The story so far
Weights, every dated article, and repo commits on one rail.
11 repo commits since the weights landed
Teal is the weights (2026-09-24), grey ticks are repo commits.
interfaze-ai Apache 2.0 out 31 days ago
Text in, labels out.
Verified 2026-10-25 Rank #0 in Decision models · score 0 method v1 Sources 3 Downloads 30d 1.2k · #589 of 619 tracked
Do people use it?1.2kdownloads, 30 days#589 of 619 we track · below the median+3 likes, downloads +9% this week129 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: 1 quantized builds and 8 Spaces.
- As a service0no API provider listed
- On own hardware1quantized builds
- Built on00 fine-tunes · 0 adapters · 0 merges
- In apps8Hugging Face Spaces
Top Spaces running it: multimodalart/jev-decision-index ♥ 802
What people built on it
1 derivative on the Hub. Which build to run, by runtime: the most downloaded in each class.
1 derivative on the Hub
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Reference
Everything else, folded: the full tables, every source.
Benchmarks
19 rows
Benchmarks 19 rows
measured third-party leaderboard rowauthor-reported number from the model card or eval files (19)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 Held-out split | 0.807* | weighted accuracy | model author | 2026-10-23 |
| author-reported Held-out split | 0.061* | expected calibration error | model author | 2026-10-23 |
| author-reported Held-out split | 0.98 | banking77 (77 intents) | model author | 2026-10-23 |
| author-reported Held-out split | 0.97 | clinc_oos (151 intents) | model author | 2026-10-23 |
| author-reported Held-out split | 0.872* | FEVER claim verification | model author | 2026-10-23 |
| author-reported S1Bench | 0.73 | vitaminc-dev | model author | 2026-10-23 |
| author-reported S1Bench | 0.87 | massive-en-US | model author | 2026-10-23 |
| author-reported S1Bench | 0.83 | massive-de-DE | model author | 2026-10-23 |
| author-reported S1Bench | 0.85 | boolq | model author | 2026-10-23 |
| author-reported S1Bench | 0.82 | squad2 | model author | 2026-10-23 |
| author-reported S1Bench | 0.84 | paws | model author | 2026-10-23 |
| author-reported S1Bench | 0.85 | multinli | model author | 2026-10-23 |
| author-reported S1Bench | 0.78 | civil_comments | model author | 2026-10-23 |
| author-reported S1Bench | 0.84 | aegis2 | model author | 2026-10-23 |
| author-reported S1Bench | 0.36 | helpsteer2 | model author | 2026-10-23 |
| author-reported S1Bench | 0.37 | summeval-relevance | model author | 2026-10-23 |
| author-reported S1Bench | 0.81 | summeval-consistency | model author | 2026-10-23 |
| author-reported S1Bench | 0.72 | pubmedqa | model author | 2026-10-23 |
| author-reported S1Bench | 0.74 | macro | model author | 2026-10-23 |
What the card says
3 cited sentences
Summary generated, cited
lev is a LoRA adapter for the Qwen3.5-4B base model designed for zero-shot classification tasks.[1] The model performs routing, moderation, and intent detection by returning calibrated probabilities in a single forward pass.[2] It is released under the Apache 2.0 license and supports the TypeSafe /v1/systemone protocol.[3][4]
- “It is a LoRA adapter on Qwen3.5-4B”
- “It is built for the high-volume judgement calls inside a …”
- license
- “speaks TypeSafe's `/v1/systemone` protocol”
Generated 2026-10-07 15:18 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.
Use-case evidence
5 use cases
What it is good at 5 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.
- Classify and make typed decisionsstructuraldecision-model estimate p=1.00tag: decision-modelpipeline: zero-shot-classification+1 more
- Call tools and APIsstructuralchat template: <tool_call>
- Reason through hard problemsstructuralchat template: <think>
- Chat and follow instructionsstructuralchat template present
- Moderate and guard contentstructuraltag: moderation
Facts
12 fields
Listed in Decision models
Facts
- License: Apache 2.0 (friction: none). Permissive; patent grant. (license file, observed 2026-10-24)
- Created 2026-09-24, last modified 2026-10-15. (HF API, observed 2026-10-24)
| Pipeline tag | zero-shot-classification |
|---|---|
| Library | peft |
| MoE | no |
| Languages | en |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-25 |
Decision facts
catalog
Decision-model catalog curated
| How it was made | LoRA + new scoring/pointer head |
|---|---|
| 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 | Qwen/Qwen3.5-4B-Base |
| DI 0.2.1 score | 38.54 (independent (DI 0.2.1 board)) |
| ECE (board) | 0.0638 |
| Median latency | 70.8 ms (DI board, 1x RTX PRO 6000, single process) |
| Hosted price (input) | self-hosted/M |
| Contamination notes | 22 BANKING77 test items found in lev's training records count as wrong; 12 CLINC150+OOS test items found in lev's training records count as wrong; 1 ANLI test items found in lev's training records count as wrong; 9 ARC-Easy test items found in lev's training records count as wrong; 1 HellaSwag test items found in lev's training records count as wrong |
| Source | https://huggingface.co/interfaze-ai/lev |
All variants
1 variant
Lineage
Base model: Qwen/Qwen3.5-4B
quantized: 1
Top variants (1 shown, ranked by downloads)
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| ggml-org/lev-GGUF | quantized | GGUF | — | — | apache-2.0 | ggml-org | 13 | 7.6k |
Repository changelog
11 commits
Repository changelog 11 commits
| Date | Commit | SHA |
|---|---|---|
| 2026-10-15 | lev: step 18750 of preset 4b-instruct | 8721908 |
| 2026-10-02 | Update README.md | 7bdc748 |
| 2026-10-01 | Update README.md | df2865a |
| 2026-09-25 | docs: add a hosted classification section linking Interfaze | f8ef711 |
| 2026-09-25 | Update README.md | 481f5bd |
| 2026-09-24 | docs: report lev vs Jev on all 13 S1Bench subsets | d045c05 |
| 2026-09-24 | Card: inline-block badges so the Hub keeps them on one row | 4ac3955 |
| 2026-09-24 | Card: badges and links on one line; padded, rounded, 2x charts; vertical diagram | cd391f9 |
| 2026-09-24 | Model card in the JEVfire style: hero, Jev results, charts, how it works | f1b230a |
| 2026-09-24 | lev: step 18750 of preset 4b-instruct | b9a47f2 |
| 2026-09-24 | initial commit | fe90e3e |
Similar models
3 models
Similar models
- kev-4b permissive license
- Bespoke-Nimble-9B permissive license
- kev-9b permissive license
Sources
3 sources
Sources
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:43 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:46 UTC · Summary: 2026-10-07 15:18 UTC · Last verified 2026-10-25
Follow lev
Watch this model
- /feeds/models/interfaze-ai--lev.xml(RSS 2.0) · every change to lev
- /feeds/families/interfaze-ai--lev.xml(RSS 2.0) · every change across the lev family
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.