes/param) | 0.3 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 | full fine-tune (whole weights published) |
|---|---|
| Lineage bucket | F - fine-tune of an open model (public/synthetic data) |
| Lineage evidence | board file inspection (adapter/head/weight files) + HF base_model field |
| Base model | answerdotai/ModernBERT-large |
| DI 0.2.1 score | 8.69 (independent (DI 0.2.1 board)) |
| ECE (board) | 0.1492 |
| Median latency | 20 ms (DI board, 1x RTX PRO 6000, single process) |
| Hosted price (input) | self-hosted/M |
| Source | https://huggingface.co/moganai/lavoir |
All variants
lineage
Lineage
Repository changelog
13 commits
Repository changelog 13 commits
| Date | Commit | SHA |
|---|---|---|
| 2026-09-25 | Card metadata: link training dataset, drop base_model (heads trained from scratch) | 4c5eaeb |
| 2026-09-25 | Model card: dataset and language badges; install from GitHub | 0195de7 |
| 2026-09-25 | Card: blog link to moganai.github.io/lavoir-web | 65b1289 |
| 2026-09-25 | Model card: heads trained from scratch, no Laya weights; encoder initialized from its base model | bc060d6 |
| 2026-09-25 | Revert bc63557: restore the model card (dataset card was uploaded by mistake) | c7467cd |
| 2026-09-25 | Update README.md | bc63557 |
| 2026-09-25 | tokenizer_config: portable tokenizer_class (loads on transformers 4.x and 5.x) | 39a94c1 |
| 2026-09-24 | Update banner | 840882b |
| 2026-09-24 | Model card: Lavoir banner | 07f46dd |
| 2026-09-24 | Model card: badges, clearer chart with captions, citation | 562c69a |
| 2026-09-24 | Model card: load with Lavoir.from_pretrained("moganai/lavoir") | 9e5b656 |
| 2026-09-24 | Lavoir: initial private upload (weights, tokenizer, encoder config, model card) | f5c713b |
| 2026-09-24 | initial commit | cee5cc8 |
Similar models
3 models
Similar models
- 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
- GLiNER2.5-Decide · 486M same size band (S), same task (text-classification), permissive license
Sources
3 sources
Sources
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
Follow lavoir
Watch this model
- /feeds/models/moganai--lavoir.xml(RSS 2.0) · every change to lavoir
- /feeds/families/moganai--lavoir.xml(RSS 2.0) · every change across the lavoir family
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.