Up to date
fastino Apache 2.0 out 32 days ago
486M parameters, text in, labels out.
Verified 2026-10-25 Rank #0 in Decision models · score 0 method v1 Sources 3 Downloads 30d 151.4k · #270 of 619 tracked
Picked by rule from 1 tracked piece: the lab's own post, the most shared, a field report, the latest.
GLiNER2.5-Decide is a 340M-parameter DeBERTa-v3-large encoder released under Apache-2.0 that runs natively on Macs via MLX Swift.
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: 12 quantized builds and 15 Spaces.
Top Spaces running it: multimodalart/jev-decision-index ♥ 802 · fastino/gliner25-decide-playground ♥ 14 · hugging-apps/gliner25-decide-demo ♥ 3 · davanstrien/topic-feed-lab ♥ 3
Weights only, an estimate: leave room for the KV cache. Method
About 0.3 GB at INT4, 0.6 GB at FP8, 1.2 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 (2026-09-23), amber triangles are 1 dated article (each links to it), grey ticks are repo commits.
28 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Everything else, folded: the full tables, every source.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| author-reported Exact-match accuracy on fastino/fast-decisions | 60.2% percent | GLiNER2.5-Decide (340M) | model author | 2026-10-09 |
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 | text-classification |
|---|---|
| Library | transformers |
| Architectures | SpanExtractor |
| Model type | extractor |
| Params by dtype | F32: 486,444,053 |
| 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 GLiNER2.5-Decide: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 1.2 GB · Edge (<4 GB) |
|---|---|
| FP8 (1 byte/param) | 0.6 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.
| How it was made | full fine-tune of vendor own GLiNER2 encoder |
|---|---|
| Lineage bucket | O - original architecture or own base |
| Lineage evidence | HF base_model fastino/gliner2-large-v1 (vendor own architecture) |
| Base model | fastino/gliner2-large-v1 |
| DI 0.2.1 score | 11.21 (independent (DI 0.2.1 board)) |
| ECE (board) | 0.0884 |
| Median latency | 23.3 ms (DI board, 1x RTX PRO 6000, single process) |
| Hosted price (input) | self-hosted/M |
| Source | https://huggingface.co/fastino/GLiNER2.5-Decide |
Base model: fastino/gliner2-large-v1
quantized: 12finetune: 14adapter: 2
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| FluidInference/gliner2-5-decide-coreml | quantized | — | — | — | apache-2.0 | FluidInference | 24 | 659 |
| litert-community/GLiNER2.5-Decide-LiteRT | finetune | — | — | — | apache-2.0 | litert-community | 11 | 597 |
| onnx-community/GLiNER2.5-Decide-ONNX | quantized | — | — | — | apache-2.0 | onnx-community | 2 | 937 |
| antflydb/GLiNER2.5-Decide | quantized | GGUF | — | — | apache-2.0 | antflydb | 0 | 574 |
| onnx-community/GLiNER2.5-Decide-mobile-ONNX | quantized | — | — | — | apache-2.0 | onnx-community | 1 | 350 |
| yale-cultural-heritage/GLiNER2.5-Decide-linkedart-recon | finetune | — | 486M | S | cc-by-nc-4.0 | yale-cultural-heritage | 4 | 146 |
| aufklarer/GLiNER2.5-Decide-340M-MLX-8bit | finetune | MLX | — | — | apache-2.0 | aufklarer | 0 | 189 |
| yannelli/gliner2.5-decide-code-review-decisions | finetune | — | 486M | S | apache-2.0 | yannelli | 2 | 48 |
| kgrozdanovski/caldec-v1-gliner2.5-decide | finetune | — | 486M | S | apache-2.0 | kgrozdanovski | 2 | 37 |
| mlboydaisuke/GLiNER2.5-Decide-CoreAI | quantized | — | — | — | apache-2.0 | mlboydaisuke | 1 | 84 |
| aufklarer/GLiNER2.5-Decide-340M-MLX | finetune | MLX | — | — | apache-2.0 | aufklarer | 0 | 129 |
| aufklarer/GLiNER2.5-Decide-340M-MLX-fp16 | finetune | MLX | — | — | apache-2.0 | aufklarer | 0 | 128 |
