| batch_size: 256 | |
| checkpoint_dir: checkpoints | |
| data: data/processed/classic_triplets.parquet | |
| device: cpu | |
| embedding: | |
| embedding_dim: 64 | |
| type: lookup | |
| encoder: | |
| arch: mean | |
| hidden_dim: 128 | |
| tied_weights: true | |
| epochs: 3 | |
| huggingface: | |
| private: false | |
| push_to_hub: true | |
| repo_id: mlx7-two-tower-retrieval | |
| learning_rate: 1e-3 | |
| loss: | |
| margin: 0.2 | |
| type: triplet | |
| max_sequence_length: 64 | |
| optimizer: | |
| lr: 0.001 | |
| type: adamw | |
| tokeniser: | |
| max_len: 64 | |
| type: char | |
| use_wandb: true | |
| wandb: | |
| entity: azuremis | |
| project: two-tower-retrieval | |