base_model: openaccess-ai-collective/tiny-mistral
base_model_config: openaccess-ai-collective/tiny-mistral
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
is_mistral_derived_model: true
model_config:
  dropout_p: 0.1

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: emrgnt-cmplxty/sciphi-textbooks-are-all-you-need
    type: completion
    field: completion

dataset_prepared_path: last_run_prepared
val_set_size: 0.001
output_dir: ./mistral-textbooks
hub_model_id: winglian/tiny-mistral-textbooks

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

wandb_project: tiny-mistral-textbooks
wandb_entity:
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 8
micro_batch_size: 4
num_epochs: 1
optimizer: adamw_torch
lr_scheduler: cosine
adam_beta2: 0.95
adam_epsilon: 0.00001
max_grad_norm: 1.0
learning_rate: 0.000025

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
eval_steps: 0.1
eval_table_size: 
eval_table_max_new_tokens: 
save_steps: 0.1
debug:
deepspeed:
weight_decay: 0.1
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"