At a glance
choice selects among labeled options, noul is yes/no and score uses an ordinal scale.- The training mixture covers English, Japanese, Chinese, Spanish, French, German, Portuguese, Russian and a smaller Korean share. Performance varies by language.
- Probabilities are fitted by answer type; the measured endpoint configuration and calibration results are below.
- The base, adapter, merged weights and shared server code are Apache-2.0.
Repositories
Quick start
Use this merged checkpoint. Follow the server setup on a CUDA-capable Linux host to install SGLang 0.5.20 and the adapter in separate virtual environments. Start the engine and the adapter in separate terminals. Use the 3B commands below; the linked guide also covers another model size.
Engine:
CUDA_VISIBLE_DEVICES=0 SGLANG_VLM_CACHE_SIZE_MB=0 .venv-sglang/bin/python -m sglang.launch_server \ --model-path ./StandardOne-3B --served-model-name standard-one-3b \ --host 127.0.0.1 --port 30000 --tp-size 1 --model-impl sglang --dtype bfloat16 \ --context-length 32768 --max-running-requests 32 --mem-fraction-static 0.8 \ --chunked-prefill-size -1 --disable-radix-cache --mm-preprocess-cache-size-mb 0 \ --model-config-parser hf --load-format safetensors
Adapter:
.venv-native/bin/jev-adapter --engine-url http://127.0.0.1:30000 --model standard-one-3b --alias jev-latest \ --host 0.0.0.0 --port 30120 --max-concurrency 1 \ --tokenizer-model mistralai/Ministral-3-3B-Instruct-2512-BF16 \ --tokenizer-revision b6d637bef2393152b3da2b2fde72eecdee30557e \ --prompt-wording served --label-scheme upper --default-temperature 1.65
The endpoint accepts choice (labeled options), noul (yes/no) and score (ordinal) questions. It returns probabilities for the caller's labels in one forward pass without decoding answer text. For example:
curl -s http://127.0.0.1:30120/v1/systemone -X POST -H 'content-type: application/json' -d '{ "model": "jev-latest", "state": "Refunds require a receipt. The customer has no receipt.", "questions": {"decision": {"type": "noul", "instructions": "Is a refund allowed under the rule?", "criteria": {"true": "The requirement is met.", "false": "A requirement is missing."}}}}'
usage.output_tokens is 0. The server applies the default temperature shown above unless a request supplies options.temperature. Each probability vector is normalized over exactly the caller's labels.
Prompt wording and temperatures
served is the adapter's default wording and the one used for the v2.2 measurements (--label-scheme upper, --default-temperature 1.65); native accepts at most 26 options per question. The table below is a v2.1 measurement of both wordings through the same served endpoint. Temperatures were fitted separately for each wording and answer type on three calibration suites only and have not been refit for v2.2. No JevBench item was used for these decisions.
Table with columns: Wording, Default temperature, Choice, Yes/no, Ordinal, Mean accuracy, 10 suites| Wording | Default temperature | Choice | Yes/no | Ordinal | Mean accuracy, 10 suites |
|---|
native | 0.90 | 0.90 | 1.15 | 1.20 | 69.91 % |
served (adapter default) | 1.00 | 1.00 | 0.90 | 0.80 | 70.21 % |
To use native with its v2.1-fitted temperatures, pass --prompt-wording native --native-system-prompt none --default-temperature 0.90 --temperature-by-type choice=0.90,noul=1.15,score=1.20.
Changes in v2.2
v2.2 continues training from v2.1 with additional decision data. Questions with more than 26 options now use the labels A–Z, then AA, AB, …; the adapter in server/ uses this order by default (--label-scheme upper). Both versions were measured the same way: merged BF16 weights through SGLang 0.5.20 and jev-adapter, served wording, one option order, accuracy of the most probable answer; measured 3 October 2026.
Table with columns: Suite, v2.1, v2.2, Change (points)| Suite | v2.1 | v2.2 | Change (points) |
|---|
| many-option questions, 53–151 options (18,000) | 48.00 % | 79.97 % | +31.97 |
| the same question set, at most 26 options (750) | 79.07 % | 85.47 % | +6.40 |
| long-document questions (150) | 32.67 % | 38.00 % | +5.33 |
| held-out decision set (600) | 73.50 % | |
Decision Index 0.2.1 (balanced skill): 34.85, measured through the adapter in server/ (served wording, default temperature 1.65). The tables below are the v2.1 measurements with native wording and are not directly comparable with the table above.
Benchmarks
Served endpoint, v2.1
Merged BF16 weights through SGLang 0.5.20 and jev-adapter, one option order, native wording, no system prompt. Accuracy is independent of the fitted temperature.
