| model_name: Llama-3.2-1B | |
| out_dir: /home/aiops/zhuty/litgpt_out/pretrain/llama3-1b-proweb-50ksteps-diffusion | |
| precision: bf16-mixed | |
| initial_checkpoint_dir: checkpoints/meta-llama/Llama-3.2-1B | |
| resume: auto | |
| data: | |
| class_path: litgpt.data.TextFiles | |
| init_args: | |
| train_data_path: /home/aiops/zhuty/cont_data/proweb/train | |
| val_data_path: /home/aiops/zhuty/cont_data/proweb/test | |
| seed: 42 | |
| num_workers: 0 | |
| add_eos: true | |
| train: | |
| save_interval: 2500 | |
| save_optimizer_state: true | |
| max_optimizer_state: 1 | |
| log_interval: 1 | |
| global_batch_size: 1024 | |
| micro_batch_size: 8 | |
| lr_warmup_fraction: 0.01 | |
| max_steps: 50000 | |
| max_seq_length: 1024 | |
| max_norm: 1.0 | |
| min_lr: 5.0e-06 | |
| eval: | |
| interval: 1000 | |
| max_iters: 100 | |
| initial_validation: true | |
| final_validation: true | |
| evaluate_example: first | |
| num_generation_examples: 1 | |
| calculate_exact_match: false | |
| log: | |
| project: mathcont | |
| optimizer: | |
| class_path: torch.optim.AdamW | |
| init_args: | |
| lr: 5.0e-05 | |
| weight_decay: 0.1 | |
| betas: | |
| - 0.9 | |
| - 0.95 | |
| devices: auto | |
| num_nodes: 1 | |
| tokenizer_dir: checkpoints/meta-llama/Llama-3.2-1B | |
| logger_name: wandb | |
| seed: 42 | |
| compiler: torch | |
| executors: | |
| - sdpa | |
| - torchcompile | |
| - torch | |
| strategy: fsdp | |
| diffusion: true | |
| mask_token_id: 811 | |
| sampling_eps: 0.001 | |
| intradoc: false | |
| block_diffusion: false | |
| block_size: 8 | |
| timestep_sampler: uniform | |