See axolotl config

axolotl version: 0.4.1

adapter: lora
auto_find_batch_size: true
base_model: beomi/polyglot-ko-12.8b-safetensors
bf16: auto
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 1e5c36a8359001d2_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/1e5c36a8359001d2_train_data.json
  type:
    field_instruction: text
    field_output: ru_text
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
do_eval: true
early_stopping_patience: 3
eval_max_new_tokens: 128
eval_steps: 50
evals_per_epoch: null
flash_attention: true
fp16: false
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 2
gradient_checkpointing: false
group_by_length: true
hub_model_id: lesso14/b7543fbe-88cd-4e2a-9b3d-51a0fd6d2dc3
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.000214
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 10
lora_alpha: 32
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 16
lora_target_linear: true
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: 500
micro_batch_size: 4
mlflow_experiment_name: /tmp/1e5c36a8359001d2_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 1
optimizer: adamw_bnb_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
save_steps: 50
saves_per_epoch: null
seed: 140
sequence_len: 512
strict: false
tf32: true
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: b0536107-325b-4206-a1a9-f2d9eefc9a8e
wandb_project: 14a
wandb_run: your_name
wandb_runid: b0536107-325b-4206-a1a9-f2d9eefc9a8e
warmup_steps: 50
weight_decay: 0.0
xformers_attention: null

b7543fbe-88cd-4e2a-9b3d-51a0fd6d2dc3

This model is a fine-tuned version of beomi/polyglot-ko-12.8b-safetensors on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4444

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: 0.000214
• train_batch_size: 4
• eval_batch_size: 4
• seed: 140
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 50
• training_steps: 500

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0000	1	1.0745
1.6012	0.0020	50	0.7476
1.1744	0.0040	100	0.6696
1.2569	0.0060	150	0.6170
1.1339	0.0080	200	0.5507
1.058	0.0100	250	0.5240
0.9907	0.0120	300	0.4914
0.9367	0.0140	350	0.4695
0.8386	0.0160	400	0.4530
0.9271	0.0180	450	0.4458
0.894	0.0200	500	0.4444

Framework versions

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