| coreai-community/GLiNER2.5-Decide-CoreAI | quantized | — | — | — | apache-2.0 | coreai-community | 0 | 100 |
| nativ-community/GLiNER2.5-Decide | finetune | — | 436M | S | apache-2.0 | nativ-community | 0 | 86 |
| TyrusRC/haruspex | finetune | — | — | — | apache-2.0 | TyrusRC | 0 | 81 |
| souuzaa/GLiNER2.5-Decide-8bit | quantized | -8bit | 486M | S | apache-2.0 | souuzaa | 0 | 74 |
| souuzaa/GLiNER2.5-Decide-4bit | quantized | -4bit | 486M | S | apache-2.0 | souuzaa | 0 | 69 |
| abdelstark/exitreceipt-gliner2.5-decide-lora | adapter | — | — | — | apache-2.0 | abdelstark | 0 | 63 |
| souuzaa/GLiNER2.5-Decide-bf16 | quantized | -bf16 | 486M | S | apache-2.0 | souuzaa | 0 | 62 |
| nishparadox/gliner2.5-decide-onnx | quantized | — | — | — | apache-2.0 | nishparadox | 1 | 0 |
| hmarchant/gliner2.5-decide-person-resolution | finetune | — | 486M | S | apache-2.0 | hmarchant | 0 | 49 |
| augustoFranke/GLiNER2.5-Decide-CoreML-FP16-MultiFn-L64-512 | quantized | -FP16 | — | — | apache-2.0 | augustoFranke | 0 | 13 |
| smdesai/GLiNER25-Decide-FP16-CoreML | quantized | -FP16 | — | — | apache-2.0 | smdesai | 0 | 11 |
| coral-bricks-ai/gliner2.5-decide-call-actions | finetune | — | 486M | S | apache-2.0 | coral-bricks-ai | 0 | 8 |
| smdesai/GLiNER25-Decide-FP16-CoreAI | finetune | -FP16 | — | — | apache-2.0 | smdesai | 0 | 0 |
| openbles/firewall-1 | adapter | — | — | — | mit | openbles | 0 | 0 |
| TyrusRC/haruspex-adapters | finetune | — | — | — | apache-2.0 | TyrusRC | 0 | 0 |
| brenorb/doc-quality | finetune | — | — | — | — | brenorb | 0 | 0 |
| Date | Commit | SHA |
|---|---|---|
| 2026-10-08 | Install with the local extra; tag as text-classification (#8) | 9d1bfb8 |
| 2026-10-08 | Install with the local extra: pip install "gliner2[local]" (#13) | b0c78b4 |
| 2026-10-08 | Add a Fine-tuning section with training-row formats (#12) | f87a56c |
| 2026-10-08 | Load this checkpoint as a Transformers schema-extraction model. (#11) | 2125c8a |
| 2026-10-08 | Add decision-model tag (#10) | a6ef71c |
| 2026-09-28 | Point the fine-tune button to agent.fastino.ai | 5a7adf7 |
| 2026-09-28 | Point the fine-tune banner to agent.fastino.ai | ec2a5d1 |
| 2026-09-24 | Normalize model card and move benchmark near top | 7ee5da4 |
| 2026-09-24 | Update the Fastino GLiNER skill | af5e41e |
| 2026-09-24 | Update GLiNER2 deployment badge link (#6) | bbe10ff |
| 2026-09-24 | Update GLiNER2.5 fine-tuning badge link (#5) | 01eaa61 |
| 2026-09-24 | Update Fastino API link (#4) | 55dba8d |
| 2026-09-24 | Update the Fastino GLiNER skill | 65624f1 |
| 2026-09-24 | Mention hosted fine-tuning via the Fastino API (#3) | 0872ab1 |
| 2026-09-24 | Update the Fastino GLiNER skill | 4069ac8 |
| 2026-09-24 | Add the Fastino GLiNER skill | 79c3fb9 |
| 2026-09-24 | Point the benchmark at fastino/fast-decisions | 8d7170d |
| 2026-09-24 | Add question, book, label, and ordinal examples | b49028a |
| 2026-09-24 | Explain each classification use case on the model card | 1e2df4e |
| 2026-09-24 | Rewrite the model card for release | ee4abe5 |
| 2026-09-24 | Add classification task examples to the model card | f05b122 |
| 2026-09-24 | Update to step 126500 with benchmark table | da05b29 |
| 2026-09-24 | upload banner (#1) | bb8e3f9 |
| 2026-09-24 | update banner (#2) | 3846d9d |
| 2026-09-23 | Upload large classification finetune at step 95500 | 91c8b6b |
| 2026-09-23 | initial commit | e4ab809 |
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.
arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).
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
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.
