Standard One 3B (LoRA adapter)
Updated weights (v2, 2026-09-26). If you downloaded this adapter before, download it again or pin
revision="v2". Earlier versions stay available under the tags v1 and v1.1.
Version: v2
Standard One scores a bounded set of answers for a supplied scenario and returns probabilities through
POST /v1/systemone. It does not generate free-form response text. This repository contains the 3B
LoRA adapter and a merge recipe; serving requires a merged checkpoint and the server code.
For ready-to-serve merged BF16 weights, use
StandardOne-3B. The server code is in
StandardOne-8B.
In the reported served evaluations, 3B has a lower median latency on the measured short-request
profile; 8B scores higher on the public standard and hard tiers. See Benchmarks
for the measurement conditions and limitations.
The figure combines results from different measurement paths. See Benchmarks for
served versus offline conditions; measured 24–26 September 2026.
At a glance
- Send a state and a bounded rubric to receive probabilities for the supplied labels:
choice selects among labeled options, noul is yes/no, and score uses an ordinal scale.
The endpoint scores the labels in one forward pass without decoding answer text after this adapter
is merged and served.
- In the same-run offline comparison with the untuned base, 3B improves on all six suites.
- Probabilities are temperature-scaled and calibration-checked (hard-tier ECE, distribution
total-variation) — see Benchmarks below.
- The card metadata lists English and Korean. The training mixture is multilingual, and the reported
nine-language MASSIVE intent evaluation has per-language results in
docs/public-classification-suites.md; quality should not
be assumed uniform across languages. The merged checkpoint retains the Pixtral vision tower and
accepts image data URLs; this card does not report a separate image-input benchmark.
- Apache-2.0 throughout: base model, adapter and merged weights.
Merge the adapter
import torch
from transformers import Mistral3ForConditionalGeneration
from peft import PeftModel
base = Mistral3ForConditionalGeneration.from_pretrained(
"mistralai/Ministral-3-3B-Instruct-2512-BF16",
revision="b6d637bef2393152b3da2b2fde72eecdee30557e",
torch_dtype=torch.bfloat16,
)
model = PeftModel.from_pretrained(base, ".").merge_and_unload()
model.save_pretrained("./StandardOne-3B-merged", safe_serialization=True)
# then copy the base snapshot's tokenizer / chat template / preprocessor / generation config
# files into ./StandardOne-3B-merged alongside the merged weights.
Then serve ./StandardOne-3B-merged with the sglang.launch_server and jev-adapter setup in
StandardOne-8B's QUICKSTART.md,
substituting the merged directory for --model-path (served model name standard-one-3b, no system
prompt, --prompt-wording native --native-system-prompt none --default-temperature 0.95 --temperature-by-type choice=0.95,noul=1.05,score=0.85). That guide also has the virtual-environment installation,
request format and curl example.
Prompt wording
jev-adapter can phrase a request in two ways. Both were measured on the same served endpoint (no system prompt) and each has its
own fitted temperatures; the recommended default is native unless served scores at least 1.0 percentage point higher on
the suites below and an offline check agrees.
--prompt-wording |
What the prompt looks like |
Temperatures (default; choice / noul / score) |
Mean accuracy, 10 suites |
native (recommended default) |
State: / Question: / Options: headers, options as A. name: description |
0.95; 0.95 / 1.05 / 0.85 |
69.9 % |
served |
Context: / Question: / Options: headers, options as A: name: description |
0.90; 0.90 / 1.05 / 0.95 |
69.8 % |
The 10 suites: judge proxy, hard proxy, stated-distribution probability, realistic transfer set, MuSiQue (multiple choice), SQuAD 2.0 unanswerable questions, ContractNLI, PAWS-X (English), a held-out hard decision set and a consistency set. None of them is a JevBench tier, and no JevBench item was used to choose the wording or the temperatures.
To use served, pass --prompt-wording served --default-temperature 0.90 --temperature-by-type choice=0.90,noul=1.05,score=0.95.
