base_model: /root/autodl-tmp/SillyTilly/mistralai_Mistral-Nemo-Base-2407
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

#hub_model_id: taozi555/hiwaifu-12b
#hub_strategy: "all_checkpoints"
#push_dataset_to_hub:
#hf_use_auth_token: true

plugins:
  - axolotl.integrations.liger.LigerPlugin
liger_rope: true
liger_rms_norm: true
liger_swiglu: true
liger_fused_linear_cross_entropy: true

load_in_8bit: false
load_in_4bit: false
strict: false
max_steps: 5000

save_total_limit: 5
pretraining_dataset:
  - path: taozi555/novel_text
    name: default
    type: pretrain
  - path: ToastyPigeon/SpringDragon
    type: pretrain
  - path: allura-org/sugarquill-10k
    type: pretrain
  - path: allura-org/fujin-cleaned-stage-2
    type: pretrain
    
shuffle_merged_datasets: true
#default_system_message: "You are an assistant that responds to the user."
dataset_prepared_path: /root/autodl-tmp/data/
val_set_size: 0.0
output_dir: /root/autodl-tmp/hiwaifu-12b-Instruct/

sequence_len: 32768
sample_packing: true
pad_to_sequence_len: true

adapter:
lora_model_dir:
lora_r:
lora_alpha:
lora_dropout:
lora_target_linear:
lora_fan_in_fan_out:


wandb_project: hiwaifu-12b
wandb_entity:
wandb_watch:
wandb_name: hiwaifu-12b-pretrain
wandb_log_model:

gradient_accumulation_steps: 2
micro_batch_size: 1
num_epochs: 3
optimizer: paged_adamw_8bit
warmup_ratio: 0.05
learning_rate: 0.0002
lr_scheduler: cosine


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

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

## Evaluation
val_set_size: 0.0
#evals_per_epoch: 4
eval_table_size:
#eval_max_new_tokens: 128
saves_per_epoch: 2
debug:
deepspeed: /root/autodl-tmp/zero2.json
weight_decay: 0.1
fsdp:
fsdp_config:
special_tokens:
  pad_token: <pad>