See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: bigscience/bloom-560m
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 21ec2939d9fc2533_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/21ec2939d9fc2533_train_data.json
  type:
    field_instruction: input
    field_output: reference_answer
    format: '{instruction}'
    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: 100
eval_table_size: null
flash_attention: false
gradient_accumulation_steps: 8
gradient_checkpointing: true
group_by_length: false
hub_model_id: Alphatao/22727b12-d31f-4226-a2a9-02abb2c6e793
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:
- query_key_value
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: 6528
micro_batch_size: 4
mlflow_experiment_name: /tmp/21ec2939d9fc2533_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
save_steps: 100
sequence_len: 1024
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: dee939d9-c79b-4a3e-944f-022f31bbb8c0
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: dee939d9-c79b-4a3e-944f-022f31bbb8c0
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

22727b12-d31f-4226-a2a9-02abb2c6e793

This model is a fine-tuned version of bigscience/bloom-560m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0002

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• 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: 4481

Training results

Training Loss	Epoch	Step	Validation Loss
19.6167	0.0004	1	2.5468
15.8246	0.0446	100	1.8829
10.6629	0.0893	200	1.4239
7.8789	0.1339	300	0.9419
4.6349	0.1786	400	0.5631
2.6567	0.2232	500	0.3090
2.854	0.2678	600	0.1852
2.0261	0.3125	700	0.1162
1.3446	0.3571	800	0.0873
0.3713	0.4018	900	0.0635
0.2305	0.4464	1000	0.0454
0.2785	0.4910	1100	0.0338
0.4851	0.5357	1200	0.0274
0.126	0.5803	1300	0.0203
0.254	0.6250	1400	0.0188
0.1323	0.6696	1500	0.0115
0.0483	0.7142	1600	0.0100
0.0887	0.7589	1700	0.0080
0.0767	0.8035	1800	0.0066
0.0597	0.8482	1900	0.0049
0.037	0.8928	2000	0.0054
0.034	0.9374	2100	0.0033
0.0309	0.9821	2200	0.0038
0.0285	1.0267	2300	0.0027
0.0329	1.0714	2400	0.0019
0.038	1.1160	2500	0.0016
0.0306	1.1606	2600	0.0013
0.0135	1.2053	2700	0.0011
0.0088	1.2499	2800	0.0007
0.006	1.2946	2900	0.0006
0.0032	1.3392	3000	0.0005
0.0049	1.3839	3100	0.0010
0.0077	1.4285	3200	0.0003
0.0028	1.4731	3300	0.0003
0.0027	1.5178	3400	0.0003
0.0033	1.5624	3500	0.0002
0.0035	1.6071	3600	0.0002
0.0035	1.6517	3700	0.0002
0.0021	1.6963	3800	0.0002

Framework versions

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