base_model: alpindale/Mistral-7B-v0.2-hf
tokenizer_type: AutoTokenizer
is_mistral_derived_model: true
load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: json
    data_files: hidden_pretraining_manners.jsonl
    ds_type: json
    type: completion


dataset_prepared_path: last_run_prepared
output_dir: ./army-pretraining

sequence_len: 4096
sample_packing: false
pad_to_sequence_len: true
shuffle_merged_datasets: true

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

gradient_accumulation_steps: 6
micro_batch_size: 2
eval_batch_size: 1
num_epochs: 11
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 0.000020
weight_decay: 0
# Gradient clipping max norm
max_grad_norm: 1.0
noisy_embedding_alpha: 0
train_on_inputs: false
group_by_length: false
bf16: true
fp16: false
tf32: false

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

chat_template: chatml

warmup_ratio: 0.5
auto_resume_from_checkpoints: false
#warmup_ratio: 0.5
eval_steps: 10
saves_per_epoch: 1
eval_sample_packing: false
save_total_limit: 3
debug:
deepspeed: deepspeed_configs/zero2.json
special_tokens:
  pad_token: "<|end_of_text|>"

pretrained base for Mannerstral 7b, only use if you are finetuning something on top of it.

Heralax/Mannerstral-base

Heralax/
Mannerstral-base