Advanced NLP -- Assignment 2 checkpoints
| file | what it is |
|---|---|
part1_v1_dense.pt |
54.8 MB |
part1_v2_moe_4e_top1.pt |
54.9 MB |
part1_v3_moe_4e_top2.pt |
54.9 MB |
part1_v4_moe_1shared_3routed_top1.pt |
54.8 MB |
part1_v5_moe_4e_top2_iso_active.pt |
83.2 MB |
part2_adamw.pt |
54.8 MB |
part2_lion.pt |
54.8 MB |
part2_mars.pt |
54.8 MB |
part2_muon.pt |
54.8 MB |
part2_sophia.pt |
54.8 MB |
part1_*.pt-- decoder-only translation models (vi/ja -> en) that differ only in their feed-forward block (dense MLP / MoE variants).part2_*.pt-- the dense model pretrained on HAP-E with each hand-written optimizer (AdamW, MARS, Lion, Muon, Sophia).
Each checkpoint stores {"config": ..., "model": state_dict}; load it with
the code in the accompanying repository (src/part1/model.py).
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support