See axolotl config

axolotl version: 0.5.2

adapter: lora
auto_find_batch_size: true
base_model: Qwen/Qwen2-0.5B-Instruct
bf16: auto
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 4ab73156d9f21816_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/4ab73156d9f21816_train_data.json
  type:
    field_instruction: sentence1
    field_output: sentence2
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: /workspace/axolotl/configs/deepspeed_stage2.json
early_stopping: true
early_stopping_patience: 3
eval_max_new_tokens: 128
eval_sample_packing: false
eval_steps: 10
eval_strategy: steps
eval_table_size: null
flash_attention: true
fp16: false
gpu_memory_limit: 80GiB
gradient_accumulation_steps: 4
gradient_checkpointing: true
greater_is_better: false
group_by_length: true
hub_model_id: PhoenixB/5a009059-1c4f-4d37-9278-3826b60a39e4
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 1e-4
liger_fused_linear_cross_entropy: true
liger_glu_activation: true
liger_layer_norm: true
liger_rms_norm: true
liger_rope: true
load_best_model_at_end: true
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 5
lora_alpha: 16
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 16
lora_target_linear: true
lr_scheduler: cosine
max_steps: 150
metric_for_best_model: loss
micro_batch_size: 2
mlflow_experiment_name: /tmp/4ab73156d9f21816_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 3
optimizer: adamw_torch_fused
output_dir: miner_id_24
pad_to_sequence_len: true
plugins:
- axolotl.integrations.liger.LigerPlugin
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
save_steps: 20
sequence_len: 8196
strict: false
tf32: true
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: 2d9ccce4-8136-4915-8917-88a69776a1fe
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 2d9ccce4-8136-4915-8917-88a69776a1fe
warmup_steps: 20
weight_decay: 0.0

5a009059-1c4f-4d37-9278-3826b60a39e4

This model is a fine-tuned version of Qwen/Qwen2-0.5B-Instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.7922

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: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• total_eval_batch_size: 4
• 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: 20
• training_steps: 150

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0002	1	4.2868
3.668	0.0019	10	4.0048
2.648	0.0038	20	2.2974
2.324	0.0057	30	1.9736
1.8779	0.0077	40	1.9136
1.6871	0.0096	50	1.8672
1.8798	0.0115	60	1.8412
1.8296	0.0134	70	1.8306
2.1119	0.0153	80	1.8141
1.7267	0.0172	90	1.8057
1.6512	0.0192	100	1.8047
1.6577	0.0211	110	1.7996
1.7361	0.0230	120	1.7982
2.2023	0.0249	130	1.7961
1.8032	0.0268	140	1.7953
1.5624	0.0287	150	1.7922

Framework versions

• PEFT 0.13.2
• Transformers 4.46.3
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.20.3