base_model: PocketDoc/Dans-PersonalityEngine-V1.2.0-24b
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

hub_model_id: NewEden/DansPE-LN
hub_strategy: "all_checkpoints"
push_dataset_to_hub:
hf_use_auth_token: true

plugins:
  - axolotl.integrations.liger.LigerPlugin
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin
liger_rope: true
liger_rms_norm: true
liger_layer_norm: true
liger_glu_activation: true
liger_fused_linear_cross_entropy: false
cut_cross_entropy: true

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: NewEden/Wata-Oshi-LNs-ShareGPT-Final
    type: dan-chat-advanced
  - path: NewEden/Playground-RP
    type: dan-chat-advanced
  - path: NewEden/Light-Novels-ShareGPT
    type: dan-chat-advanced
dataset_prepared_path: prepared_data
val_set_size: 0.0
output_dir: ./output
sequence_len: 16384
sample_packing: true
pad_to_sequence_len: true

wandb_project: Hamanasu-LN
wandb_entity:
wandb_watch:
wandb_name: 
wandb_log_model:


gradient_accumulation_steps: 2
micro_batch_size: 2
num_epochs: 4
optimizer: apollo_adamw
lr_scheduler: rex
learning_rate: 2e-6
max_grad_norm: 0.1

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

warmup_steps: 35
saves_per_epoch: 2
debug:
deepspeed: ./deepspeed_configs/zero3_bf16.json
weight_decay: 0.02
fsdp:
fsdp_config:
special_tokens: