LuxAI-1-Nemo-Checkpoints / training_config.json
luxopes's picture
LuxAI Nemo resume checkpoint, group 1
0ee8bdf verified
Raw History Blame Contribute Delete
1.24 kB
{
"model_id": "mistralai/Ministral-3-14B-Base-2512",
"model_revision": "5b0ceedbb42dff466ae60b258ba296f32da51384",
"prepared_dir": "training/prepared",
"run_dir": "training/runs/nemo-unsloth-bf16-64k",
"backend": "unsloth",
"freeze_vision": true,
"auto_stop_pod": true,
"budget_usd": 150.0,
"finalization_reserve_usd": 10.0,
"other_cost_reserve_usd": 5.0,
"storage_hourly_allowance_usd": 0.03,
"max_gpu_hourly_usd": 2.2,
"budget_action": "save_and_stop",
"required_gpu": "RTX PRO 6000",
"pod_name_prefix": "luxai-nemo",
"watchdog_poll_seconds": 15,
"stall_timeout_minutes": 60,
"max_seq_length": 65536,
"oom_fallback_seq_length": 51200,
"seed": 3407,
"lora_rank": 64,
"lora_alpha": 128,
"lora_dropout": 0.0,
"learning_rate": 0.0001,
"min_lr_ratio": 0.1,
"warmup_ratio": 0.03,
"weight_decay": 0.01,
"max_grad_norm": 1.0,
"loss_chunk_size": 256,
"checkpoint_minutes": 15,
"checkpoint_loss_tokens": 50000000,
"hf_checkpoint_repo": "luxopes/LuxAI-1-Nemo-Checkpoints",
"hf_checkpoint_hours": 6,
"hf_checkpoint_retry_minutes": 15,
"hf_model_repo": "luxopes/LuxAI-1-Nemo",
"hf_adapter_repo": "luxopes/LuxAI-1-Nemo-LoRA",
"hf_private": true,
"upload_after_training": true
}