See axolotl config

axolotl version: 0.5.2

adapter: lora
base_model: NousResearch/Llama-3.2-1B
bf16: auto
datasets:
- data_files:
  - 5b5d462a3104f60a_train_data.json
  ds_type: json
  format: custom
  path: 5b5d462a3104f60a_train_data.json
  preprocessing:
  - shuffle: true
  type:
    field: null
    field_input: text
    field_instruction: document_type
    field_output: entities
    field_system: null
    format: null
    no_input_format: null
    system_format: '{system}'
    system_prompt: ''
debug: null
device_map: auto
early_stopping_patience: null
eval_max_new_tokens: 32
evals_per_epoch: 8
flash_attention: null
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 4
gradient_checkpointing: true
group_by_length: false
hub_model_id: taopanda/0c1875b8-6e01-45de-816e-58e722dc0eb1
learning_rate: 0.0005
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 128
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 64
lora_target_modules:
- gate_proj
- down_proj
- up_proj
- q_proj
- v_proj
- k_proj
- o_proj
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: 500
micro_batch_size: 4
model_type: LlamaForCausalLM
num_epochs: 1
optimizer: adamw_torch
output_dir: ./outputs/lora-out/taopanda_28ee0c8e-531a-42b6-813c-c18fc5b57bf8
pad_to_sequence_len: true
resume_from_checkpoint: null
save_steps: 0.25
save_total_limit: 1
seed: 42
sequence_len: 1024
special_tokens:
  pad_token: <|end_of_text|>
strict: false
tf32: true
tokenizer_type: AutoTokenizer
torch_compile: true
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.1
wandb_entity: fatcat87-taopanda
wandb_log_model: null
wandb_mode: online
wandb_name: taopanda_28ee0c8e-531a-42b6-813c-c18fc5b57bf8
wandb_project: subnet56-test
wandb_runid: taopanda_28ee0c8e-531a-42b6-813c-c18fc5b57bf8
wandb_watch: null
warmup_ratio: 0.06
weight_decay: 0.01
xformers_attention: null

0c1875b8-6e01-45de-816e-58e722dc0eb1

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: 0.0423

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: 0.0005
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• total_eval_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH 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: 30
• training_steps: 500

Training results

Training Loss	Epoch	Step	Validation Loss
1.0533	0.0012	1	1.1479
0.0907	0.0748	63	0.0931
0.084	0.1496	126	0.0918
0.0813	0.2243	189	0.0753
0.071	0.2991	252	0.0611
0.0479	0.3739	315	0.0536
0.0529	0.4487	378	0.0457
0.0367	0.5234	441	0.0423

Framework versions

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