We present Kenji-Endo, a BabyLM pretrained on a dedicated Italian dataset, which participated in four tasks at the 9th edition of EVALITA: DeSegMa, MultiPRIDE, IMPOLS, and FadeIT. Kenji-Endo achieved competitive performance across all tasks, demonstrating that language modeling with limited data and compact model sizes can represent a viable alternative to Large Language Models.
Kenji-Endo: a BabyLM @EVALITA
Calogero Jerik Scozzaro
First
;Matteo Rinaldi;Gianluca Mittone;Marco Antonio StranisciLast
2026-01-01
Abstract
We present Kenji-Endo, a BabyLM pretrained on a dedicated Italian dataset, which participated in four tasks at the 9th edition of EVALITA: DeSegMa, MultiPRIDE, IMPOLS, and FadeIT. Kenji-Endo achieved competitive performance across all tasks, demonstrating that language modeling with limited data and compact model sizes can represent a viable alternative to Large Language Models.
File in questo prodotto:
| File | Dimensione | Formato | |
|---|---|---|---|
|
36.pdf
Accesso aperto
Tipo di file:
PDF EDITORIALE
Dimensione
234.36 kB
Formato
Adobe PDF
|
234.36 kB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
