thalovant-m2v-intents

thalovant/thalovant-m2v-intents

thalovant-m2v-intents is a text classification model from thalovant. Use it when you need a label for a piece of text. It is set up for model2vec. The card lists the license as apache-2.0.

A static-embedding intent classifier for the Thalovant hub. It maps an utterance to one of 364 labels of the form <skillid:<intent, exactly as ovos-m2v-pipeline registers them. The labels are the fleet's skills and th…

Downloads · 30 days

1.6K

34% of all-time downloads

All-time downloads

4.7K

Public

Parameters

30.6M

14.2 GB on disk

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.safetensors246 MB · 77%

At a glance

Task: Text Classification
Library: model2vec
License: apache-2.0
Access: Public
Created: Sep 8, 2026
Updated: Nov 15, 2026
SHA: 160c4da1

Try a prompt

Base models

Jarbas/m2v-256-distiluse-base-multilingual-cased-v2

Model card

From the Hugging Face model README

thalovant/thalovant-m2v-intents

A static-embedding intent classifier for the Thalovant hub. It maps an utterance to one of 364 labels of the form <skill_id>:<intent>, exactly as ovos-m2v-pipeline registers them. The labels are the fleet's skills and the other intents the hubs register (the persona and reading pipelines, wikihow), from the fleet corpus and nothing else. There is no "none of these" label: a sentence about something no skill does still gets the nearest skill's label, often with high confidence.

Base: Jarbas/m2v-256-distiluse-base-multilingual-cased-v2, fine-tuned with model2vec's classifier trainer.

How good it is

Measured on sentences the classifier never saw: 100118 held-out rows whose text, in any language, and whose intent line are both absent from the 412898 rows an evaluation model was trained on. A split per row would score the en-GB twin of an en-US training sentence as new.

Intent accuracy 91.3 %, skill accuracy 95.8 % across 68 languages; median language 90.4 %, 7 language(s) under 80 %.
0.9 % of held-out sentences are given to another skill at 0.9 or more: what thalovant-skillkit check would report about a sentence that is in fact new and the skill's own.
Languages that need a higher threshold to keep that under 1.0 %: az-AZ 0.92, cs-CZ 0.92, el-GR 0.92, et-EE 0.99, eu-ES 0.94, fi-FI 0.95, he-IL 0.92, hi-IN 0.98, hu-HU 0.98, id-ID 0.98, ja-JP 0.92, ko-KR 0.94, lv-LV 0.98, nb-NO 0.94, pl-PL 0.995, pt-AO 0.96, pt-BR 0.96, pt-MZ 0.96, pt-PT 0.96, sk-SK 0.92, sl-SI 0.92, sv-FI 0.92, sv-SE 0.92, sw-KE 0.98, th-TH 0.92, tr-TR 0.95, uk-UA 0.99, vi-VN 0.95 (thresholds in training.json).

The published weights are then trained with the same recipe on all 513016 rows, so a skill's existing sentences are the classifier's training data. At most 512 sentences per label and language are trained on (the reading pipeline's English intents alone expand to thousands); index.json keeps all 557929.

Use

from model2vec.inference import StaticModelPipeline

model = StaticModelPipeline.from_pretrained("thalovant/thalovant-m2v-intents")
model.predict(["will it rain tomorrow"])

Configure the pipeline with "mode": "classifier" and this model's path; labels.json beside the weights lists the labels the hub may route to.

index.json names every sentence in the training corpus by a digest of its language and text, with the labels that publish it. thalovant-skillkit check uses it to fail a skill that publishes a sentence another skill already owns, and asks this classifier what it makes of the rest -- without the fleet's sentences leaving their private repositories.

Held-out results per language

False warnings: held-out sentences given to another skill at 0.9 or more. Threshold: the lowest that keeps them under 1.0 %.

