See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: heegyu/WizardVicuna-open-llama-3b-v2
bf16: auto
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 0ba056ab4aea2163_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/0ba056ab4aea2163_train_data.json
  type:
    field_input: input_role
    field_instruction: target_role
    field_output: target_role_short
    format: '{instruction} {input}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
device: cuda
early_stopping_patience: null
eval_max_new_tokens: 128
eval_table_size: null
evals_per_epoch: 4
flash_attention: false
fp16: null
fsdp: null
fsdp_config:
  max_steps: 50
  weight_decay: 0.01
gradient_accumulation_steps: 4
gradient_checkpointing: false
group_by_length: false
hub_model_id: sn56a1/681012ad-17b8-4365-810a-9cf00e644424
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: 16
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 8
lora_target_linear: true
lr_scheduler: cosine
max_memory:
  0: 70GiB
max_steps: 50
micro_batch_size: 2
mlflow_experiment_name: /tmp/0ba056ab4aea2163_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 3
optimizer: adamw_torch
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
save_steps: 70
sequence_len: 1024
special_tokens:
  pad_token: </s>
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: sn56-miner
wandb_mode: disabled
wandb_name: 681012ad-17b8-4365-810a-9cf00e644424
wandb_project: god
wandb_run: k3li
wandb_runid: 681012ad-17b8-4365-810a-9cf00e644424
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

681012ad-17b8-4365-810a-9cf00e644424

This model is a fine-tuned version of heegyu/WizardVicuna-open-llama-3b-v2 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0001

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: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• total_eval_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_TORCH 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: 50

Training results

Training Loss	Epoch	Step	Validation Loss
5.8478	0.0013	1	5.8559
5.4022	0.0065	5	5.2092
2.2725	0.0130	10	1.3080
0.2262	0.0195	15	0.1639
0.0172	0.0260	20	0.0119
0.0034	0.0325	25	0.0019
0.0005	0.0390	30	0.0005
0.0002	0.0455	35	0.0003
0.0002	0.0520	40	0.0002
0.0001	0.0586	45	0.0002
0.0001	0.0651	50	0.0001

Framework versions

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