| model: | |
| _component_: torchtune.models.qwen2_5.qwen2_5_0_5b | |
| tokenizer: | |
| _component_: torchtune.models.qwen2_5.qwen2_5_tokenizer | |
| path: qwen2_5.qwen2_5_0_5b_instruct/vocab.json | |
| merges_file: qwen2_5.qwen2_5_0_5b_instruct/merges.txt | |
| max_seq_len: null | |
| checkpointer: | |
| _component_: torchtune.training.FullModelHFCheckpointer | |
| checkpoint_dir: qwen2_5.qwen2_5_0_5b_instruct | |
| checkpoint_files: | |
| - model.safetensors | |
| recipe_checkpoint: null | |
| output_dir: qwen2_5.qwen2_5_0_5b_instruct | |
| model_type: QWEN2 | |
| resume_from_checkpoint: false | |
| dataset: | |
| _component_: torchtune.datasets.alpaca_cleaned_dataset | |
| packed: false | |
| seed: null | |
| shuffle: true | |
| epochs: 1 | |
| max_steps_per_epoch: 60 | |
| batch_size: 16 | |
| gradient_accumulation_steps: 8 | |
| optimizer: | |
| _component_: torch.optim.AdamW | |
| fused: true | |
| lr: 2.0e-05 | |
| optimizer_in_bwd: false | |
| loss: | |
| _component_: torchtune.modules.loss.CEWithChunkedOutputLoss | |
| device: cuda | |
| enable_activation_checkpointing: false | |
| enable_activation_offloading: false | |
| dtype: bf16 | |
| compile: true | |
| output_dir: qwen2_5.qwen2_5_0_5b_instruct | |
| metric_logger: | |
| _component_: torchtune.training.metric_logging.WandBLogger | |
| project: hallucination | |
| entity: c-metrics | |
| log_every_n_steps: 1 | |
| log_peak_memory_stats: true | |
| profiler: | |
| _component_: torchtune.training.setup_torch_profiler | |
| enabled: false | |
| output_dir: ${output_dir}/profiling_outputs | |
| cpu: true | |
| cuda: true | |
| profile_memory: false | |
| with_stack: false | |
| record_shapes: true | |
| with_flops: false | |
| wait_steps: 5 | |
| warmup_steps: 3 | |
| active_steps: 2 | |
| num_cycles: 1 | |