See axolotl config

axolotl version: 0.4.1

adapter: qlora
base_model: unsloth/Phi-3-medium-4k-instruct
bf16: auto
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 9cd9d6ddd992cd94_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/9cd9d6ddd992cd94_train_data.json
  type:
    field_input: input
    field_instruction: system_prompt
    field_output: reference_answer
    format: '{instruction} {input}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 5
eval_max_new_tokens: 128
eval_steps: 200
eval_table_size: null
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 4
gradient_checkpointing: true
group_by_length: true
hub_model_id: error577/65665304-fb0e-4e82-8266-aba26a9f6ca0
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0002
load_in_4bit: true
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 16
lora_dropout: 0.15
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 8
lora_target_linear: true
loraplus_lr_ratio: 16
lr_scheduler: constant_with_warmup
micro_batch_size: 2
mlflow_experiment_name: /tmp/9cd9d6ddd992cd94_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 3
optimizer: paged_ademamix_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
restore_best_weights: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
save_steps: 200

sequence_len: 256
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.002
wandb_entity: null
wandb_mode: online
wandb_name: 23ddb23c-7a61-40c1-9f72-48b94618385b
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 23ddb23c-7a61-40c1-9f72-48b94618385b
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

65665304-fb0e-4e82-8266-aba26a9f6ca0

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

• Loss: 0.0604

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.PAGED_ADEMAMIX_8BIT and the args are: No additional optimizer arguments
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 10
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
4.8043	0.0001	1	1.1370
3.1671	0.0212	200	0.3959
1.3667	0.0424	400	0.1787
1.3113	0.0637	600	0.0892
0.5292	0.0849	800	0.2110
0.3196	0.1061	1000	0.0461
0.5194	0.1273	1200	0.0534
3.7961	0.1485	1400	0.0516
0.3986	0.1698	1600	0.0575
0.3479	0.1910	1800	0.0679
0.777	0.2122	2000	0.0604

Framework versions

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