add README
Browse files
README.md
ADDED
|
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags: [anlp-assignment-2]
|
| 3 |
+
---
|
| 4 |
+
# ANLP Assignment 2 — model checkpoints
|
| 5 |
+
|
| 6 |
+
One folder per trained model; each holds `model.safetensors`, `config.json`,
|
| 7 |
+
`tokenizer.json` and a README describing the run.
|
| 8 |
+
|
| 9 |
+
| Folder | Contents |
|
| 10 |
+
|---|---|
|
| 11 |
+
| `part1/<variant>` | vi/ja→en translation LM; FFN variants `mlp`, `moe_top1`, `moe_top2`, `moe_shared`, `moe_top2_active` |
|
| 12 |
+
| `part2/<optimizer>` | causal LM trained with `adamw`, `mars`, `adam_mini`, `muon`, `sophia` |
|
| 13 |
+
|
| 14 |
+
Load with `load_model_folder` from `src/part1/evaluate.py` of the assignment repository.
|