Language	Rows	Intent	Skill	False warnings	Threshold
`an-ES`	928	91.9 %	97.0 %	0.4 %	0.9
`ar-SA`	728	82.7 %	92.7 %	1.0 %	0.9
`ast-ES`	688	95.6 %	97.4 %	0.3 %	0.9
`az-AZ`	500	80.4 %	89.8 %	1.6 %	0.92
`bg-BG`	587	90.1 %	95.6 %	0.7 %	0.9
`ca-ES`	3510	93.6 %	97.0 %	0.7 %	0.9
`cs-CZ`	974	83.8 %	92.7 %	1.0 %	0.92
`da-DK`	1244	89.8 %	94.4 %	0.9 %	0.9
`de-AT`	2529	93.7 %	96.7 %	0.7 %	0.9
`de-CH`	2471	93.7 %	96.7 %	0.7 %	0.9
`de-DE`	3132	91.9 %	96.4 %	0.6 %	0.9
`el-GR`	1087	85.8 %	91.1 %	1.3 %	0.92
`en-AU`	2877	94.4 %	98.1 %	0.2 %	0.9
`en-CA`	2818	94.3 %	98.1 %	0.2 %	0.9
`en-GB`	2877	94.4 %	98.1 %	0.2 %	0.9
`en-IE`	45	93.3 %	93.3 %	2.2 %	0.9
`en-NZ`	2877	94.4 %	98.1 %	0.2 %	0.9
`en-US`	3800	94.6 %	98.3 %	0.2 %	0.9
`es-AR`	1925	95.8 %	98.1 %	0.3 %	0.9
`es-CO`	2000	95.8 %	98.0 %	0.4 %	0.9
`es-ES`	2696	95.9 %	97.7 %	0.6 %	0.9
`es-MX`	1983	95.8 %	98.0 %	0.4 %	0.9
`es-US`	1983	95.8 %	98.0 %	0.4 %	0.9
`et-EE`	548	80.1 %	88.7 %	3.5 %	0.99
`eu-ES`	1233	83.5 %	91.2 %	1.5 %	0.94
`fa-IR`	943	81.7 %	92.6 %	0.6 %	0.9
`fi-FI`	668	83.7 %	90.7 %	1.5 %	0.95
`fr-BE`	2710	95.5 %	97.5 %	0.7 %	0.9
`fr-CA`	2735	95.6 %	97.4 %	0.7 %	0.9
`fr-CH`	2710	95.5 %	97.5 %	0.7 %	0.9
`fr-FR`	3257	95.3 %	97.7 %	0.5 %	0.9
`gl-ES`	2542	96.7 %	98.3 %	0.2 %	0.9
`he-IL`	521	85.2 %	91.5 %	1.1 %	0.92
`hi-IN`	2170	78.6 %	87.8 %	2.1 %	0.98
`hr-HR`	558	88.0 %	94.1 %	0.5 %	0.9
`hu-HU`	896	80.8 %	90.2 %	1.9 %	0.98
`id-ID`	665	87.1 %	95.5 %	1.4 %	0.98
`it-CH`	61	88.5 %	90.2 %	4.9 %	0.9
`it-IT`	2603	93.2 %	97.2 %	0.7 %	0.9
`ja-JP`	679	69.7 %	92.9 %	1.2 %	0.92
`kab`	72	90.3 %	97.2 %	2.8 %	0.9
`ko-KR`	863	59.6 %	88.6 %	1.5 %	0.94
`lt-LT`	547	77.0 %	88.8 %	0.9 %	0.9
`lv-LV`	536	80.4 %	86.4 %	2.1 %	0.98
`ms-MY`	577	90.3 %	94.6 %	0.5 %	0.9
`nb-NO`	701	90.9 %	96.3 %	1.3 %	0.94
`nl-BE`	1496	92.2 %	95.7 %	0.7 %	0.9
`nl-NL`	2262	93.6 %	96.4 %	0.7 %	0.9
`oc-FR`	726	86.8 %	95.9 %	0.8 %	0.9
`pl-PL`	1135	73.2 %	78.8 %	6.6 %	0.995
`pt-AO`	2645	95.1 %	96.7 %	1.5 %	0.96
`pt-BR`	1474	92.2 %	95.0 %	1.7 %	0.96
`pt-MZ`	2645	95.1 %	96.7 %	1.5 %	0.96
`pt-PT`	3139	95.0 %	96.9 %	1.3 %	0.96
`ro-RO`	678	87.6 %	93.4 %	0.6 %	0.9
`ru-RU`	1024	89.8 %	95.9 %	0.9 %	0.9
`sk-SK`	878	83.8 %	94.7 %	1.1 %	0.92
`sl-SI`	661	90.2 %	94.4 %	1.1 %	0.92
`sv-FI`	728	90.4 %	94.6 %	1.0 %	0.92
`sv-SE`	853	88.6 %	94.0 %	1.2 %	0.92
`sw-KE`	515	78.5 %	88.7 %	2.1 %	0.98
`th-TH`	936	81.2 %	88.3 %	1.3 %	0.92
`tr-TR`	628	81.0 %	90.5 %	1.8 %	0.95
`uk-UA`	1086	77.5 %	92.4 %	2.5 %	0.99
`vi-VN`	654	86.1 %	91.4 %	2.0 %	0.95
`zh-CN`	776	97.2 %	99.1 %	0.0 %	0.9
`zh-SG`	776	97.2 %	99.1 %	0.0 %	0.9
`zh-TW`	1349	96.4 %	98.4 %	0.1 %	0.9

