base_model: meta-llama/Llama-2-7b-chat-hf
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: ../upsampled_train.json
    ds_type: json
    type: alpaca
    split: train

test_datasets:
  - path: ../val.json
    ds_type: json
    type: alpaca
    split: train

load_best_model_at_end: False
early_stopping_patience: 
dataset_prepared_path: 
val_set_size: 0
output_dir: ./qlora-out-llama2-balance-1st

adapter: qlora
lora_model_dir:

sequence_len: 4096
sample_packing: false
pad_to_sequence_len: false

lora_r: 32
lora_alpha: 16
lora_dropout: 0.15
lora_target_linear: true
lora_fan_in_fan_out:
lora_target_modules:
  - gate_proj
  - down_proj
  - up_proj
  - q_proj
  - v_proj
  - k_proj
  - o_proj

wandb_project: fine-tune-sal
wandb_entity:
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 2
micro_batch_size: 1
num_epochs: 5
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0001

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

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

wandb_project: fine-tune-sal
wandb_entity:
wandb_watch:
wandb_run_id:
wandb_log_model:

warmup_steps: 10
eval_steps: 0.05
eval_table_size:
eval_table_max_new_tokens: 128
save_steps: 
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"