See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: EleutherAI/pythia-410m-deduped
batch_size: 8
bf16: true
chat_template: tokenizer_default_fallback_alpaca
datasets:
- data_files:
  - 7b17ba91d0356ec9_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/7b17ba91d0356ec9_train_data.json
  type:
    field_instruction: prompt
    field_output: completion
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
early_stopping_patience: 3
eval_steps: 50
flash_attention: true
gpu_memory_limit: 80GiB
gradient_checkpointing: true
group_by_length: true
hub_model_id: willtensora/bf58c716-f1ab-4447-897b-55ade88ff479
hub_strategy: checkpoint
learning_rate: 0.0002
logging_steps: 10
lora_alpha: 256
lora_dropout: 0.1
lora_r: 128
lora_target_linear: true
lr_scheduler: cosine
micro_batch_size: 1
model_type: AutoModelForCausalLM
num_epochs: 100
optimizer: adamw_bnb_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
resize_token_embeddings_to_32x: false
sample_packing: false
save_steps: 50
sequence_len: 2048
special_tokens:
  pad_token: <|endoftext|>
tokenizer_type: GPTNeoXTokenizerFast
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.1
wandb_entity: ''
wandb_mode: online
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: default
warmup_ratio: 0.05
xformers_attention: true

bf58c716-f1ab-4447-897b-55ade88ff479

This model is a fine-tuned version of EleutherAI/pythia-410m-deduped on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.4376

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: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 8
• total_eval_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: 153
• num_epochs: 100

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0041	1	1.8043
1.4581	0.2033	50	1.4781
1.7801	0.4065	100	1.7743
1.6691	0.6098	150	1.7577
2.3973	0.8130	200	2.4376

Framework versions

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