Held-out results per skill

Skill	Rows	Intent	Skill	False warnings
`ovos-common-reading-pipeline-plugin`	695	95.8 %	99.0 %	0.0 %
`ovos-persona-pipeline-plugin`	1576	94.0 %	94.5 %	0.8 %
`ovos-skill-wikihow.openvoiceos`	387	95.3 %	95.6 %	1.3 %
`thalovant-skill-alarm.thalovant`	1975	94.4 %	97.6 %	0.7 %
`thalovant-skill-camera.thalovant`	693	76.0 %	83.7 %	2.7 %
`thalovant-skill-christmas-companion.thalovant`	457	79.9 %	89.1 %	1.3 %
`thalovant-skill-christmas-games.thalovant`	455	98.9 %	99.6 %	0.0 %
`thalovant-skill-christmas-protocol.thalovant`	169	79.9 %	97.6 %	0.0 %
`thalovant-skill-coming-soon.thalovant`	2975	97.7 %	97.8 %	0.2 %
`thalovant-skill-custos-query.thalovant`	340	84.4 %	89.1 %	2.4 %
`thalovant-skill-custos-shadow.thalovant`	291	95.2 %	95.9 %	1.4 %
`thalovant-skill-date-time.thalovant`	14071	96.2 %	98.9 %	0.1 %
`thalovant-skill-easter-eggs.thalovant`	1542	85.8 %	89.0 %	1.9 %
`thalovant-skill-fart.thalovant`	2130	85.2 %	93.9 %	1.5 %
`thalovant-skill-guide.thalovant`	2944	85.6 %	92.7 %	0.9 %
`thalovant-skill-home.thalovant`	18230	87.0 %	94.5 %	1.7 %
`thalovant-skill-interaction-modes.thalovant`	468	82.3 %	92.3 %	1.3 %
`thalovant-skill-joke-garden.thalovant`	2683	91.3 %	93.7 %	0.9 %
`thalovant-skill-language-buddy.thalovant`	421	81.2 %	81.2 %	1.9 %
`thalovant-skill-learning-lounge.thalovant`	452	83.2 %	83.2 %	4.9 %
`thalovant-skill-lists.thalovant`	85	68.2 %	95.3 %	0.0 %
`thalovant-skill-local-pulse.thalovant`	1783	94.2 %	94.2 %	1.1 %
`thalovant-skill-media.thalovant`	7981	96.5 %	98.9 %	0.1 %
`thalovant-skill-music-assistant.thalovant`	2726	83.8 %	96.2 %	0.3 %
`thalovant-skill-news.thalovant`	1811	96.6 %	97.7 %	0.3 %
`thalovant-skill-on-this-day.thalovant`	1277	97.1 %	97.2 %	0.2 %
`thalovant-skill-ops-copilot.thalovant`	693	87.0 %	87.0 %	1.9 %
`thalovant-skill-pop-facts.thalovant`	3253	91.7 %	95.9 %	1.3 %
`thalovant-skill-pop-quiz.thalovant`	3227	99.0 %	99.0 %	0.2 %
`thalovant-skill-recall.thalovant`	1430	87.5 %	92.7 %	0.8 %
`thalovant-skill-reminder.thalovant`	1615	89.2 %	94.2 %	1.5 %
`thalovant-skill-rewrite-forge.thalovant`	1632	93.0 %	93.0 %	2.5 %
`thalovant-skill-routines.thalovant`	1097	87.5 %	99.5 %	0.0 %
`thalovant-skill-safety-guide.thalovant`	494	87.5 %	87.5 %	2.4 %
`thalovant-skill-source-scout.thalovant`	690	83.5 %	83.5 %	2.9 %
`thalovant-skill-stopwatch.thalovant`	1012	85.6 %	94.5 %	0.3 %
`thalovant-skill-timer.thalovant`	1643	94.2 %	97.3 %	0.6 %
`thalovant-skill-twenty-questions.thalovant`	263	90.9 %	90.9 %	0.4 %
`thalovant-skill-volume.thalovant`	1275	81.1 %	94.9 %	0.7 %
`thalovant-skill-weather.thalovant`	10192	90.9 %	96.7 %	0.8 %
`thalovant-skill-what-next.thalovant`	2985	97.0 %	97.0 %	0.6 %

