See axolotl config

axolotl version: 0.4.1

adapter: lora
auto_find_batch_size: true
base_model: microsoft/Phi-3-mini-4k-instruct
bf16: auto
chat_template: llama3
dataloader_num_workers: 12
dataset_prepared_path: null
datasets:
- data_files:
  - 11bb3328a39885eb_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/11bb3328a39885eb_train_data.json
  type:
    field_instruction: query
    field_output: atom
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 3
early_stopping_threshold: 0.001
eval_max_new_tokens: 128
eval_steps: 20
flash_attention: false
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 2
gradient_checkpointing: false
group_by_length: false
hub_model_id: sn56t0/385d0209-5a48-4c55-afeb-fa0021266d80
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0003
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 100
lora_alpha: 16
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 8
lora_target_linear: true
lr_scheduler: cosine
micro_batch_size: 32
mlflow_experiment_name: /tmp/11bb3328a39885eb_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 5
optimizer: adamw_bnb_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
s2_attention: null
sample_packing: false
save_steps: 20
saves_per_epoch: 0
seed: 754886094
sequence_len: 512
shuffle: true
strict: false
tf32: false
tokenizer_type: AutoTokenizer
torch_compile: true
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: sn56-miner
wandb_mode: disabled
wandb_name: null
wandb_project: god
wandb_run: 38ik
wandb_runid: null
warmup_ratio: 0.05
weight_decay: 0.0
xformers_attention: null

385d0209-5a48-4c55-afeb-fa0021266d80

This model is a fine-tuned version of microsoft/Phi-3-mini-4k-instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.6789

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: 0.0003
• train_batch_size: 32
• eval_batch_size: 32
• seed: 754886094
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 256
• total_eval_batch_size: 128
• 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: 16
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0075	1	1.6674
No log	0.1509	20	0.9842
No log	0.3019	40	0.7868
No log	0.4528	60	0.7475
No log	0.6038	80	0.7317
1.7686	0.7547	100	0.7231
1.7686	0.9057	120	0.7148
1.7686	1.0566	140	0.7108
1.7686	1.2075	160	0.7068
1.7686	1.3585	180	0.7043
1.3649	1.5094	200	0.7011
1.3649	1.6604	220	0.6980
1.3649	1.8113	240	0.6959
1.3649	1.9623	260	0.6925
1.3649	2.1132	280	0.6920
1.3269	2.2642	300	0.6920
1.3269	2.4151	320	0.6893
1.3269	2.5660	340	0.6882
1.3269	2.7170	360	0.6864
1.3269	2.8679	380	0.6846
1.2806	3.0189	400	0.6831
1.2806	3.1698	420	0.6832
1.2806	3.3208	440	0.6837
1.2806	3.4717	460	0.6820
1.2806	3.6226	480	0.6815
1.2517	3.7736	500	0.6806
1.2517	3.9245	520	0.6802
1.2517	4.0755	540	0.6800
1.2517	4.2264	560	0.6797
1.2517	4.3774	580	0.6792
1.2416	4.5283	600	0.6792
1.2416	4.6792	620	0.6789
1.2416	4.8302	640	0.6789
1.2416	4.9811	660	0.6789

Framework versions

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