Benchmarks
Served endpoint results (the release configuration). Merged BF16 weights through SGLang 0.5.20
and jev-adapter, native wording, no system prompt, one option order, per-answer-type temperatures (choice 0.95, noul 1.05, score 0.85). The wording and the temperatures were chosen on non-JevBench data. Jev 1.13 was
measured on the same items through its hosted endpoint; its probabilities are raw, with no
temperature applied. These are our measurements, not official sealed-set JevBench scores.
| Suite |
Standard One 3B |
Jev 1.13 |
| JevBench public easy (48) |
100.00 % |
100.00 % |
| JevBench public standard (72) |
88.89 % |
98.61 % |
| JevBench public hard (111) |
46.85 % |
72.07 % |
| judge proxy (600: routing + answer adequacy) |
88.17 % |
90.50 % |
| realistic transfer set (600) |
90.67 % |
86.67 % |
| stated-distribution probability (1,036) |
78.76 % |
72.97 % |
| hard proxy (600) |
44.67 % |
54.83 % |
Offline comparison with the untuned base. This separate transformers runner used native
wording, no system prompt, one option order, T=1. The base and tuned checkpoint were scored by the
same offline path; these numbers are indicative of the base-model change, not the served scores above.
| Suite |
Untuned base |
Standard One 3B |
| JevBench public easy (48) |
97.92 % |
100.00 % |
| JevBench public standard (72) |
70.83 % |
93.06 % |
| JevBench public hard (111) |
47.75 % |
49.55 % |
| judge proxy (600: routing + answer adequacy) |
63.17 % |
88.00 % |
| realistic transfer set (600) |
62.50 % |
88.50 % |
| stated-distribution probability (1,036) |
31.56 % |
79.25 % |
Against the untuned base, all six suites improve on this offline run. Served and offline probabilities differ even on identical prompts, so use
the served table for expected endpoint behavior. Hard-tier ECE at the served temperatures is 0.225 against
Jev 1.13's 0.099 raw, and mean TV to the stated distributions is 0.117 at served T against Jev's
0.192 raw; full calibration table: docs/BENCHMARKS.md.
Speed — raw serial latency on one H200 with SGLang 0.5.20, using a 242-decision profile averaging
about 280 input tokens per decision. The 22.6 ms figure is p50 for this profile, not a latency
guarantee for other request lengths, concurrency or hardware. Qwen checkpoints are untuned and shown
for speed only; no accuracy comparison is implied.
| Model |
p50 |
p95 |
Input tokens/decision |
| Standard One 3B |
22.6 ms |
33.2 ms |
≈280 |
| Standard One 8B |
25.8 ms |
41.9 ms |
≈280 |
| Qwen3-8B (untuned) |
28.5 ms |
57.2 ms |
278 |
| Qwen3.5-4B (untuned) |
48.8 ms |
72.7 ms |
283 |
Throughput has been measured for the 8B only (see docs/BENCHMARKS.md); no equivalent sweep has been
run for the 3B.
On the public classification and decision suites (400 cases/suite, seed 13, served endpoints): AG News
84.2 %, typed decisions 67.8 %, MASSIVE intent mean 82.9 %, email spam 93.2 %, phishing 89.5 %. Full
table, per-language and per-workflow breakdown: docs/BENCHMARKS.md
and docs/public-classification-suites.md.
A JevBench v1.4.1 run has been requested; the sealed-set result is not yet available.
Full report: docs/BENCHMARKS.md.
Model details
- Base model:
mistralai/Ministral-3-3B-Instruct-2512-BF16, revision b6d637bef2393152b3da2b2fde72eecdee30557e (Apache-2.0).
- Adapter: LoRA r=16, α=32, dropout 0, on
q_proj k_proj v_proj o_proj gate_proj up_proj down_proj of the language-model projections only (vision tower and multimodal projector excluded), 24,707,072 trainable parameters, PEFT 0.21.0. Adapter file adapter_model.safetensors, 135,113,048 bytes, sha256 a8e3eb341e27c1a49a282e327c2a2038906abb0eee771e80debbb4a4c45f7afc.
- Merged BF16 checkpoint (as published in
StandardThinking/StandardOne-3B): merging this adapter into the base changes 182 tensors, none outside the language-model projections, maximum absolute weight change 0.0023.