What the classifier confuses

Each line is a held-out sentence whose true intent the classifier read as another intent. Two skills on one line is a collision the fleet should resolve.

thalovant-skill-home.thalovant:home.turn.on read as thalovant-skill-home.thalovant:home.turn.off (347): "ممكن تشغل الضوء في something" (ar-SA)
thalovant-skill-home.thalovant:home.turn.on read as thalovant-skill-media.thalovant:media.play.title (248): "enzende a televisión de something" (an-ES)
thalovant-skill-home.thalovant:home.state read as thalovant-skill-home.thalovant:home.turn.off (195): "son as luzes de something amortadas" (an-ES)
thalovant-skill-home.thalovant:home.state read as thalovant-skill-home.thalovant:home.turn.on (161): "ventilyator işləyir mi" (az-AZ)
thalovant-skill-home.thalovant:home.state read as thalovant-skill-weather.thalovant:air_quality.weather (136): "공기청정기 꺼져 있나요" (ko-KR)
thalovant-skill-date-time.thalovant:weekday.for.date read as thalovant-skill-date-time.thalovant:holiday.when (101): "an welchem tag ist weihnachten dieses jahr" (de-AT)
thalovant-skill-home.thalovant:home.turn.on read as thalovant-skill-home.thalovant:home.brightness (90): "enzende a luz de something" (an-ES)
thalovant-skill-weather.thalovant:condition.weather read as thalovant-skill-weather.thalovant:snow.weather (89): "budou bouře?" (cs-CZ)
thalovant-skill-weather.thalovant:alerts.weather read as thalovant-skill-weather.thalovant:condition.weather (77): "are there any storm advisories today" (en-AU)
thalovant-skill-home.thalovant:home.state read as thalovant-skill-weather.thalovant:wind.weather (55): "cik silts ir šeit" (lv-LV)
thalovant-skill-weather.thalovant:condition.weather read as thalovant-skill-weather.thalovant:sunrise.weather (53): "habrá tormenta esta noche" (es-AR)
thalovant-skill-home.thalovant:home.state read as thalovant-skill-home.thalovant:home.cover.open (51): "are the blinds open" (en-AU)
thalovant-skill-home.thalovant:home.turn.off read as thalovant-skill-home.thalovant:home.state (51): "kondisioneri qapat" (az-AZ)
thalovant-skill-home.thalovant:home.state read as thalovant-skill-home.thalovant:home.thermostat (49): "o aire acondicionado de something está acendido" (gl-ES)
thalovant-skill-joke-garden.thalovant:party.host read as thalovant-skill-joke-garden.thalovant:joke.garden (47): "udělej nám skupinovou výzvu." (cs-CZ)

Trained from the corpus built 2026-11-15T16:02:14Z covering 41 skills and sources, expanded by thalovant-skillkit 0.38.0.

Task: Text Classification
Library: model2vec
License: apache-2.0
Created: Sep 8, 2026
Updated: Nov 15, 2026