See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: Qwen/Qwen2-7B-Instruct
bf16: true
chat_template: llama3
dataloader_num_workers: 8
dataloader_pin_memory: true
dataset_prepared_path: null
datasets:
- data_files:
  - e042e1b993a4ecfe_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/
  type:
    field_instruction: instruct
    field_output: output
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
device_map: auto
dynamic_lora_per_layer: true
early_stopping_patience: 2
eval_max_new_tokens: 128
eval_steps: 100
eval_table_size: null
evaluation_strategy: steps
flash_attention: true
fp16: false
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 2
gradient_checkpointing: true
group_by_length: false
hub_model_id: JoshMe1/fdf4cd1b-53b8-4f10-9e66-20dd67cab3ca
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0002
load_in_4bit: false
load_in_8bit: true
local_rank: null
logging_steps: 10
lora_alpha: 128
lora_dropout: 0.1
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 64
lora_target_linear: true
lr_finder: true
lr_scheduler: cosine
lr_scheduler_args: []
max_grad_norm: 1.0
max_memory:
  0: 130GB
max_steps: 1534
micro_batch_size: 8
mixed_precision: bf16
mlflow_experiment_name: /tmp/e042e1b993a4ecfe_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 4
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: 100
save_strategy: steps
save_total_limit: 3
scheduler:
  factor: 0.5
  monitor: eval_loss
  patience: 1
  threshold: 0.01
  type: ReduceLROnPlateau
sequence_len: 512
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
training_stages:
- learning_rate: 0.0002
  name: warmup
  num_train_epochs: 1
- learning_rate: 2.0e-05
  name: main
trl:
  ema: true
  ema_decay: 0.999
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: a8dd5d6f-03c0-4539-a4f4-1f162f583d8b
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: a8dd5d6f-03c0-4539-a4f4-1f162f583d8b
warmup_steps: 153
weight_decay: 0.01
xformers_attention: true

fdf4cd1b-53b8-4f10-9e66-20dd67cab3ca

This model is a fine-tuned version of Qwen/Qwen2-7B-Instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.3822

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• 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: 153
• training_steps: 1534

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0014	1	1.7261
1.4081	0.1372	100	1.4116
1.3922	0.2743	200	1.3941
1.3977	0.4115	300	1.3838
1.4132	0.5487	400	1.3759
1.3838	0.6859	500	1.3699
1.3789	0.8230	600	1.3652
1.3592	0.9602	700	1.3588
1.1585	1.0974	800	1.3761
1.1329	1.2346	900	1.3822

Framework versions

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