See axolotl config

axolotl version: 0.10.0.dev0

adapter: lora
base_model: Qwen/Qwen3-4B
bf16: true
datasets:
- data_files:
  - e07d306d62489726_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/
  type:
    field_input: input
    field_instruction: instruct
    field_output: output
    format: '{instruction} {input}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
eval_max_new_tokens: 128
evals_per_epoch: 4
flash_attention: false
fp16: false
gradient_accumulation_steps: 1
gradient_checkpointing: true
group_by_length: true
hf_upload_public: true
hf_upload_repo_type: model
hub_model_id: apriasmoro/2d7701d2-57a5-4412-9b17-a3dee3aef4ed
learning_rate: 0.0001
load_in_4bit: false
logging_steps: 10
lora_alpha: 32
lora_dropout: 0.15
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 32
lora_target_linear: true
loraplus_lr_embedding: 1.0e-06
loraplus_lr_ratio: 16
lr_scheduler: cosine
max_steps: 1620
micro_batch_size: 28
mlflow_experiment_name: /tmp/e07d306d62489726_train_data.json
model_card: false
optimizer: adamw_torch_fused
output_dir: miner_id_24
push_to_hub: true
rl: null
sample_packing: true
save_steps: 243
sequence_len: 2048
tf32: true
tokenizer_type: AutoTokenizer
train_on_inputs: false
trl: null
trust_remote_code: true
wandb_name: b2262a8e-1631-49db-894d-8c1276b94651
wandb_project: Gradients-On-Demand
wandb_run: apriasmoro
wandb_runid: b2262a8e-1631-49db-894d-8c1276b94651
warmup_steps: 162
weight_decay: 0.02

2d7701d2-57a5-4412-9b17-a3dee3aef4ed

This model is a fine-tuned version of Qwen/Qwen3-4B on an unknown dataset.

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: 0.0001
• train_batch_size: 28
• eval_batch_size: 28
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• total_train_batch_size: 56
• total_eval_batch_size: 56
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 162
• training_steps: 1620

Training results

Framework versions

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