See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: microsoft/phi-2
bf16: auto
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - c46283424f0b5d2a_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/c46283424f0b5d2a_train_data.json
  type:
    field_instruction: question
    field_output: query
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: null
eval_max_new_tokens: 128
eval_steps: 50
eval_table_size: null
flash_attention: false
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 4
gradient_checkpointing: false
group_by_length: false
hub_model_id: oliverchang/18dac2dc-fb84-47b9-aaf7-7ceb1242f230
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0002
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 1
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: 20000000
micro_batch_size: 2
mlflow_experiment_name: /tmp/c46283424f0b5d2a_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 1
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: 50
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: online
wandb_name: 18dac2dc-fb84-47b9-aaf7-7ceb1242f230
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 18dac2dc-fb84-47b9-aaf7-7ceb1242f230
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

18dac2dc-fb84-47b9-aaf7-7ceb1242f230

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: 0.4429

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.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: 1128

Training results

Training Loss	Epoch	Step	Validation Loss
1.8994	0.0009	1	2.0234
0.939	0.0444	50	0.8451
0.7808	0.0887	100	0.7711
0.9628	0.1331	150	0.7313
0.6268	0.1774	200	0.6124
0.8939	0.2218	250	0.5876
0.6337	0.2661	300	0.5524
0.513	0.3105	350	0.5673
0.7696	0.3548	400	0.5626
0.5483	0.3992	450	0.5483
0.4969	0.4436	500	0.5449
0.4714	0.4879	550	0.5080
0.4638	0.5323	600	0.4898
0.3513	0.5766	650	0.4703
0.4209	0.6210	700	0.4609
0.541	0.6653	750	0.4660
0.412	0.7097	800	0.4570
0.4849	0.7540	850	0.4576
0.4414	0.7984	900	0.4516
0.5111	0.8428	950	0.4461
0.53	0.8871	1000	0.4435
0.4098	0.9315	1050	0.4421
0.4856	0.9758	1100	0.4429

Framework versions

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