See axolotl config

axolotl version: 0.4.1

base_model: google/gemma-2b-it
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
# tokenizer_config: google/gemma-2b-it

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: /scratch/jiarui14/dpo-ood/LLM-finetune/WX-Reward-Modeling/pair-pm/ultrafeedback-single
    conversation: gemma
    type: sharegpt.load_ultrachat
    split: "train"
    train_on_split: "train"

warmup_steps: 40
val_set_size: 0.0
output_dir: ./pm_models/gemma-2b-it_lr1e-5_ultrafeedback
#wandb_project: preference-models
#wandb_entity: domain-generalization
wandb_watch:
wandb_name: "gemma-2b-it_lr1e-5_ultrafeedback"
#_response_only
wandb_log_model:

train_on_inputs: false

save_safetensors: true
#noisy_embedding_alpha: 10.0 # default for sharegpt type
dataset_prepared_path: data/gemma-2b-it/ultrafeedback


dataset_processes: 48
#torch_compile: true
sequence_len: 3072
sample_packing: true
pad_to_sequence_len: true

trust_remote_code: True
adapter:
lora_model_dir:




gradient_checkpointing: false

#warmup_ratio: 0.1
gradient_accumulation_steps: 8
micro_batch_size: 2
num_epochs: 1
optimizer: adamw_torch_fused
lr_scheduler: cosine
learning_rate: 1e-5

weight_decay: 0.0
max_grad_norm: 1.0


group_by_length: false
bf16: auto
fp16: false
tf32: true

early_stopping_patience:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true


eval_steps:
eval_table_size:
eval_table_max_new_tokens:
save_steps: 50
save_strategy: "steps"
save_total_limit: 2
debug:


ddp: #true
deepspeed: #deepspeed/zero1.json # multi-gpu only

fsdp:
fsdp_config:
special_tokens:

pm_models/gemma-2b-it_lr1e-5_ultrafeedback

This model is a fine-tuned version of google/gemma-2b-it 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: 1e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 40
• num_epochs: 1

Training results

Framework versions

• Transformers 4.43.3
• Pytorch 2.1.2+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1