See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: elyza/Llama-3-ELYZA-JP-8B
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 8a34c82f4fcec0a6_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/8a34c82f4fcec0a6_train_data.json
  type:
    field_instruction: question
    field_output: answer
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: null
eval_max_new_tokens: 256
eval_table_size: null
evals_per_epoch: 4
flash_attention: false
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 32
gradient_checkpointing: true
group_by_length: false
hub_model_id: sn56/d52f364b-0870-4bf0-90f7-7d4940e808fe
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0002
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 3
lora_alpha: 64
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 32
lora_target_linear: true
lora_target_modules:
- q_proj
- k_proj
- v_proj
- o_proj
- gate_proj
- down_proj
- up_proj
lr_scheduler: cosine
max_grad_norm: 2
max_steps: 90
micro_batch_size: 2
mlflow_experiment_name: /tmp/8a34c82f4fcec0a6_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 3
optim_args:
  adam_beta1: 0.9
  adam_beta2: 0.95
  adam_epsilon: 1.0e-05
optimizer: adamw_torch
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
saves_per_epoch: 4
seed: 2122986924
sequence_len: 2048
shuffle: true
special_tokens:
  pad_token: <|eot_id|>
strict: false
tf32: false
tokenizer_type: AutoTokenizer
torch_compile: true
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: sn56-miner
wandb_mode: disabled
wandb_name: null
wandb_project: god
wandb_run: ivjl
wandb_runid: null
warmup_steps: 20
weight_decay: 0.02
xformers_attention: false

d52f364b-0870-4bf0-90f7-7d4940e808fe

This model is a fine-tuned version of elyza/Llama-3-ELYZA-JP-8B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3648

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 2
• eval_batch_size: 2
• seed: 2122986924
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 32
• total_train_batch_size: 256
• total_eval_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=adam_beta1=0.9,adam_beta2=0.95,adam_epsilon=1e-05
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 20
• training_steps: 90

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0054	1	0.6262
0.5582	0.0433	8	0.5080
0.4572	0.0866	16	0.4493
0.4166	0.1299	24	0.4204
0.3891	0.1731	32	0.4024
0.3761	0.2164	40	0.3908
0.3759	0.2597	48	0.3820
0.3721	0.3030	56	0.3755
0.3617	0.3463	64	0.3706
0.3583	0.3896	72	0.3670
0.3537	0.4329	80	0.3653
0.3521	0.4762	88	0.3648

Framework versions

• PEFT 0.13.2
• Transformers 4.46.0
• Pytorch 2.5.0+cu124
• Datasets 3.0.1
• Tokenizers 0.20.1