model: name: openbmb/MiniCPM5-1B attn_implementation: flash_attention_3 chat_template: /mnt/vast/maxence/code-search/configuration/training/templates/minicpm5_generation.jinja datasets: - name: /mnt/vast/maxence/code-search/artifacts/training/rebalanced15k-v2/grep/weighted.jsonl split: train validation: split_ratio: 0.0 seed: 42 training: output_dir: /mnt/vast/maxence/code-search/artifacts/models/grep-rebalanced15k-v2-weighted-minicpm5-1b num_train_epochs: 1 per_device_train_batch_size: 4 gradient_accumulation_steps: 8 learning_rate: 2.0e-05 lr_scheduler_type: cosine warmup_ratio: 0.03 weight_decay: 0.0 max_grad_norm: 1.0 bf16: true max_seq_length: 32768 packing: true assistant_only_loss: true gradient_checkpointing: true use_liger_kernel: true optim: adamw_torch logging_steps: 10 save_strategy: steps save_steps: 500 eval_strategy: 'no' report_to: none dataset_num_proc: 8 wandb: enabled: false project: code-search-sft run_name: grep-rebalanced15k-v2-weighted-minicpm5-1b