added training_info.json
Browse files- training_info.json +86 -0
training_info.json
ADDED
|
@@ -0,0 +1,86 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"training_metadata": {
|
| 3 |
+
"timestamp": "20251021_084709",
|
| 4 |
+
"training_date": "2025-10-21",
|
| 5 |
+
"training_time": "15:45:44",
|
| 6 |
+
"final_epoch": 5.0,
|
| 7 |
+
"total_steps": 2355,
|
| 8 |
+
"status": "completed"
|
| 9 |
+
},
|
| 10 |
+
"model_config": {
|
| 11 |
+
"base_model": "Kwaipilot/KAT-Dev",
|
| 12 |
+
"model_type": "causal_lm",
|
| 13 |
+
"architecture": "Qwen2ForCausalLM"
|
| 14 |
+
},
|
| 15 |
+
"lora_config": {
|
| 16 |
+
"r": 64,
|
| 17 |
+
"lora_alpha": 128,
|
| 18 |
+
"lora_dropout": 0.05,
|
| 19 |
+
"target_modules": [
|
| 20 |
+
"q_proj",
|
| 21 |
+
"k_proj",
|
| 22 |
+
"v_proj",
|
| 23 |
+
"o_proj",
|
| 24 |
+
"gate_proj",
|
| 25 |
+
"up_proj",
|
| 26 |
+
"down_proj"
|
| 27 |
+
]
|
| 28 |
+
},
|
| 29 |
+
"training_config": {
|
| 30 |
+
"num_epochs": 5,
|
| 31 |
+
"per_device_train_batch_size": 2,
|
| 32 |
+
"per_device_eval_batch_size": 2,
|
| 33 |
+
"gradient_accumulation_steps": 8,
|
| 34 |
+
"effective_batch_size": 32,
|
| 35 |
+
"learning_rate": 5e-05,
|
| 36 |
+
"lr_scheduler_type": "cosine",
|
| 37 |
+
"warmup_ratio": 0.03,
|
| 38 |
+
"weight_decay": 0.1,
|
| 39 |
+
"max_grad_norm": 0.5,
|
| 40 |
+
"bf16": true,
|
| 41 |
+
"gradient_checkpointing": true,
|
| 42 |
+
"optim": "adamw_torch",
|
| 43 |
+
"logging_steps": 10,
|
| 44 |
+
"save_steps": 50,
|
| 45 |
+
"eval_steps": 25
|
| 46 |
+
},
|
| 47 |
+
"dataset_info": {
|
| 48 |
+
"train_samples": 15057,
|
| 49 |
+
"eval_samples": 1674,
|
| 50 |
+
"max_seq_length": 8192,
|
| 51 |
+
"sample_packing": false
|
| 52 |
+
},
|
| 53 |
+
"hardware_config": {
|
| 54 |
+
"num_gpus": 2,
|
| 55 |
+
"gpu_model": "NVIDIA H200",
|
| 56 |
+
"distributed_strategy": "DeepSpeed ZeRO-2",
|
| 57 |
+
"flash_attention": "2.8.3"
|
| 58 |
+
},
|
| 59 |
+
"performance_metrics": {
|
| 60 |
+
"final_train_loss": 0.4799,
|
| 61 |
+
"final_eval_loss": 0.46369343996047974,
|
| 62 |
+
"final_train_perplexity": 1.6159128028327767,
|
| 63 |
+
"final_eval_perplexity": 1.5899354850369571,
|
| 64 |
+
"final_token_accuracy": 0.8899670094251633,
|
| 65 |
+
"initial_loss": 1.6335,
|
| 66 |
+
"initial_perplexity": 5.121769577787628,
|
| 67 |
+
"initial_accuracy": 0.6116928663104773
|
| 68 |
+
},
|
| 69 |
+
"framework_versions": {
|
| 70 |
+
"torch": "2.4.1+cu124",
|
| 71 |
+
"transformers": "4.57.1",
|
| 72 |
+
"peft": "0.17.1",
|
| 73 |
+
"trl": "0.23.1",
|
| 74 |
+
"deepspeed": "0.18.0",
|
| 75 |
+
"flash_attn": "2.8.3",
|
| 76 |
+
"python": "3.12.3"
|
| 77 |
+
},
|
| 78 |
+
"special_features": {
|
| 79 |
+
"flash_attention_2": true,
|
| 80 |
+
"gradient_checkpointing": true,
|
| 81 |
+
"bf16_training": true,
|
| 82 |
+
"sample_packing": false,
|
| 83 |
+
"deepspeed_zero2": true,
|
| 84 |
+
"distributed_training": true
|
| 85 |
+
}
|
| 86 |
+
}
|