See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: beomi/polyglot-ko-12.8b-safetensors
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 7da945e3f9b506b1_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/7da945e3f9b506b1_train_data.json
  type:
    field_instruction: premise
    field_output: hypothesis
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 30
eval_max_new_tokens: 128
eval_steps: 50
eval_table_size: null
flash_attention: false
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 16
gradient_checkpointing: true
group_by_length: false
hub_model_id: Romain-XV/f8e4cfbf-cc68-45fe-9b4b-70f932bb5ef9
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: 1
lora_alpha: 32
lora_dropout: 0.05
lora_fan_in_fan_out: true
lora_model_dir: null
lora_r: 16
lora_target_linear: true
lora_target_modules:
- q_proj
- k_proj
- v_proj
lr_scheduler: cosine
micro_batch_size: 4
mlflow_experiment_name: /tmp/7da945e3f9b506b1_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 2
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: 100
sequence_len: 1024
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: d4310d46-c3d3-43c3-bc02-992c85afbc77
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: d4310d46-c3d3-43c3-bc02-992c85afbc77
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

f8e4cfbf-cc68-45fe-9b4b-70f932bb5ef9

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: 0.4591

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 64
• 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: 10
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss
24.7371	0.0012	1	1.5328
9.709	0.0607	50	0.5939
8.5961	0.1215	100	0.5613
9.4501	0.1822	150	0.5476
8.9883	0.2429	200	0.5326
9.0582	0.3037	250	0.5257
8.4664	0.3644	300	0.5199
8.4412	0.4251	350	0.5100
7.3829	0.4859	400	0.5047
8.3055	0.5466	450	0.4967
8.1007	0.6073	500	0.4941
7.4946	0.6681	550	0.4876
8.7649	0.7288	600	0.4849
8.1825	0.7896	650	0.4809
7.1683	0.8503	700	0.4767
7.505	0.9110	750	0.4727
8.1852	0.9718	800	0.4703
6.1613	1.0325	850	0.4719
5.7592	1.0932	900	0.4750
5.8067	1.1540	950	0.4704
6.2207	1.2147	1000	0.4701
5.3496	1.2754	1050	0.4685
5.8719	1.3362	1100	0.4654
6.3027	1.3969	1150	0.4635
5.8079	1.4576	1200	0.4642
5.3594	1.5184	1250	0.4612
5.7123	1.5791	1300	0.4619
5.4205	1.6398	1350	0.4621
5.2061	1.7006	1400	0.4612
5.3493	1.7613	1450	0.4601
5.2286	1.8220	1500	0.4591
6.0298	1.8828	1550	0.4591
5.2035	1.9435	1600	0.4591

Framework versions

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