Beko2210/statim-decide-multilingual-base-emotion

Statim Decide Multilingual Base: Emotion adapter

Licence notice (2026-10-03). The base model this adapter requires (multilingual 0.7.0) was trained partly on data that is non-commercial, ShareAlike or under an unknown licence. From 2026-10-03 the base and this adapter are offered only under PolyForm Noncommercial 1.0.0; the Small Business and Free Trial licences and the commercial licence do not apply to them. A replacement trained only on cleared data will follow. Details: DATA_LICENSES.md.

A LoRA adapter that improves the Emotion decisions of Beko2210/statim-decide-multilingual-base 0.7.0. Checked with Statim 0.8.3: the published f32 file loads it merged at load, the q8_0 file as runtime LoRA. LoRA adapters need Statim 0.8.0 or later.

Quick start

hf download Beko2210/statim-decide-multilingual-base statim-decide-multilingual-base-q8_0.gguf --local-dir models
hf download Beko2210/statim-decide-multilingual-base-emotion statim-decide-multilingual-base-emotion.lora.gguf --local-dir models
statim serve -m multilingual=models/statim-decide-multilingual-base-q8_0.gguf --adapter multilingual:emotion=models/statim-decide-multilingual-base-emotion.lora.gguf --port 8080

curl -s localhost:8080/v1/systemone -d '{"state": {"text": "They cancelled my flight again and nobody even apologised!"}, "questions": {"emotion": {"type": "choice", "instructions": "Which emotion is expressed most strongly?", "criteria": ["anger", "disgust", "fear", "joy", "sadness", "surprise"]}}, "adapter": "emotion"}'
curl -s localhost:8080/v1/systemone -d '{"state": {"text": "They cancelled my flight again and nobody even apologised!"}, "questions": {"emotion": {"type": "choice", "instructions": "Which emotion is expressed most strongly?", "criteria": ["anger", "disgust", "fear", "joy", "sadness", "surprise"]}}, "adapter": "auto"}'

Python (Python SDK 0.8.3 or later):

from statim import Client
client = Client("http://127.0.0.1:8080")
client.decide({"text": "They cancelled my flight again and nobody even apologised!"}, {"emotion": {"type": "choice", "instructions": "Which emotion is expressed most strongly?", "criteria": ["anger", "disgust", "fear", "joy", "sadness", "surprise"]}}, adapter="emotion")

Results (experiment's f32 run)

Each cell uses 150 items (seed 20260927).

Language	Base	Adapter	Change (points)	Verdict	Qwen3-8B zero-shot
de	0.4533	0.4600	+0.67	within noise	0.587
en	0.5733	0.6200	+4.67	within noise	0.693
es	0.5733	0.5800	+0.67	within noise	0.740
fr	0.6600	0.7867	+12.67	gain (2 SE)	0.800
hi	0.7133	0.8267	+11.34	gain (2 SE)	0.880
pt	0.5067	0.5400	+3.33	within noise	0.587
ru	0.6933	0.7200	+2.67	within noise	0.887
zh	0.5400	0.5600	+2.00	within noise	0.653
Mean	0.5892	0.6367	+4.75	—	0.728

The pooled family change is +4.75 points with 2 SE = 3.88 points (gain).

A single cell of 150 items rarely clears 2 SE on its own; the decision uses the pooled family.

Decision rule: promote when the category family gains more than 2 standard errors, pooled by rows or by suites, and nothing regresses. A regression is a pooled drop beyond 2 standard errors under either pooling, or a drop in one language cell that stays significant after Holm-Bonferroni (family-wise 5 %).

Statim is trained on this category, while Qwen3-8B runs zero-shot.

Checked on the published files

Weights	Adapter mode	Base mean	Adapter mean	Change (points)	Cell for cell as in the experiment
f32	merged at load	0.5892	0.6367	+4.75	yes
q8_0	runtime LoRA	0.5892	0.6425	+5.33	— (experiment: f32)

Training

Source	Rows	Licence
Johnson8187/Chinese_Multi-Emotion_Dialogue_Dataset (default)	6,200	MIT
JusteLeo/French-emotion (default)	6,200	MIT
brighter-dataset/BRIGHTER-emotion-categories (hin)	3,629	CC-BY-4.0
brighter-dataset/BRIGHTER-emotion-categories (mar)	3,768	CC-BY-4.0
github:wwbp/empathic_reactions (data/responses/data/messages.csv)	3,719	CC-BY-4.0
jmccardle/pulse-sofroniew-emotion-concept-texts (default)	6,200	CC-BY-4.0
llm-for-emotion/Cultural-Emo (`ara	deu	eng
shreyaspullehf/emotion-dataset-20-emotions (default)	6,200	MIT
tanaos/synthetic-emotion-detection-dataset-v1 (default)	6,200	MIT

LoRA rank 16, alpha 32.0, dropout 0.05; target modules Wqkv, Wo, Wi; 88 wrapped modules and 3,379,200 trainable parameters.

