See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: microsoft/Phi-3.5-mini-instruct
bf16: auto
dataset_prepared_path: null
datasets:
- data_files:
  - 71a792e171cec430_train_data.json
  ds_type: json
  format: custom
  path: /root/G.O.D-test/core/data/71a792e171cec430_train_data.json
  type:
    field_instruction: question
    field_output: answer
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: null
eval_max_new_tokens: 128
eval_table_size: null
evals_per_epoch: 10
flash_attention: false
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 4
gradient_checkpointing: false
group_by_length: false
hub_model_id: souging/6db42b4d-a926-41a4-97e5-f6919b4ae8e6
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: 1024
micro_batch_size: 2
mlflow_experiment_name: /tmp/71a792e171cec430_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 4
optimizer: adamw_bnb_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
saves_per_epoch: 10
sequence_len: 1024
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: a1fb7aa4-ffe3-49f2-8699-288a3b7ff1bc
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: a1fb7aa4-ffe3-49f2-8699-288a3b7ff1bc
warmup_steps: 150
weight_decay: 0.01
xformers_attention: null

6db42b4d-a926-41a4-97e5-f6919b4ae8e6

This model is a fine-tuned version of microsoft/Phi-3.5-mini-instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1737

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
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• total_eval_batch_size: 16
• 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: 150
• training_steps: 1024

Training results

Training Loss	Epoch	Step	Validation Loss
0.9874	0.0014	1	0.2389
0.8952	0.0356	26	0.2239
0.8988	0.0712	52	0.1942
0.7657	0.1068	78	0.1873
0.6981	0.1424	104	0.1839
0.7229	0.1780	130	0.1826
0.7599	0.2136	156	0.1820
0.7184	0.2491	182	0.1811
0.7584	0.2847	208	0.1804
0.6986	0.3203	234	0.1800
0.687	0.3559	260	0.1795
0.6937	0.3915	286	0.1790
0.7793	0.4271	312	0.1789
0.7881	0.4627	338	0.1781
0.7438	0.4983	364	0.1778
0.7136	0.5339	390	0.1775
0.6419	0.5695	416	0.1772
0.715	0.6051	442	0.1768
0.6999	0.6407	468	0.1766
0.6511	0.6762	494	0.1763
0.727	0.7118	520	0.1761
0.7074	0.7474	546	0.1757
0.6984	0.7830	572	0.1755
0.7964	0.8186	598	0.1752
0.7321	0.8542	624	0.1750
0.7149	0.8898	650	0.1747
0.7095	0.9254	676	0.1745
0.642	0.9610	702	0.1743
0.8211	0.9966	728	0.1741
0.7495	1.0322	754	0.1743
0.6667	1.0678	780	0.1743
0.6425	1.1034	806	0.1741
0.6368	1.1389	832	0.1739
0.6285	1.1745	858	0.1739
0.6018	1.2101	884	0.1738
0.6186	1.2457	910	0.1737
0.7654	1.2813	936	0.1739
0.6569	1.3169	962	0.1737
0.6282	1.3525	988	0.1737
0.6579	1.3881	1014	0.1737

Framework versions

• PEFT 0.13.2
• Transformers 4.46.0
• Pytorch 2.4.1+cu124
• Datasets 3.0.1
• Tokenizers 0.20.3