See axolotl config

axolotl version: 0.3.0

base_model: justinj92/phi2-platypus
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
is_llama_derived_model: false
trust_remote_code: true

load_in_8bit: false
load_in_4bit: false
strict: false

rl: true
datasets:
  - path: Intel/orca_dpo_pairs
    split: train
    type: intel_apply_chatml
  - path: argilla/ultrafeedback-binarized-preferences
    split: train
    type: argilla_apply_chatml
dataset_prepared_path: ./dpoplatypus-phi2/last_run_prepared
val_set_size: 0.0
output_dir: ./dpoplatypus-phi2/
#'Wqkv', 'out_proj', 'fc2', 'linear', 'fc1'
adapter:
sequence_len: 2048
sample_packing: false
pad_to_sequence_len:

lora_r: 64
lora_alpha: 32
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:
lora_modules_to_save:
  - embd
  - lm_head
hub_model_id: justinj92/phi2-platypus-dpo


wandb_project: phi2-platypus-dpo
wandb_entity: justinjoy-5
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 1
num_epochs: 1
optimizer: adamw_torch
adam_beta2: 0.95
adam_epsilion: 0.00001
lr_scheduler: cosine
max_grad_norm: 1.0
learning_rate: 0.00002

train_on_inputs: false
group_by_length: false
bf16: true
fp16: false
tf32: true

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

warmup_steps: 100
eval_steps:
evals_per_epoch: 4
saves_per_epoch: 2
eval_table_size:
debug:
deepspeed:
weight_decay: 0.1
fsdp:
fsdp_config:
resize_token_embeddings_to_32x: true
special_tokens:
  pad_token: "<|endoftext|>"

dpoplatypus-phi2

This model is a fine-tuned version of justinj92/phi2-platypus 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-05
• train_batch_size: 1
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 4
• optimizer: Adam with betas=(0.9,0.95) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 19120

Training results

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.0.1+cu117
• Datasets 2.16.1
• Tokenizers 0.15.0