base_model: NousResearch/Meta-Llama-3-8B
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer  # PreTrainedTokenizerFast

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: alexandrainst/lexdk-open
    type: completion
  - path: syvai/wiki-da
    type: completion


dataset_prepared_path: last_run_prepared
val_set_size: 0.001
output_dir: ./out/qlora-llama3-70b

adapter: qlora

lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:
lora_target_modules:


wandb_project: llama-3
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true
eval_sample_packing: false


gradient_accumulation_steps: 1
micro_batch_size: 1
num_epochs: 2
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

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

warmup_steps: 10
evals_per_epoch: 10
eval_table_size:
saves_per_epoch: 2
debug:
deepspeed:
weight_decay: 0.1
fsdp:
fsdp_config:
special_tokens:
  pad_token: <|end_of_text|>