adapter: lora
base_model: JackFram/llama-160m
bf16: auto
datasets:
- data_files:
  - ab80a554070cdc53_train_data.json
  ds_type: json
  format: custom
  path: ab80a554070cdc53_train_data.json
  type:
    field: null
    field_input: rejected
    field_instruction: prompt
    field_output: chosen
    field_system: null
    format: null
    no_input_format: null
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: null
eval_max_new_tokens: 128
eval_sample_packing: false
eval_table_size: null
evals_per_epoch: 4
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 4
gradient_checkpointing: true
group_by_length: false
hub_model_id: FatCat87/1b054327-a6b9-45e0-98ef-2d5fb503dfb8
learning_rate: 0.0002
load_in_4bit: false
load_in_8bit: true
local_rank: null
logging_steps: 1
lora_alpha: 16
lora_dropout: 0.05
lora_r: 32
lora_target_linear: true
lr_scheduler: cosine
micro_batch_size: 2
model_type: AutoModelForCausalLM
num_epochs: 1
optimizer: adamw_bnb_8bit
output_dir: ./outputs/out
pad_to_sequence_len: true
resume_from_checkpoint: null
sample_packing: true
saves_per_epoch: 1
seed: 701
sequence_len: 4096
special_tokens:
  pad_token: </s>
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
val_set_size: 0.1
wandb_entity: fatcat87-taopanda
wandb_log_model: null
wandb_mode: online
wandb_name: 1b054327-a6b9-45e0-98ef-2d5fb503dfb8
wandb_project: subnet56
wandb_runid: 1b054327-a6b9-45e0-98ef-2d5fb503dfb8
wandb_watch: null
warmup_ratio: 0.05
weight_decay: 0.0
xformers_attention: null

Visualize in Weights & Biases

1b054327-a6b9-45e0-98ef-2d5fb503dfb8

This model is a fine-tuned version of JackFram/llama-160m on the None dataset. It achieves the following results on the evaluation set:

Loss: 2.7122

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: 701
distributed_type: multi-GPU
num_devices: 2
gradient_accumulation_steps: 4
total_train_batch_size: 16
total_eval_batch_size: 4
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 2
num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
3.7539	0.0308	1	3.8105
3.4309	0.2462	8	3.3958
3.0223	0.4923	16	2.9836
2.7418	0.7385	24	2.7555
2.6571	0.9846	32	2.7122

Framework versions

PEFT 0.11.1
Transformers 4.42.3
Pytorch 2.3.0+cu121
Datasets 2.19.1
Tokenizers 0.19.1