python-llm-phase0_5 / phase2_report.md
Hak5's picture
Upload SFT artifact (adapter_only) from phase2-codellama-7b-lora-kaggle-1h
868831d verified

Phase 2 SFT Report

  • Run directory: /kaggle/working/python-llm-grpo-new-techniques/training/output/phase2-codellama-7b-lora-kaggle-1h
  • Trainer state: /kaggle/working/python-llm-grpo-new-techniques/training/output/phase2-codellama-7b-lora-kaggle-1h/checkpoint-43/trainer_state.json

Metrics

  • global_step: 43
  • best_metric: None
  • last_train_loss: 0.7746126174926757
  • best_eval_loss: None
  • last_eval_loss: None
  • last_learning_rate: 0.00015600000000000002
  • last_step_logged: 40
  • num_train_loss_logs: 4
  • num_eval_loss_logs: 0