See axolotl config

axolotl version: 0.4.1

adapter: qlora
auto_resume_from_checkpoints: true
base_model: Qwen/Qwen2.5-3B
bf16: auto
chat_template: llama3
dataloader_num_workers: 6
dataset_prepared_path: null
datasets:
- data_files:
  - 24b4fa4ad9cceed3_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/24b4fa4ad9cceed3_train_data.json
  type:
    field_instruction: prompt
    field_output: chosen
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 3
eval_max_new_tokens: 128
eval_steps: 50
eval_table_size: null
evals_per_epoch: null
flash_attention: false
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 16
gradient_checkpointing: true
group_by_length: true
hub_model_id: error577/83740774-80d7-4f6e-876d-da910c5abc61
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: 16
lora_dropout: 0.3
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 8
lora_target_linear: true
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: 1000
micro_batch_size: 1
mlflow_experiment_name: /tmp/24b4fa4ad9cceed3_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: 50
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: 6996457f-13d0-46b6-8cd4-8047ff829794
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 6996457f-13d0-46b6-8cd4-8047ff829794
warmup_steps: 10
weight_decay: 0.01
xformers_attention: null

83740774-80d7-4f6e-876d-da910c5abc61

This model is a fine-tuned version of Qwen/Qwen2.5-3B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.2125

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: 1000

Training results

Training Loss	Epoch	Step	Validation Loss
0.9046	0.0061	1	1.6886
1.3796	0.3059	50	1.2486
1.0926	0.6119	100	1.1687
0.9194	0.9178	150	1.1075
0.7725	1.2237	200	1.1503
0.9649	1.5296	250	1.1441
1.0847	1.8356	300	1.0945
0.7307	2.1415	350	1.1109
0.6989	2.4474	400	1.1534
1.2336	2.7533	450	1.2125

Framework versions

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