• Items: 45,714 train, 401 dev.
• Updates: 984.
• Dev accuracy: 0.6309 before, 0.7007 after.
• Time: 1124.1 seconds; peak memory: 2,319 MB.

Provenance

• Adapter GGUF SHA-256: c832dc6aada9bf0b07e7f481b554d51ddff2fe0c1365ba49145de13f3952bdc8
• PEFT safetensors SHA-256: a347899cb7e45d1130add3de9fddb278705d7ce8ea0901f7412c5dd1b0cefd34
• Training checkpoint model.safetensors SHA-256: c44425f14ac9d55508f73a6f371e4e2802ed59e646287c3abbec10b653a19840
• Base fingerprint: e7a8fa743b4920850167b587226e339d651be0244fcc6a9e24fff8ed5073bae5
• Training mixture SHA-256: b8a87e85f3bf72b61509555aba9148358760d3329ecf37a36cebd94f9358cb03
• Source registry SHA-256: 309ce8ec262886b3bfaa529934151ae0c1879595c26301e4269da276d95cbfcd

Experiment commands (paths relative to the Statim repository):

• train: '.venv-train/bin/python' 'tools/finetune/train_lora.py' 'models/laya-multilingual-v9' --mixture 'data/mixture-v8.jsonl.gz' --category emotion --registry 'tools/finetune/sources/v6-keep.json' --out 'models/lora-exp1/emotion' --device cuda --epochs 2
• convert: '.venv/bin/python' 'tools/convert_lora.py' 'models/lora-exp1/emotion' -o 'models/lora-exp1/emotion.lora.gguf' --base 'models/laya-multilingual-v9-f32.gguf' --category emotion --name emotion
• serve: 'build-vk/statim' serve -m 'multilingual=models/laya-multilingual-v9-f32.gguf' --adapter 'multilingual:emotion=models/lora-exp1/emotion.lora.gguf' --device vulkan --threads 16 --port 8098 --no-access-log --inference-timeout 600
• eval_base: '.venv/bin/python' 'bench/eval_categories.py' --strict --url http://127.0.0.1:8098 --model multilingual --suites emotion --n 150 --seed 20260927 --exclude-mixture 'data/mixture-v8.jsonl.gz' --out 'models/lora-exp1/emotion.base.jsonl'
• eval_adapter: '.venv/bin/python' 'bench/eval_categories.py' --strict --url http://127.0.0.1:8098 --model multilingual --suites emotion --n 150 --seed 20260927 --exclude-mixture 'data/mixture-v8.jsonl.gz' --adapter emotion --out 'models/lora-exp1/emotion.adapter.jsonl'

Protocol and experiment: docs/ADAPTERS.md. Full reproduction instructions: REPRODUCE.md.

Intended use and limits

• This adapter only helps its category; route requests with "emotion" or "auto".
• Bound to Beko2210/statim-decide-multilingual-base 0.7.0 by the base fingerprint e7a8fa743b492085… (statim.lora.base_fingerprint, SHA-256 over the checkpoint's norm and bias tensors); Statim refuses the adapter on a base whose fingerprint differs.
• Languages outside the evaluated list are untested.
• Each language cell has 150 items.
• Do not automate decisions about people without human review.
• Evaluated languages with no category training data: pt, ru.

Licence

The weights may be used under any one of: PolyForm Noncommercial 1.0.0, PolyForm Small Business 1.0.0 (free commercial use below 100 people and 1 M USD revenue), PolyForm Free Trial 1.0.0 (any company, fewer than 32 days), or a Statim commercial licence (COMMERCIAL.md). Texts in LICENSE-MODEL.md. The Statim engine is Apache-2.0.

Training data attribution is listed source by source above, with the row count and licence read from the experiment registry.