CheesefireV3lora / training_parameters.json
foilfoilfoil's picture
Upload 6 files
ea000b9
raw
history blame contribute delete
726 Bytes
{
"lora_name": "discordchat7b",
"always_override": true,
"save_steps": 1000.0,
"micro_batch_size": 2,
"batch_size": 128,
"epochs": 2.0,
"learning_rate": "1e-5",
"lr_scheduler_type": "cosine",
"lora_rank": 128,
"lora_alpha": 256,
"lora_dropout": 0.05,
"cutoff_len": 1024,
"dataset": "mini-10k",
"eval_dataset": "None",
"format": "vicuna-format",
"eval_steps": 500.0,
"raw_text_file": "formatted_messages",
"overlap_len": 64,
"newline_favor_len": 64,
"higher_rank_limit": false,
"warmup_steps": 100.0,
"optimizer": "adamw_bnb_8bit",
"hard_cut_string": "\\n\\n\\n",
"train_only_after": "",
"stop_at_loss": 0,
"add_eos_token": false,
"min_chars": 0.0,
"report_to": "wandb"
}