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:
  - 37c72b217abcec12_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/37c72b217abcec12_train_data.json
  type:
    field_instruction: input
    field_output: output
    format: '{instruction}'
    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: 1
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 8
gradient_checkpointing: true
group_by_length: false
hub_model_id: error577/e2a3724d-0804-4a0c-ab1e-8004e89575a3
hub_repo: null
hub_strategy: end
hub_token: null
learning_rate: 0.0001
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 16
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 32
lora_target_linear: true
lr_scheduler: cosine
max_steps: 3000
micro_batch_size: 1
mlflow_experiment_name: /tmp/37c72b217abcec12_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 12
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: 1
sequence_len: 4096
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: null
wandb_mode: online
wandb_name: 9b409c46-5502-443f-9955-550b9ba58a9e
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 9b409c46-5502-443f-9955-550b9ba58a9e
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

e2a3724d-0804-4a0c-ab1e-8004e89575a3

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: 2.6628

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: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 8
• 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: 3000

Training results

Training Loss	Epoch	Step	Validation Loss
2.9193	0.0006	1	3.6011
3.0411	0.1421	250	2.9425
2.5867	0.2842	500	2.8315
2.5223	0.4263	750	2.7849
2.6007	0.5684	1000	2.7457
3.0427	0.7105	1250	2.7177
2.6826	0.8526	1500	2.6940
2.4515	0.9947	1750	2.6883
2.0376	1.1368	2000	2.6751
2.8151	1.2790	2250	2.6700
2.6503	1.4211	2500	2.6634
2.3753	1.5632	2750	2.6627
2.9334	1.7053	3000	2.6628

Framework versions

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