base_model: HuggingFaceH4/zephyr-7b-beta
# model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
is_mistral_derived_model: true

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: klima7/polish-prose
    type: completion
  - path: Lajonbot/alpaca-dolly-chrisociepa-instruction-only-polish
    type: alpaca
  
dataset_prepared_path:   
val_set_size: 0.01
output_dir: ./nondzu/Mistral-7B-codealpaca-test17
adapter: qlora
lora_model_dir:
# 16384 8192 4096  2048
sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true
lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_modules:
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project: mistral-code
wandb_entity:
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 1
micro_batch_size: 2
num_epochs: 8
optimizer: paged_adamw_32bit
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
main_process_port: 0
warmup_steps: 10
eval_steps: 20
save_steps:
debug:
# deepspeed:
deepspeed: deepspeed/zero2.json
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"