Table with columns: Suite, Questions, Accuracy| Suite | Questions | Accuracy |
|---|
| Judge proxy | 600 | 87.50 % |
| Hard proxy | 600 | 44.17 % |
| Stated-distribution probability | 1,036 | 78.38 % |
| Realistic transfer | 600 | 90.17 % |
| MuSiQue multiple choice | 500 | 69.80 % |
| SQuAD 2.0 unanswerable | 500 |
The following public tiers are reported separately and were excluded from wording, temperature, checkpoint and quantization decisions. These are our measurements, not official sealed-set scores.
Table with columns: Suite, Questions, Accuracy| Suite | Questions | Accuracy |
|---|
| JevBench public easy | 48 | 100.00 % |
| JevBench public standard | 72 | 86.11 % |
| JevBench public hard | 111 | 47.75 % |
At the fitted temperatures, public-hard 10-bin ECE is 0.212 and mean total-variation distance on the stated-distribution validation set is 0.124.
Offline six-suite result, v2.1
This separate transformers run used native wording, no system prompt and T=1. It is not the served endpoint result. The three public JevBench tiers below are report-only and were excluded from wording, temperature, checkpoint and quantization decisions.
Table with columns: Suite, Questions, Accuracy| Suite | Questions | Accuracy |
|---|
| JevBench public easy | 48 | 100.00 % |
| JevBench public standard | 72 | 93.06 % |
| JevBench public hard | 111 | 47.75 % |
| Judge proxy | 600 | 88.00 % |
| Realistic transfer | 600 | 89.33 % |
| Stated-distribution probability | 1,036 |
Public classification and decision suites, v2.1
Served endpoint, 400 cases per suite, seed 13: AG News 84.5 %, typed decisions 67.3 %, nine-language MASSIVE intent mean 83.4 %, email spam 92.0 %, phishing 88.8 %. Results vary by language and task.
The full report in docs/BENCHMARKS.md and the classification tables in docs/public-classification-suites.md are historical; for this checkpoint, see Changes in v2.2. The historical v2 benchmark chart shows earlier-version measurements.
Model details
- Base model:
mistralai/Ministral-3-3B-Instruct-2512-BF16, revision b6d637bef2393152b3da2b2fde72eecdee30557e (Apache-2.0).
- Adapter architecture: LoRA r=16, α=32, dropout 0 on
q_proj, k_proj, v_proj, o_proj, gate_proj, up_proj and down_proj of the language model.
- Adapter file:
adapter_model.safetensors, 135,113,048 bytes, SHA-256 cfe06432d85efbe99df53dbfd3fa3881e5d6ef177c6e26d5800d8bf83afe79e4.
- Merge: 182 language-model tensors changed, none outside the language-model projections; maximum absolute weight change 0.0022.
Training data
The public source inventory shared with Standard One 8B is listed below. Training data is synthetic and format-augmented decision data plus decision items converted from public
datasets (listed below); the JevBench public tiers used only for evaluation carry MIT. The linked full report is historical.
Public datasets used (train splits where the dataset has one; licence as stated by each dataset; labels come from the
datasets, distractor options are generated by code):
Table with columns: Dataset, Licence| Dataset | Licence |
|---|
| SQuAD 2.0 | CC BY-SA 4.0 |
| ARC | CC BY-SA 4.0 |
| BoolQ | CC BY-SA 3.0 |
| CommonsenseQA | MIT |
| HellaSwag | MIT |
| Banking77 | CC BY 4.0 |
| Bias in Bios | MIT |
| Bitext customer support | CDLA-Sharing-1.0 |
| CLINC150 | CC BY 3.0 |
Historical upstream-id mapping: docs/BENCHMARKS.md.
Limitations
- Results depend on the task distribution and the serving path. Refit temperatures when using a materially different distribution.
- With
native wording, at most 26 options per question (one uppercase letter per option, A–Z). With served wording the adapter continues with AA, AB, … up to 255 single-token labels (--label-scheme upper); v2.2 was evaluated with up to 151 options.
- Served and offline probabilities can differ on identical prompts. Use the served results to estimate endpoint behavior.
- Korean is a small share of the multilingual training mixture.
- The card reports text benchmarks; it does not establish accuracy on image inputs.
- The sealed JevBench set has not been measured for this version.
Licence
Adapter weights, merged weights and this card: Apache-2.0. Base model mistralai/Ministral-3-3B-Instruct-2512 (and -BF16): Apache-2.0 per its Hugging Face model card, which adds that the model must not be used in a way that infringes, misappropriates, or otherwise violates any third party's rights. jev-adapter and SGLang: Apache-2.0. The JevBench harness and public tiers used for evaluation: MIT; other benchmark items keep their own upstream terms.
Citation
StandardThinking/StandardOne-3B (merged weights) · StandardThinking/StandardOne-3B-LoRA (adapter + merge recipe).