See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: zake7749/gemma-2-2b-it-chinese-kyara-dpo
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 8c2dd3c11c63229d_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/8c2dd3c11c63229d_train_data.json
  type:
    field_instruction: premise
    field_output: hypothesis
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
device_map:
  ? ''
  : 0,1,2,3,4,5,6,7
early_stopping_patience: 2
eval_max_new_tokens: 128
eval_steps: 100
eval_table_size: null
flash_attention: false
gradient_accumulation_steps: 8
gradient_checkpointing: true
group_by_length: false
hub_model_id: Alphatao/5b2c40df-bedb-47c3-b4a1-28953e907c67
hub_repo: null
hub_strategy: null
hub_token: null
learning_rate: 0.0002
load_best_model_at_end: true
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: null
lora_model_dir: null
lora_r: 16
lora_target_linear: true
lora_target_modules:
- q_proj
- k_proj
- v_proj
- o_proj
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: 2427
micro_batch_size: 4
mlflow_experiment_name: /tmp/8c2dd3c11c63229d_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: 0.029679755438815184
wandb_entity: null
wandb_mode: online
wandb_name: 60492357-69eb-4e2c-a118-9f38faabd1d2
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 60492357-69eb-4e2c-a118-9f38faabd1d2
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

5b2c40df-bedb-47c3-b4a1-28953e907c67

This model is a fine-tuned version of zake7749/gemma-2-2b-it-chinese-kyara-dpo on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.6759

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: 8
• total_train_batch_size: 32
• 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
• training_steps: 2427

Training results

Training Loss	Epoch	Step	Validation Loss
4.2567	0.0002	1	5.2654
1.8257	0.0196	100	1.8478
2.1918	0.0392	200	1.8134
1.713	0.0587	300	1.7998
2.3756	0.0783	400	1.7872
2.1478	0.0979	500	1.7767
1.6895	0.1175	600	1.7681
1.6317	0.1370	700	1.7571
1.8421	0.1566	800	1.7521
1.5517	0.1762	900	1.7412
1.7238	0.1958	1000	1.7356
1.9622	0.2153	1100	1.7270
1.6885	0.2349	1200	1.7207
1.5344	0.2545	1300	1.7143
1.4848	0.2741	1400	1.7047
1.477	0.2936	1500	1.6972
1.6738	0.3132	1600	1.6928
1.6056	0.3328	1700	1.6890
1.7096	0.3524	1800	1.6852
2.2766	0.3719	1900	1.6821
1.5262	0.3915	2000	1.6786
1.6859	0.4111	2100	1.6771
1.5937	0.4307	2200	1.6764
1.4678	0.4502	2300	1.6757
1.5568	0.4698	2400	1.6759

Framework versions

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