base_model: HuggingFaceH4/mistral-7b-sft-beta
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
is_mistral_derived_model: true

load_in_8bit: false
load_in_4bit: false
strict: false

rl: kto_pair
datasets:
  - path: winglian/deita-nectar
    split: train_dpo
    type: zephyr.nectar
_test_datasets:
  - path: winglian/deita-nectar
    split: test_dpo
    type: zephyr.nectar
dataset_prepared_path: last_run_prepared
val_set_size: 0.0
output_dir: ./zephyr-deita-kto-3ep-v3-r512-bsz16-cosine
save_total_limit: 3
hub_model_id: openaccess-ai-collective/kto-zephyr-deita-nectar-final

adapter: lora
lora_model_dir:

sequence_len: 2048
sample_packing: false
pad_to_sequence_len: false

lora_r: 512
lora_alpha: 256
lora_dropout: 0.05
lora_target_linear: true
lora_modules_to_save:
lora_fan_in_fan_out:
lora_target_modules:
  - gate_proj
  - down_proj
  - up_proj
  - q_proj
  - v_proj
  - k_proj
  - o_proj

wandb_project: dpo-zephyr-deita-nectar
wandb_entity: oaaic
wandb_watch:
wandb_run_id:
wandb_name: kto-3ep-v3-r512-bsz16-lr2e-5-cosine
wandb_log_model:
wandb_disabled: true

gradient_accumulation_steps: 1
micro_batch_size: 4
num_epochs: 3
optimizer: paged_adamw_8bit
adam_beta2: 0.95
adam_epsilion: 0.00001
lr_scheduler: cosine
learning_rate: 2.0e-5

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

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

warmup_steps: 10
eval_steps:
eval_table_size:
eval_table_max_new_tokens: 128
save_steps: 538
debug:
deepspeed:
weight_decay: 0.1
fsdp:
fsdp_config:
special_tokens:
save_safetensors: true

dataloader_num_workers: 16
dataloader_pin_memory: true