mt5-autonomous-workspace / training_state.json
abdoosh1000's picture
Update training state - Sample 6000, Chunk 5, Fine-tunings: 6
46b573f verified
{
"session_id": "session_20250901_105031",
"last_processed_sample": 6000,
"total_samples_processed": 6001,
"current_chunk": 5,
"latest_model_name": "mt5-websrc-chunk5-20250901_111931",
"latest_model_version": 1,
"training_start_time": "2025-09-01T10:50:31.022781",
"last_update_time": "2025-09-01T11:23:05.186083",
"completed_datasets": [],
"current_dataset": "websrc",
"training_metrics": {
"last_loss": 0.0,
"best_loss": 0.0
},
"system_info": {
"accelerator": "GPU",
"config": {
"batch_size": 8,
"gradient_accumulation_steps": 2,
"learning_rate": 0.0001,
"weight_decay": 0.01,
"warmup_ratio": 0.1,
"max_grad_norm": 1.0,
"fp16": true,
"bf16": false,
"dataloader_num_workers": 4,
"remove_unused_columns": true,
"use_compile": false,
"save_safetensors": true,
"use_lora": false,
"multi_gpu": true,
"gpu_count": 2,
"identical_gpus": true,
"ddp_find_unused_parameters": false,
"ddp_bucket_cap_mb": 25,
"dataloader_pin_memory": true
},
"device_agnostic_tracking": true,
"workspace_version": "2.0",
"created_with_persistence": true,
"current_accelerator": "GPU",
"current_config": {
"batch_size": 8,
"gradient_accumulation_steps": 2,
"learning_rate": 0.0001,
"weight_decay": 0.01,
"warmup_ratio": 0.1,
"max_grad_norm": 1.0,
"fp16": true,
"bf16": false,
"dataloader_num_workers": 4,
"remove_unused_columns": true,
"use_compile": false,
"save_safetensors": true,
"use_lora": false,
"multi_gpu": true,
"gpu_count": 2,
"identical_gpus": true,
"ddp_find_unused_parameters": false,
"ddp_bucket_cap_mb": 25,
"dataloader_pin_memory": true
},
"last_device_update": "2025-09-01T11:23:05.186090",
"last_saved_to_hub": "2025-09-01T11:23:05.186092"
},
"dataset_progress": {
"websrc": {
"last_processed_sample": 6000,
"total_samples_processed": 16000,
"current_chunk": 5,
"completion_percentage": 37.5,
"last_update_time": "2025-09-01T11:19:21.526886",
"last_sample": 6000
}
},
"model_history": [
{
"model_name": "mt5-websrc-chunk2-20250901_110428",
"chunk_number": 2,
"dataset_name": "websrc",
"timestamp": "2025-09-01T11:04:45.351161",
"upload_time": "2025-09-01T11:04:45.351163",
"samples_trained": 3001,
"metrics": {
"upload_time": "2025-09-01T11:04:45.351147",
"model_path": "models/mt5-websrc-chunk2-20250901_110428",
"local_path": "./model_chunk_2",
"milestone_finetunings": 3
},
"repo_path": "models/mt5-websrc-chunk2-20250901_110428",
"training_session_id": "session_20250901_105031"
},
{
"model_name": "mt5-websrc-chunk2-20250901_110428",
"chunk_number": 2,
"dataset_name": "websrc",
"timestamp": "2025-09-01T11:04:56.725441",
"upload_time": "2025-09-01T11:04:56.725444",
"samples_trained": 3001,
"metrics": {
"milestone_finetunings": 3,
"milestone_type": "regular_3",
"upload_time": "2025-09-01T11:04:56.725414",
"chunk_samples": 1000,
"total_samples_trained": 3001,
"milestone_achievement": true,
"chunk_number": 2,
"dataset_name": "websrc"
},
"repo_path": "models/mt5-websrc-chunk2-20250901_110428",
"training_session_id": "session_20250901_105031"
},
{
"model_name": "mt5-websrc-chunk3-20250901_110932",
"chunk_number": 3,
"dataset_name": "websrc",
"timestamp": "2025-09-01T11:09:56.195952",
"upload_time": "2025-09-01T11:09:56.195954",
"samples_trained": 3001,
"metrics": {
"upload_time": "2025-09-01T11:09:56.195937",
"model_path": "models/mt5-websrc-chunk3-20250901_110932",
"local_path": "./model_chunk_3",
"milestone_finetunings": 3
},
"repo_path": "models/mt5-websrc-chunk3-20250901_110932",
"training_session_id": "session_20250901_105031"
},
{
"model_name": "mt5-websrc-chunk5-20250901_111931",
"chunk_number": 5,
"dataset_name": "websrc",
"timestamp": "2025-09-01T11:19:45.814548",
"upload_time": "2025-09-01T11:19:45.814553",
"samples_trained": 6001,
"metrics": {
"upload_time": "2025-09-01T11:19:45.814529",
"model_path": "models/mt5-websrc-chunk5-20250901_111931",
"local_path": "./model_chunk_5",
"milestone_finetunings": 6
},
"repo_path": "models/mt5-websrc-chunk5-20250901_111931",
"training_session_id": "session_20250901_105031"
},
{
"model_name": "mt5-websrc-chunk5-20250901_111931",
"chunk_number": 5,
"dataset_name": "websrc",
"timestamp": "2025-09-01T11:19:56.574490",
"upload_time": "2025-09-01T11:19:56.574492",
"samples_trained": 6001,
"metrics": {
"milestone_finetunings": 6,
"milestone_type": "regular_3",
"upload_time": "2025-09-01T11:19:56.574446",
"chunk_samples": 1000,
"total_samples_trained": 6001,
"milestone_achievement": true,
"chunk_number": 5,
"dataset_name": "websrc"
},
"repo_path": "models/mt5-websrc-chunk5-20250901_111931",
"training_session_id": "session_20250901_105031"
}
],
"training_sessions": [
{
"session_id": "session_20250901_105031",
"start_time": "2025-09-01T10:50:31.022802",
"dataset": "websrc",
"status": "active",
"accelerator": "GPU"
}
],
"workspace_version": "2.0",
"last_model_path": "models/mt5-websrc-chunk5-20250901_111931",
"next_action": "continue",
"successful_finetunings_count": 6
}