See axolotl config

axolotl version: 0.4.1

adapter: qlora
auto_resume_from_checkpoints: true
base_model: furiosa-ai/mlperf-gpt-j-6b
bf16: auto
chat_template: llama3
dataloader_num_workers: 12
dataset_prepared_path: null
datasets:
- data_files:
  - b051361718f88ddb_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/b051361718f88ddb_train_data.json
  type:
    field_input: input
    field_instruction: instruction
    field_output: output
    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: 100
eval_table_size: null
evals_per_epoch: null
flash_attention: false
fp16: false
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 4
gradient_checkpointing: true
group_by_length: true
hub_model_id: error577/060d3dbf-6ed8-4093-b68f-bc293f1307f5
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: 10
lora_alpha: 128
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 64
lora_target_linear: true
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: null
micro_batch_size: 2
mlflow_experiment_name: /tmp/b051361718f88ddb_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 3
optimizer: adamw_torch_4bit
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
save_steps: 100
sequence_len: 512
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.001
wandb_entity: null
wandb_mode: online
wandb_name: 966d5a5a-e872-4bc8-805b-43448f9e513a
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 966d5a5a-e872-4bc8-805b-43448f9e513a
warmup_steps: 30
weight_decay: 0.0
xformers_attention: null

060d3dbf-6ed8-4093-b68f-bc293f1307f5

This model is a fine-tuned version of furiosa-ai/mlperf-gpt-j-6b on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.4625

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_TORCH_4BIT 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: 30
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0000	1	4.5361
11.7356	0.0039	100	2.8860
11.498	0.0078	200	2.7240
11.5482	0.0116	300	2.7027
11.2156	0.0155	400	2.6278
10.884	0.0194	500	2.5724
10.8781	0.0233	600	2.5185
11.1212	0.0272	700	2.5492
10.52	0.0310	800	2.4964
11.77	0.0349	900	2.5035
9.9839	0.0388	1000	2.4881
10.7824	0.0427	1100	2.4743
10.2496	0.0466	1200	2.4504
10.418	0.0504	1300	2.4633
10.3929	0.0543	1400	2.4624
10.5461	0.0582	1500	2.4625

Framework versions

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