Arihant25/anlp-a2-moe

Checkpoints for Advanced NLP Assignment 2 (Monsoon 2026). Each folder has model.safetensors, config.json (src.model.TransformerConfig), tokenizer.json and meta.json. Load with src.utils.load_checkpoint(folder) from the assignment repository.

run	total params	active params	training tokens
p1-dense	37.46M	37.46M	150028257
p1-moe4-top1	37.48M	24.90M	150028257
p1-moe4-top2	37.48M	29.09M	150028257
p1-moe4-top2-active	54.26M	37.48M	150028257
p1-shared1-moe3-top1	37.47M	29.09M	150028257

Training logs: https://wandb.ai/arihanttr-iiit-hyderabad/anlp-a2