See axolotl config

axolotl version: 0.4.1

base_model: JackFram/llama-68m
batch_size: 32
bf16: true
chat_template: tokenizer_default_fallback_alpaca
datasets:
- data_files:
  - ff3a521d02fa72b2_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/ff3a521d02fa72b2_train_data.json
  type:
    field_instruction: context
    field_output: question
    format: '{instruction}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
eval_steps: 20
flash_attention: true
gpu_memory_limit: 80GiB
gradient_checkpointing: true
group_by_length: true
hub_model_id: willtensora/4ada8092-cc1e-445c-9260-a580ef2586ae
hub_strategy: checkpoint
learning_rate: 0.0002
logging_steps: 10
lr_scheduler: cosine
max_steps: 2500
micro_batch_size: 4
model_type: AutoModelForCausalLM
optimizer: adamw_bnb_8bit
output_dir: /workspace/axolotl/configs
pad_to_sequence_len: true
resize_token_embeddings_to_32x: false
sample_packing: false
save_steps: 40
save_total_limit: 1
sequence_len: 2048
special_tokens:
  pad_token: </s>
tokenizer_type: LlamaTokenizerFast
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.1
wandb_entity: ''
wandb_mode: online
wandb_name: JackFram/llama-68m-/workspace/input_data/ff3a521d02fa72b2_train_data.json
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: default
warmup_ratio: 0.05
xformers_attention: true

4ada8092-cc1e-445c-9260-a580ef2586ae

This model is a fine-tuned version of JackFram/llama-68m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2208

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: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 32
• total_eval_batch_size: 32
• 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: 205

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0006	1	6.7193
1.5212	0.0122	20	1.0774
0.7826	0.0244	40	0.6352
0.5492	0.0366	60	0.4713
0.3663	0.0488	80	0.3924
0.3533	0.0610	100	0.3112
0.2434	0.0732	120	0.2761
0.2989	0.0854	140	0.2445
0.2464	0.0976	160	0.2251
0.2233	0.1098	180	0.2203
0.2213	0.1220	200	0.2208

Framework versions

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