CoolFace
Modelpublic

afk-live/afk-setfit-camembert-large-v1

sourceHugging Facemitupdated 8mo agoView on Hugging Face
0likes13downloads
Model Card

afk-setfit-camembert-large-v1

French news headline classifier for AFK.live, fine-tuned with SetFit.

Quick Start

python
from setfit import SetFitModel

model = SetFitModel.from_pretrained("afk-live/afk-setfit-camembert-large-v1")
model.predict(["Macron annonce une réforme majeure des retraites"])
# => ["Politique France"]

Model Details

AttributeValue
Base Modeldangvantuan/sentence-camembert-large
Accuracy80.09%
Mean F10.7954
Training Time1576s
Datasetafk-live/afk-news-fr-classification-202601

Categories (12 labels)

IDLabelDescriptionLive
1Politique FR, EU, QuébecDomestic francophone politics: France, Belgium, Switzerland, Quebec, EU institutionsYes
2Géopolitique mondialeInternational relations, conflicts, diplomacy outside francophone sphereYes
3Economie & Marchés & Business TechPure finance, macroeconomics, non-tech corporations, banking, marketsYes
4SociétéYes
5Science & EnvironnementResearch discoveries, climate, biodiversity, health research (non-clickbait)Yes
6CultureCultural works and events: cinema, music, books, theater, visual arts, video gamesYes
7SportAll sports news (competitions, transfers, results)Yes
8Faits divers & Actualité localeAccidents, local news without national relevanceNo
10Gadgets, pub, articles sponsorisésMode, cooking, travel, wellness, fashion, product reviews, dealsNo
11MétéoWeather forecasts, daily weather reportsNo
12Rediffusion, replay, programme TVTV schedules, replay announcements, program gridsNo
13Code & InfrastructureDeveloper-focused technical content: programming, ML engineering, DevOps, software architectureYes

Note: "Live" indicates whether the category is displayed on AFK.live (Yes) or filtered out (No).

Training Details

  • —Train samples: 720 (60 per class)
  • —Test samples: 211
  • —Hyperparameters: batchsize=(32,16), epochs=(1,16), endto_end=True, sampling=unique

About AFK.live

AFK.live is a French news aggregation project that creates trends from journalistic work.

Last updated: 2026-01-17