See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: JackFram/llama-68m
bf16: auto
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 425c6bf4bb96a710_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/425c6bf4bb96a710_train_data.json
  type:
    field_input: paragraph
    field_instruction: question
    field_output: answer
    format: '{instruction} {input}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: null
eval_max_new_tokens: 128
eval_table_size: null
evals_per_epoch: 4
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 4
gradient_checkpointing: true
gradient_clipping: 1.0
group_by_length: false
hub_model_id: Nexspear/2453e11c-1506-4485-a4fd-b29c0185bb57
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 5.0e-05
load_in_4bit: false
load_in_8bit: false
local_rank: 0
logging_steps: 3
lora_alpha: 32
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: 100
micro_batch_size: 8
mlflow_experiment_name: /tmp/425c6bf4bb96a710_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 3
optimizer: adamw_bnb_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
saves_per_epoch: 4
sequence_len: 1024
special_tokens:
  pad_token: </s>
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: leixa-personal
wandb_mode: online
wandb_name: 832d80d5-251c-46fe-b13b-7cc0427dac5a
wandb_project: Gradients-On-Four
wandb_run: your_name
wandb_runid: 832d80d5-251c-46fe-b13b-7cc0427dac5a
warmup_steps: 10
weight_decay: 0.01
xformers_attention: null

2453e11c-1506-4485-a4fd-b29c0185bb57

This model is a fine-tuned version of JackFram/llama-68m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.1608

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• 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: 100

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0052	1	4.9409
5.2272	0.0466	9	4.8344
4.4814	0.0931	18	4.3345
3.3043	0.1397	27	3.8139
3.7606	0.1863	36	3.3687
2.1562	0.2329	45	2.9507
2.9386	0.2794	54	2.5972
2.4696	0.3260	63	2.3809
2.0397	0.3726	72	2.2495
1.8449	0.4191	81	2.1891
2.1496	0.4657	90	2.1661
2.2016	0.5123	99	2.1608

Framework versions

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