See axolotl config

axolotl version: 0.4.1

adapter: qlora
base_model: unsloth/Llama-3.2-3B
bf16: auto
chat_template: llama3
dataloader_num_workers: 6
dataset_prepared_path: null
datasets:
- data_files:
  - c9b7d4a6209ec024_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/c9b7d4a6209ec024_train_data.json
  type:
    field_instruction: context
    field_output: question
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping:
  metric: eval_loss
  mode: min
  patience: 3
eval_max_new_tokens: 128
eval_steps: 20
eval_table_size: null
evals_per_epoch: null
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 16
gradient_checkpointing: true
group_by_length: true
hub_model_id: error577/856ed31c-62d9-4ed5-9989-8f6e5d4c01e3
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0003
load_in_4bit: true
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 64
lora_dropout: 0.3
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 32
lora_target_linear: true
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: 300
micro_batch_size: 1
mlflow_experiment_name: /tmp/c9b7d4a6209ec024_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 50
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: 20
sequence_len: 512
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.02
wandb_entity: null
wandb_mode: online
wandb_name: 7f5dd85d-b076-4c99-a5b9-664ad388f4ef
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 7f5dd85d-b076-4c99-a5b9-664ad388f4ef
warmup_steps: 10
weight_decay: 0.01
xformers_attention: null

856ed31c-62d9-4ed5-9989-8f6e5d4c01e3

This model is a fine-tuned version of unsloth/Llama-3.2-3B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.7131

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: 0.0003
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 16
• 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: 300

Training results

Training Loss	Epoch	Step	Validation Loss
4.0814	0.0009	1	4.7141
1.5844	0.0172	20	1.8652
1.6637	0.0344	40	1.8204
1.5636	0.0515	60	1.8356
1.4937	0.0687	80	1.8590
1.6937	0.0859	100	1.8064
1.5605	0.1031	120	1.7988
1.4112	0.1202	140	1.7772
1.6521	0.1374	160	1.7552
1.5504	0.1546	180	1.7561
1.8864	0.1718	200	1.7429
1.6297	0.1889	220	1.7317
1.5704	0.2061	240	1.7266
1.6367	0.2233	260	1.7198
1.4851	0.2405	280	1.7132
1.6715	0.2576	300	1.7131

Framework versions

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