See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: JackFram/llama-160m
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 2750bead1adffdf3_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/2750bead1adffdf3_train_data.json
  type:
    field_input: input
    field_instruction: instruction
    field_output: output
    format: '{instruction} {input}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
device_map:
  ? ''
  : 0,1,2,3,4,5,6,7
early_stopping_patience: 2
eval_max_new_tokens: 128
eval_steps: 400
eval_table_size: null
flash_attention: true
gradient_accumulation_steps: 4
gradient_checkpointing: true
group_by_length: false
hub_model_id: Alphatao/d4d74e47-e280-409c-8e49-f68226bea56b
hub_repo: null
hub_strategy: null
hub_token: null
learning_rate: 0.0002
load_best_model_at_end: true
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 32
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 16
lora_target_linear: true
lora_target_modules:
- q_proj
- k_proj
- v_proj
- o_proj
- down_proj
- up_proj
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: 25681
micro_batch_size: 2
mlflow_experiment_name: /tmp/2750bead1adffdf3_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 10
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: 400
sequence_len: 2048
special_tokens:
  pad_token: </s>
strict: false
tf32: true
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: 2cb63b6b-8245-4658-a5db-bef379134a64
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 2cb63b6b-8245-4658-a5db-bef379134a64
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

d4d74e47-e280-409c-8e49-f68226bea56b

This model is a fine-tuned version of JackFram/llama-160m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.5597

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• 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: 25681

Training results

Training Loss	Epoch	Step	Validation Loss
3.4657	0.0003	1	3.3169
2.0325	0.1339	400	1.8400
1.7227	0.2678	800	1.7728
2.0482	0.4017	1200	1.7269
1.8619	0.5356	1600	1.6967
1.8042	0.6695	2000	1.6784
1.6159	0.8033	2400	1.6537
1.9969	0.9372	2800	1.6380
1.5519	1.0711	3200	1.6279
1.8034	1.2050	3600	1.6202
1.1583	1.3389	4000	1.6118
2.0106	1.4728	4400	1.6035
1.1075	1.6067	4800	1.5968
2.2332	1.7406	5200	1.5909
1.6162	1.8745	5600	1.5831
1.2261	2.0084	6000	1.5807
1.6027	2.1423	6400	1.5816
1.5871	2.2762	6800	1.5782
1.1991	2.4100	7200	1.5711
1.6726	2.5439	7600	1.5675
1.3379	2.6778	8000	1.5628
1.4592	2.8117	8400	1.5580
1.9095	2.9456	8800	1.5543
1.2568	3.0795	9200	1.5595
0.9862	3.2134	9600	1.5597

Framework versions

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