The story so far
Weights, every dated article, and repo commits on one rail.
22 repo commits since the weights landed
Teal is the weights (2026-09-20), grey ticks are repo commits.
Winnow-12B
EldanRing Apache 2.0 out 35 days ago
Images and text in, text out, 256K tokens of context.
Verified 2026-10-25 Rank #3140 in Decision models · score 54.6 method v1 Sources 3 Downloads 30d 62.7k · #365 of 619 tracked
Do people use it?62.7kdownloads, 30 days#365 of 619 we track · below the median+2 likes, downloads −4% this week73 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: 2 quantized builds and 6 Spaces.
- As a service0no API provider listed
- On own hardware2quantized builds
- Built on11 fine-tunes · 0 adapters · 0 merges
- In apps6Hugging Face Spaces
Top Spaces running it: multimodalart/jev-decision-index ♥ 802
What people built on it
3 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
3 derivatives on the Hub, most of them quantized
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
7 rows
Benchmarks 7 rows
measured third-party leaderboard rowauthor-reported number from the model card or eval files (7)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 Jev public, 231 decisions | 85.71% percent | Q8 GGUF | model author | 2026-10-07 |
| author-reported JevBench public subset, 231 items | 85.28% percent | Winnow-12B BF16 | model author | 2026-10-07 |
| author-reported JevBench public subset, 231 items | 85.71% percent | Winnow-12B Q8 | model author | 2026-10-07 |
| author-reported Kev-clean, 1,046 decisions | 81.45% percent | Q8 GGUF | model author | 2026-10-07 |
| author-reported Kev-v9 clean, 1,046 items | 81.45% percent | Winnow-12B BF16 | model author | 2026-10-07 |
| author-reported Kev-v9 clean, 1,046 items | 81.55% percent | Winnow-12B Q8 | model author | 2026-10-07 |
| author-reported Typed teacher agreement, 2,000 decisions / 400 groups | 69.90% percent | Q8 GGUF | model author | 2026-10-07 |
What the card says
1 cited sentence
Summary generated, cited
Winnow-12B is an EldanRing fine-tune of Gemma 4 12B IT designed for local typed decisions, chat, and image input.[1]
Generated 2026-10-07 15:17 UTC by parasail-gemma-4-26b-a4b-it; each numeral points at the fact it rests on. Not a verdict.
Use-case evidence
7 use cases
What it is good at 7 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.
- Understand images and videostructuraldecision-model estimate p=0.98tag: visionpipeline: image-text-to-text+1 more
- Chat and follow instructionsstructuraldecision-model estimate p=0.79tag: conversationalchat template present
- Classify and make typed decisionsstructuraldecision-model estimate p=0.99tag: decision-modeldecision catalog
- Run on a laptop or phonestructuraldecision-model estimate p=0.96tag: gguf
- Call tools and APIsstructuralchat template: tool_calls
- Reason through hard problemsstructuralchat template: <|think|>
- Read long documentsstructuralconfig: 262,144 context
Facts
17 fields
Listed in Decision models · LLMs
Facts
- Context length is 262,144 tokens (256K). (config.json max_position_embeddings, observed 2026-10-24)
- License: Apache 2.0 (friction: none). Permissive; patent grant. (license file, observed 2026-10-24)
- Created 2026-09-20, last modified 2026-10-13. (HF API, observed 2026-10-24)
| Pipeline tag | image-text-to-text |
|---|---|
| Library | gguf |
| Architectures | Gemma4UnifiedForConditionalGeneration |
| Model type | gemma4_unified |
| Dtype | bfloat16 |
| MoE | no |
| Context length | 262,144 (config.json max_position_embeddings) |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-25 |
Servability
vLLM · SGLang
Servability
vLLM: supported · Gemma4UnifiedForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Gemma4UnifiedForConditionalGeneration · 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.
Decision facts
catalog
Decision-model catalog curated
| How it was made | LoRA/adapter fine-tune |
|---|---|
| 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 | google/gemma-4-12B |
| DI 0.2.1 score | 50.02 (independent (DI 0.2.1 board)) |
| ECE (board) | 0.1679 |
| Median latency | 72.5 ms (DI board, 1x RTX PRO 6000, single process) |
| Hosted price (input) | self-hosted/M |
| Notes | Board raised a capacity limit ("lifted"). |
| Source | https://huggingface.co/EldanRing/Winnow-12B |
All variants
3 variants
Lineage
Base model: google/gemma-4-12B-it
quantized: 2finetune: 1
Top variants (3 shown, ranked by downloads)
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| Piotr1215/Winnow-12B-GGUF | quantized | GGUF | — | — | apache-2.0 | Piotr1215 | 0 | 1.3k |
| ollaya-dev/winnow | finetune | — | — | — | apache-2.0 | ollaya-dev | 2 | 0 |
| tielmane/Winnow-12B-NPU-LPBQ-X-Elite | quantized | — | — | — | apache-2.0 | tielmane | 1 | 0 |
Repository changelog
22 commits
Repository changelog 22 commits
| Date | Commit | SHA |
|---|---|---|
| 2026-10-13 | Add decision-model tag (#4) | 0b5b7f9 |
| 2026-10-06 | Fix checksum and license links | 859a2cb |
| 2026-10-06 | Publish approved Winnow card, usage and evidence | 6f3417c |
| 2026-10-05 | Clarify target downloads and optional MTP assistant metadata | abb2162 |
| 2026-10-04 | Fix checksum and license file links in model card | 80b99e9 |
| 2026-10-04 | Publish Winnow runtime guidance, measured results and model cards | fdaebc9 |
| 2026-10-04 | Add verified NVFP4 and matching MTP assistant assets | 3439960 |
| 2026-09-21 | Separate the vision projector from model quantization choices | b6ac22b |
| 2026-09-21 | Publish BF16 GGUF, remove safetensors, and clarify GGUF downloads | 98851d9 |
| 2026-09-21 | Focus deployment chart on inference VRAM | 61c0ace |
| 2026-09-21 | Pin complete inference release with final benchmark report | de8d4d4 |
| 2026-09-21 | Add measured inference-speed figure and final benchmark graphics | 816496f |
| 2026-09-21 | Pin final verified inference release and benchmark figures | be16591 |
| 2026-09-21 | Polish comparison figures with compact labeled cards | 396e123 |
| 2026-09-21 | Download current BF16 model card alongside merged weights | 1d16c7d |
| 2026-09-21 | Pin verified inference source and immutable model downloads | 2f2c499 |
| 2026-09-21 | Label every comparison panel and stabilize figure metadata | b703405 |
| 2026-09-21 | Add Winnow-12B model card, benchmarks, figures and quickstart | e30cf0a |
| 2026-09-21 | Add Winnow-12B merged BF16, Q8 GGUF and vision projector | b2b1421 |
| 2026-09-20 | Prepare Winnow-12B private release documentation | 07594dc |
| 2026-09-20 | Prepare Winnow-12B private release documentation | 031a8dd |
| 2026-09-20 | initial commit | b35d2da |
Similar models
3 models
Similar models
- llava-v1.5-7b same task (image-text-to-text)
- opt-125m closest by size and task
- gpt-j-6b 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 15:17 UTC · Last verified 2026-10-25
Follow Winnow-12B
Watch this model
- /feeds/models/EldanRing--Winnow-12B.xml(RSS 2.0) · every change to Winnow-12B
- /feeds/families/EldanRing--Winnow.xml(RSS 2.0) · every change across the Winnow family
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.