| { |
| "run_name": "llama32-3B-rte-bf16-lora-seed44", |
| "model_size": "3B", |
| "model_id": "meta-llama/Llama-3.2-3B", |
| "task": "rte", |
| "glue_config": "rte", |
| "n_classes": 2, |
| "bitwidth": "bf16", |
| "seed": 44, |
| "lora_init_source": "shared:lora_init_3B_seed44.pt:224tensors", |
| "trainable_params": 9175040, |
| "total_params": 3221924864, |
| "train": { |
| "steps": 105, |
| "epochs": 2.9411764705882355, |
| "train_runtime_sec": 225.0290515422821, |
| "train_loss": 0.4039383820125035, |
| "n_train_examples": 2241, |
| "n_train_tokens_per_epoch": 219447, |
| "throughput_samples_per_sec": 29.290335738493653, |
| "throughput_tokens_per_sec": 2868.217896834099 |
| }, |
| "peak_gpu_mem_gib": { |
| "allocated": 7.778663635253906, |
| "reserved": 14.548828125 |
| }, |
| "validation": { |
| "accuracy": 0.8594377510040161, |
| "macro_f1": 0.8594014680971204, |
| "loss": 0.3213284973159851, |
| "n": 249, |
| "n_classes": 2, |
| "majority_baseline": 0.5140562248995983, |
| "pred_dist": { |
| "A": 132, |
| "B": 117 |
| }, |
| "true_dist": { |
| "A": 121, |
| "B": 128 |
| } |
| }, |
| "test": { |
| "accuracy": 0.8447653429602888, |
| "macro_f1": 0.8442456814823533, |
| "loss": 0.4106067032805419, |
| "n": 277, |
| "n_classes": 2, |
| "majority_baseline": 0.5270758122743683, |
| "pred_dist": { |
| "A": 147, |
| "B": 130 |
| }, |
| "true_dist": { |
| "A": 146, |
| "B": 131 |
| } |
| }, |
| "hyperparams": { |
| "num_train_epochs": 3, |
| "learning_rate": 0.0002, |
| "max_length": 256, |
| "per_device_train_batch_size": 4, |
| "per_device_eval_batch_size": 8, |
| "gradient_accumulation_steps": 16, |
| "warmup_ratio": 0.03, |
| "lr_scheduler_type": "cosine", |
| "gradient_checkpointing": true |
| }, |
| "lora": { |
| "r": 16, |
| "alpha": 32, |
| "dropout": 0.0, |
| "bias": "none", |
| "target_modules": [ |
| "q_proj", |
| "k_proj", |
| "v_proj", |
| "o_proj" |
| ] |
| }, |
| "env": { |
| "torch": "2.4.1+cu121", |
| "cuda": "12.1", |
| "gpu": "NVIDIA GeForce RTX 4090", |
| "slurm_job": "1932474", |
| "node": "node39" |
| }, |
| "debug_caps": { |
| "max_train": null, |
| "max_eval": null, |
| "max_steps": -1 |
| } |
| } |
