See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: NousResearch/Llama-3.2-1B
bf16: true
chat_template: llama3
data_processes: 54
dataset_prepared_path: null
datasets:
- data_files:
  - a1fd3b8799300f0a_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/a1fd3b8799300f0a_train_data.json
  type:
    field_instruction: constraints
    field_output: prompt
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
device_map: auto
distributed_training:
  multi_gpu: true
  num_gpus: 2
do_eval: true
early_stopping_patience: 4
eval_batch_size: 8
eval_max_new_tokens: 128
eval_steps: 150
eval_table_size: null
evals_per_epoch: null
flash_attention: true
fp16: false
fsdp:
- full_shard
fsdp_config:
  activation_checkpointing: false
  backward_prefetch: BACKWARD_POST
  forward_prefetch: FORWARD_POST
  fsdp_min_num_params: 1000000000
  limit_all_gathers: true
  mixed_precision: bf16
  sharding_strategy: FULL_SHARD
gradient_accumulation_steps: 2
gradient_checkpointing: true
group_by_length: true
hub_model_id: cimol/a799937b-9418-4d0f-927f-7ec8073d086a
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 7.0e-05
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 64
lora_dropout: 0.04
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 32
lora_target_linear: true
lr_scheduler: cosine
lr_scheduler_warmup_steps: 100
max_grad_norm: 1.0
max_memory:
  0: 80GB
  1: 80GB
max_steps: 750
micro_batch_size: 8
mlflow_experiment_name: /tmp/a1fd3b8799300f0a_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 3
optim_args:
  adam_beta1: 0.9
  adam_beta2: 0.95
  adam_epsilon: 1e-8
optimizer: adamw_torch
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
save_steps: 300
saves_per_epoch: null
seed: 17333
sequence_len: 1024
special_tokens:
  pad_token: <|end_of_text|>
strict: false
tf32: true
tokenizer_type: AutoTokenizer
total_train_batch_size: 32
train_batch_size: 16
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: 7ba6d308-f32a-4ef8-86b8-9e9b314b6fb0
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 7ba6d308-f32a-4ef8-86b8-9e9b314b6fb0
warmup_steps: 100
weight_decay: 0.0
xformers_attention: null

a799937b-9418-4d0f-927f-7ec8073d086a

This model is a fine-tuned version of NousResearch/Llama-3.2-1B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.5127

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: 7e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 17333
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• total_eval_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=adam_beta1=0.9,adam_beta2=0.95,adam_epsilon=1e-8
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• training_steps: 750

Training results

Training Loss	Epoch	Step	Validation Loss
2.0615	0.0017	1	2.8681
1.9878	0.2564	150	1.7228
1.8938	0.5128	300	1.5987
1.8665	0.7692	450	1.5405
1.4281	1.0256	600	1.5162
1.4781	1.2821	750	1.5127

Framework versions

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