MiniBananaMind-v4-9M / train_args.json
Banaxi-Tech's picture
Upload MiniBananaMind-v4-9M
013c7e8 verified
Raw History Blame Contribute Delete
579 Bytes
{
"data": "cosmopedia_v2_3B_8k_digits.uint16.bin",
"base": "MiniBananaMind-v3-9M",
"revision": "",
"tokenizer": "",
"out": "checkpoints/MiniBananaMind-v3-9M-CosmopediaV2-3B-1024",
"seq_len": 1024,
"batch_size": 72,
"grad_accum": 16,
"max_steps": 2613,
"lr": 0.0001,
"min_lr": 1e-05,
"warmup_steps": 200,
"weight_decay": 0.1,
"grad_clip": 1.0,
"val_tokens": 5000000,
"eval_every": 250,
"eval_batches": 50,
"save_every": 250,
"device": "cuda",
"dtype": "auto",
"compile": true,
"compile_mode": "default",
"seed": 1337,
"resume": ""
}