| { | |
| "training_metadata": { | |
| "timestamp": "20251031_111636", | |
| "training_date": "2025-11-01", | |
| "training_time": "00:28:49", | |
| "final_epoch": 3.0, | |
| "total_steps": 711, | |
| "status": "completed" | |
| }, | |
| "model_config": { | |
| "base_model": "Kwaipilot/KAT-Dev-72B-Exp", | |
| "model_type": "causal_lm", | |
| "architecture": "Qwen2ForCausalLM" | |
| }, | |
| "lora_config": { | |
| "r": 64, | |
| "lora_alpha": 128, | |
| "lora_dropout": 0.05, | |
| "target_modules": [ | |
| "q_proj", | |
| "k_proj", | |
| "v_proj", | |
| "o_proj", | |
| "gate_proj", | |
| "up_proj", | |
| "down_proj" | |
| ] | |
| }, | |
| "training_config": { | |
| "num_epochs": 3, | |
| "per_device_train_batch_size": 1, | |
| "per_device_eval_batch_size": 1, | |
| "gradient_accumulation_steps": 16, | |
| "effective_batch_size": 64, | |
| "learning_rate": 5e-05, | |
| "lr_scheduler_type": "cosine", | |
| "warmup_ratio": 0.03, | |
| "weight_decay": 0.1, | |
| "max_grad_norm": 0.5, | |
| "bf16": true, | |
| "gradient_checkpointing": true, | |
| "optim": "adamw_torch", | |
| "logging_steps": 10, | |
| "save_steps": 50, | |
| "eval_steps": 25 | |
| }, | |
| "dataset_info": { | |
| "train_samples": 15129, | |
| "eval_samples": 1682, | |
| "max_seq_length": 8192, | |
| "sample_packing": false | |
| }, | |
| "hardware_config": { | |
| "num_gpus": 4, | |
| "gpu_model": "NVIDIA H200", | |
| "distributed_strategy": "DeepSpeed ZeRO-2", | |
| "flash_attention": "2.8.3" | |
| }, | |
| "performance_metrics": { | |
| "final_train_loss": 0.2641, | |
| "final_eval_loss": 0.37574875354766846, | |
| "final_train_perplexity": 1.3022584156313823, | |
| "final_eval_perplexity": 1.4560812525608204, | |
| "final_token_accuracy": 0.9259863365441561, | |
| "initial_loss": 1.6648, | |
| "initial_perplexity": 5.284616220817229, | |
| "initial_accuracy": 0.6015806214883923 | |
| }, | |
| "framework_versions": { | |
| "torch": "2.9.0+cu128", | |
| "transformers": "4.57.1", | |
| "peft": "0.17.1", | |
| "trl": "0.24.0", | |
| "deepspeed": "0.18.1", | |
| "flash_attn": "2.8.3", | |
| "python": "3.12.3" | |
| }, | |
| "special_features": { | |
| "flash_attention_2": true, | |
| "gradient_checkpointing": true, | |
| "bf16_training": true, | |
| "sample_packing": false, | |
| "deepspeed_zero2": true, | |
| "distributed_training": true | |
| } | |
| } |