ionic21/anlp-a2-checkpoints

ANLP Assignment 2 checkpoints

PyTorch state dicts for https://code.iiit.ac.in/advanced-nlp/m26-assignment2-ayush.daga

• part1/4ep/: MoE ablations c1 to c5, 4 epochs (final runs, router bug fixed)
• part1/4ep_buggy/: c2 and c4 at 4 epochs before the k=1 router gate fix
• part1/1ep/: c1 to c5, 1 epoch (c2 and c4 still have the router bug)
• part1/lr_sweep/: dense LR sweep at batch 64, 2000 steps
• part2/: dense c1 model pretrained on browndw/human-ai-parallel-corpus with AdamW, MARS, Lion, Muon, Sophia (1x dataset)
• part2/sweep/: LR sweep runs at 0.1x the dataset
• part2/smoke/: 1% smoke tests

Part 1 files hold {"cfg", "model", "args"}; load with Transformer(ck["cfg"]).load_state_dict(ck["model"]) from src/part1/model.py. Part 2 files are plain model.state_dict() for CONFIGS["c1_dense"].