RyanYr/llama31-it-preference_data_v2_800K_wsafety_v2

See axolotl config

axolotl version: 0.4.1

adapter: null
base_model: /var/lib/condor/execute/slot1/dir_2464950/llama31_pretrain_pad
bf16: true
dataset_prepared_path: /var/lib/condor/execute/slot1/dir_2464950/prepare
dataset_processes: 48
datasets:
- conversation: llama-3
  path: RLHFlow/preference_data_v2_80K_wsafety
  split: train
  train_on_split: train
  type: sharegpt.load_ultrachat
ddp: null
debug: null
deepspeed: null
early_stopping_patience: null
eval_steps: null
eval_table_max_new_tokens: null
eval_table_size: null
flash_attention: false
fp16: false
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 128
gradient_checkpointing: true
group_by_length: false
hub_model_id: RyanYr/llama31-it-preference_data_v2_800K_wsafety
hub_strategy: every_save
learning_rate: 2.0e-06
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 2
lora_model_dir: null
lr_scheduler: cosine
max_grad_norm: null
micro_batch_size: 2
model_type: AutoModelForCausalLM
num_epochs: 1
optimizer: paged_adamw_32bit
output_dir: /var/lib/condor/execute/slot1/dir_2464950/output-08-13-2024-08:47
pad_to_sequence_len: true
sample_packing: true
save_safetensors: true
save_steps: 100
save_strategy: steps
save_total_limit: 1
sequence_len: 4096
special_tokens: null
strict: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.0
wandb_entity: yyr
wandb_log_model: null
wandb_name: llama31-8b-it_preference_data_v2_80K_wsafety
wandb_project: preference-models
wandb_watch: null
warmup_ratio: 0.03
weight_decay: 0.0
xformers_attention: null

llama31-it-preference_data_v2_800K_wsafety

This model was trained from scratch on the None dataset.

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: 2e-06
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 128
• total_train_batch_size: 1024
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 4
• num_epochs: 1

Training results

Framework versions

• Transformers 4.44.0
• Pytorch 2.1.2+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1