- Serving details: native chat-template wording, no system prompt, fixed per-answer-type temperatures (choice 0.95, noul 1.05, score 0.85; fitted on held-out and public-train calibration data, no JevBench item); served model name
standard-one-3b behind stock SGLang 0.5.20 via jev-adapter (POST /v1/systemone); single caller-supplied option order, no rotation ensemble; 8,192-token context.
| Path |
Contents |
adapter_model.safetensors, adapter_config.json |
The LoRA adapter |
merge.py |
Loads the base model, applies this adapter, saves the merged BF16 checkpoint |
docs/, docs/public-classification-suites.md |
Full benchmark report, figures (shared with 8B repo), per-language/per-workflow numbers |
SHA256SUMS, release-manifest.json, MERGE_REPORT.json, evidence/, LICENSE, README.md |
File hashes, training manifest, merge report, supporting artifacts, licence, this card |
The merged BF16 checkpoint is published in StandardThinking/StandardOne-3B; the server code and full
quick-start guide live in StandardThinking/StandardOne-8B.
Training data
Trains on the same data sources as StandardThinking/StandardOne-8B-LoRA, not a reduced subset. 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. Full
per-cohort breakdown: docs/BENCHMARKS.md.
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):
| 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 |
| Amazon Counterfactual |
CC BY 4.0 |
| DBpedia-14 |
CC BY-SA 3.0 |
| Dolly 15k |
CC BY-SA 3.0 |
| GoEmotions |
Apache-2.0 |
| MASSIVE |
CC BY 4.0 |
| Twitter Financial News Sentiment |
MIT |
| HelpSteer3 |
CC BY 4.0 |
| HelpSteer2 |
CC BY 4.0 |
| 2WikiMultihopQA |
Apache-2.0 |
| HotpotQA |
CC BY-SA 4.0 |
| MuSiQue |
CC BY 4.0 |
| QASC |
CC BY 4.0 |
| DROP |
CC BY-SA 4.0 |
| GSM8K |
MIT |
| TempReason |
CC BY-SA 3.0 |
| MultiNLI |
OANC / CC BY-SA 3.0 / CC BY 3.0 |
| PAWS |
Google terms, free for any purpose |
| PAWS-X |
Google terms, free for any purpose |
| SNLI |
CC BY-SA 4.0 |
| WANLI |
CC BY 4.0 |
| ContractNLI |
CC BY 4.0 |
| CUAD |
CC BY 4.0 |
| ShARC |
CC BY-SA 3.0 |
| Jailbreak classification |
Apache-2.0 |
| Prompt injections |
Apache-2.0 |
| Aegis AI Content Safety 2.0 |
CC BY 4.0 |
| Jigsaw Toxic Comment Classification (mirror of the Kaggle data) |
CC0 (data); comment text CC BY-SA 3.0 (Wikipedia) |
| Measuring Hate Speech |
CC BY 4.0 |
| Image safety classes |
MIT |
Upstream ids and the cohort each one feeds: docs/BENCHMARKS.md.
An exact-text overlap audit against the public JevBench tiers found 0 exact scenario matches and 181
exact instruction matches — rows in two adequacy-rubric cohorts whose entire instruction field, a
generic 58-character adequacy question, is byte-identical to one public hard-tier instruction (0.03 %
of the 520,754-row training mixture). These rows are kept and disclosed here rather than regenerated,
since the overlap is limited to one rubric question's wording and never touches a scenario or an
answer.
Limitations
- Public hard tier: the served 3B score is 46.85 %, versus 72.07 % for Jev 1.13. In the
separate offline base comparison, Standard One 3B scores 49.55 %, 1.80 percentage points above the untuned base's 47.75 %.
- Served probabilities are temperature-scaled by one value per answer type; if you apply this model to a
materially different question distribution, re-fitting that temperature is advisable rather than
assuming these values transfer.
- At most 26 options per question (one uppercase letter per option,
A–Z).
- The sealed JevBench set has not been measured for this model.
- Served and offline probabilities can differ on identical prompts (mean total-variation ≈0.08 on the
hard tier); served numbers are treated as authoritative.
- Korean is a small share of multilingual training alongside English.
- The card reports text benchmarks; it does not establish decision accuracy on image inputs.
- Ten-way support triage (42 %) and RAG passage relevance (57 %) are weak zero-shot; fine-tune
for those.
Licence
Adapter weights, merge recipe 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-LoRA (this repository, adapter + merge recipe) · StandardThinking/StandardOne-3B
(merged weights) · StandardThinking/StandardOne-8B (server code).
