| model: | |
| base_model: Qwen/Qwen3-0.6B | |
| max_length: 512 | |
| pooling: last_token | |
| dropout: 0.05 | |
| freeze_backbone: true | |
| dtype: auto | |
| training: | |
| seed: 42 | |
| epochs: 5 | |
| batch_size: 8 | |
| gradient_accumulation_steps: 1 | |
| learning_rate: 0.001 | |
| weight_decay: 0.01 | |
| max_grad_norm: 1.0 | |
| soft_target_temperature: 0.15 | |
| log_every: 10 | |