See axolotl config

axolotl version: 0.12.1

adapter: lora
base_model: bigscience/bloom-560m
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - a14930b6eeb6d1b3_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/
  type:
    field_instruction: instruct
    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: 4
flash_attention: false
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 16
gradient_checkpointing: true
group_by_length: false
hub_model_id: eason668/4b509713-486e-488a-bf91-393179e986f5
learning_rate: 5.0e-05
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
lr_scheduler: cosine
max_steps: 1500
micro_batch_size: 8
mlflow_experiment_name: /tmp/a14930b6eeb6d1b3_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 2
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: 512
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.1
wandb_entity: null
wandb_mode: offline
wandb_name: 4b509713-486e-488a-bf91-393179e986f5
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 4b509713-486e-488a-bf91-393179e986f5
warmup_steps: 100
weight_decay: 0.0
xformers_attention: null

4b509713-486e-488a-bf91-393179e986f5

This model is a fine-tuned version of bigscience/bloom-560m on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 2.2940
• Memory/max Mem Active(gib): 13.09
• Memory/max Mem Allocated(gib): 13.09
• Memory/device Mem Reserved(gib): 15.75

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
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 16
• total_train_batch_size: 1024
• total_eval_batch_size: 64
• 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: 100
• training_steps: 1500

Training results

Training Loss	Epoch	Step	Validation Loss	Mem Active(gib)	Mem Allocated(gib)	Mem Reserved(gib)
No log	0	0	3.1634	10.72	10.72	11.38
2.5436	0.8716	188	2.5547	13.09	13.09	15.75
2.4108	1.7418	376	2.4190	13.09	13.09	15.75
2.3911	2.6120	564	2.3619	13.09	13.09	15.75
2.3574	3.4822	752	2.3295	13.09	13.09	15.75
2.3197	4.3524	940	2.3099	13.09	13.09	15.75
2.2956	5.2225	1128	2.2985	13.09	13.09	15.75
2.2586	6.0927	1316	2.2940	13.09	13.09	15.75

Framework versions

• PEFT 0.17.0
• Transformers 4.55.0
• Pytorch 2.6.0+cu124
• Datasets 4.0.0
• Tokenizers 0.21.4