Views
No views yet
P1 seed 42P142amosluna/babylm-2026-p1-seed42eng=0.321, nld=0.323, zho=0.35665536000010089.1326816082 sruns/2026-05-21_P1_seed4224 intermediate checkpoints as branches following
the BabyLM 2026 naming convention: chck_1M, chck_2M, ..., chck_10M, chck_20M, ..., chck_100M, chck_200M, ..., chck_600M. The eval pipeline at
babylm-org/babylm-eval pulls
these revisions automatically with:1bash multilingual/scripts/zeroshot_model_fast_all.sh \
2 --model_name amosluna/babylm-2026-p1-seed421from transformers import AutoModelForCausalLM, AutoTokenizer
2
3tok = AutoTokenizer.from_pretrained("amosluna/babylm-2026-p1-seed42")
4model = AutoModelForCausalLM.from_pretrained("amosluna/babylm-2026-p1-seed42") # final checkpoint
5# Intermediate checkpoint:
6# model = AutoModelForCausalLM.from_pretrained("amosluna/babylm-2026-p1-seed42", revision="chck_100M")normalized_excess_loss (v1, delta-based) and
cross_lingual_deficit (v2, level-based).1@inproceedings{taam2026,
2 title = {Typology-Aware Adaptive Mixing for Multilingual BabyLMs},
3 author = {Luna, Amos and collaborators},
4 year = {2026},
5 booktitle = {Proceedings of the BabyLM Workshop at EMNLP 2026}
6}