| encoder: |
| name: jhu-clsp/ettin-encoder-17m |
| max_length: 512 |
| dtype: bfloat16 |
| trust_remote_code: false |
| decoder: |
| d_model: 64 |
| n_layer: 2 |
| n_head: 4 |
| d_ff: 256 |
| max_seq_len: 128 |
| prefix_len: 1 |
| dropout: 0.0 |
| tie_embeddings: true |
| data: |
| hf_name: OLMo-Coding/starcoder-python-instruct |
| split: train |
| val_fraction: 0.1 |
| max_identifiers: 26 |
| max_tokens: 128 |
| instruction_field: instruction |
| code_field: text |
| stdlib_whitelist_path: null |
| num_proc: 16 |
| cache_dir: /workspace-SR008.nfs2/nsushko/nsushko/polygon/tiny_decoder/.cache/tiny_decoder/shared |
| vocab: |
| path: configs/vocab_98.yaml |
| builtin_aliases_path: configs/builtin_aliases.yaml |
| attr_aliases_path: configs/attr_aliases.yaml |
| train: |
| batch_size: 128 |
| epochs: 50 |
| lr: 0.0003 |
| warmup_frac: 0.05 |
| weight_decay: 0.1 |
| grad_clip: 1.0 |
| early_stop_patience: 5 |
| seed: 0 |
| log_every: 50 |
| eval_every_epochs: 1 |
| wandb_project: null |
| tensorboard_dir: null |
| eval: |
| split: val |
| max_samples: null |
| compute_exact_match: true |
| compute_syntactic_validity: true |
| run: |
| dir: /workspace-SR008.nfs2/nsushko/nsushko/polygon/tiny_decoder/runs/prefix_k1 |
| save_every_epochs: 1 |
|
|
