loadbrief / training_config.json

v8 LoRA adapter (SFT only, no RL stage)

722e529 verified about 2 months ago

271 Bytes

{
  "model_name": "meta-llama/Meta-Llama-3-8B-Instruct",
  "lora_rank": 32,
  "effective_batch": 32,
  "learning_rate": 0.0002,
  "epochs": 1,
  "max_length": 512,
  "train_examples": 20000,
  "training_minutes": 1719.6,
  "trl_version": "1.5.1",
  "device": "mps_48gb"
}

	{
	"model_name": "meta-llama/Meta-Llama-3-8B-Instruct",
	"lora_rank": 32,
	"effective_batch": 32,
	"learning_rate": 0.0002,
	"epochs": 1,
	"max_length": 512,
	"train_examples": 20000,
	"training_minutes": 1719.6,
	"trl_version": "1.5.1",
	"device": "mps_48gb"
	}