See axolotl config

axolotl version: 0.4.1

adapter: qlora
auto_resume_from_checkpoints: true
base_model: facebook/opt-125m
bf16: auto
chat_template: llama3
dataloader_num_workers: 12
dataset_prepared_path: null
datasets:
- data_files:
  - 47b36a24df61e9c5_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/47b36a24df61e9c5_train_data.json
  type:
    field_input: documents
    field_instruction: question
    field_output: answer
    format: '{instruction} {input}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 3
eval_max_new_tokens: 128
eval_steps: 50
eval_table_size: null
evals_per_epoch: null
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 16
gradient_checkpointing: true
group_by_length: true
hub_model_id: error577/5f0159c4-1008-4527-9092-4ee6e6b9e663
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0003
load_in_4bit: true
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 128
lora_dropout: 0.3
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 128
lora_target_linear: true
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: null
micro_batch_size: 1
mlflow_experiment_name: /tmp/47b36a24df61e9c5_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 3
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
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.02
wandb_entity: null
wandb_mode: online
wandb_name: a6924886-18eb-47b1-8a4b-24becc99648c
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: a6924886-18eb-47b1-8a4b-24becc99648c
warmup_steps: 10
weight_decay: 0.01
xformers_attention: null

5f0159c4-1008-4527-9092-4ee6e6b9e663

This model is a fine-tuned version of facebook/opt-125m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.2857

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: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 16
• total_train_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: 10
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
47.7848	0.0004	1	3.0276
34.159	0.0202	50	2.9019
39.2367	0.0405	100	2.5922
52.0151	0.0607	150	2.5356
33.4878	0.0809	200	2.5099
25.6957	0.1012	250	2.4814
27.5454	0.1214	300	2.4519
32.6855	0.1417	350	2.4207
25.3411	0.1619	400	2.4211
27.4427	0.1821	450	2.4128
34.6101	0.2024	500	2.3944
23.8259	0.2226	550	2.3888
23.7378	0.2428	600	2.3808
27.431	0.2631	650	2.3735
26.069	0.2833	700	2.3755
20.5981	0.3035	750	2.3722
23.1821	0.3238	800	2.3646
20.5374	0.3440	850	2.3509
22.8665	0.3642	900	2.3556
21.9577	0.3845	950	2.3418
20.0986	0.4047	1000	2.3399
29.616	0.4250	1050	2.3433
25.8536	0.4452	1100	2.3335
18.732	0.4654	1150	2.3298
21.2083	0.4857	1200	2.3250
20.2594	0.5059	1250	2.3195
14.3002	0.5261	1300	2.3196
24.714	0.5464	1350	2.3132
22.0257	0.5666	1400	2.3093
16.7176	0.5868	1450	2.3012
15.5525	0.6071	1500	2.3052
20.5451	0.6273	1550	2.2970
31.716	0.6475	1600	2.2905
23.2587	0.6678	1650	2.2938
16.72	0.6880	1700	2.2914
19.7095	0.7083	1750	2.2868
25.7639	0.7285	1800	2.2802
30.8813	0.7487	1850	2.2860
25.8737	0.7690	1900	2.2825
21.8546	0.7892	1950	2.2857

Framework versions

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