See axolotl config

axolotl version: 0.4.1

adapter: lora
auto_find_batch_size: false
base_model: microsoft/phi-2
bf16: auto
chat_template: llama3
dataloader_num_workers: 12
dataset_prepared_path: null
datasets:
- data_files:
  - 5e36b0dfd0daa960_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/5e36b0dfd0daa960_train_data.json
  type:
    field_input: ''
    field_instruction: question
    field_output: answer
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 3
eval_max_new_tokens: 128
eval_steps: 12
eval_strategy: null
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 8
gradient_checkpointing: true
group_by_length: false
hub_model_id: mrferr3t/0b701c4d-c0dd-4869-a595-dbaef1150740
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0004
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 12
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
max_steps: 
micro_batch_size: 16
mlflow_experiment_name: /tmp/5e36b0dfd0daa960_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 100
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: 12
saves_per_epoch: 0
sequence_len: 512
special_tokens:
  pad_token: <|endoftext|>
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: 
wandb_name: 5810cd80-bcf6-4aaa-80fe-c8a0864b56d8
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 5810cd80-bcf6-4aaa-80fe-c8a0864b56d8
warmup_steps: 100
weight_decay: 0.0
xformers_attention: null

0b701c4d-c0dd-4869-a595-dbaef1150740

This model is a fine-tuned version of microsoft/phi-2 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.9062

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: 0.0004
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 8
• total_train_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: 100
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0398	1	1.0693
1.1042	0.4776	12	1.0669
1.0844	0.9552	24	1.0213
1.0981	1.4328	36	0.9759
0.9907	1.9104	48	0.9481
1.044	2.3881	60	0.9411
0.9661	2.8657	72	0.9332
1.022	3.3433	84	0.9306
0.9604	3.8209	96	0.9335
1.0204	4.2985	108	0.9257
0.9454	4.7761	120	0.9301
1.0118	5.2537	132	0.9222
0.9359	5.7313	144	0.9220
1.0025	6.2090	156	0.9157
0.9201	6.6866	168	0.9116
0.9909	7.1642	180	0.9098
0.9095	7.6418	192	0.9063
0.9738	8.1194	204	0.9030
0.8978	8.5970	216	0.9029
0.9746	9.0746	228	0.9043
0.8927	9.5522	240	0.8971
0.9479	10.0299	252	0.8916
0.8778	10.5075	264	0.8962
0.8857	10.9851	276	0.8961
0.9468	11.4627	288	0.9062

Framework versions

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