adapter: lora
base_model: WhiteRabbitNeo/Llama-3-WhiteRabbitNeo-8B-v2.0
bf16: false
bnb_4bit_compute_dtype: float16
bnb_4bit_quant_type: nf4
bnb_4bit_use_double_quant: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 9b33138dcfdbbaea_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/
  type:
    field_input: None
    field_instruction: instruct
    field_output: output
    field_system: None
    format: None
    no_input_format: None
    system_format: '{system}'
    system_prompt: None
debug: null
deepspeed: null
early_stopping_patience: null
eval_max_new_tokens: 128
eval_table_size: null
evals_per_epoch: 4
flash_attention: false
fp16: true
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 1
gradient_checkpointing: true
group_by_length: true
hub_model_id: apriasmoro/83d2ef11-a360-4972-8485-b83403aeb513
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0002
load_in_4bit: true
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_steps: 5
micro_batch_size: 2
mlflow_experiment_name: /tmp/9b33138dcfdbbaea_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
sequence_len: 512
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: 6eee0287-8ac7-4224-9016-db360c2e7534
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 6eee0287-8ac7-4224-9016-db360c2e7534
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

83d2ef11-a360-4972-8485-b83403aeb513

This model is a fine-tuned version of WhiteRabbitNeo/Llama-3-WhiteRabbitNeo-8B-v2.0 on an unknown dataset. It achieves the following results on the evaluation set:

Loss: 1.3458

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: 2
total_train_batch_size: 4
total_eval_batch_size: 4
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: 5
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
1.5913	0.2	1	1.4644
1.2216	0.4	2	1.4553
1.1742	0.8	4	1.3458

Framework versions

PEFT 0.15.2
Transformers 4.51.3
Pytorch 2.5.1+cu124
Datasets 3.5.1
Tokenizers 0.21.1