base_model: mistralai/Mistral-7B-v0.1
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
is_mistral_derived_model: true

# multimodal pretrain
multimodal: true
mm_vision_tower: openai/clip-vit-large-patch14
tune_mm_mlp_adapter: true
mm_freeze_backbone: true
mm_vision_select_layer: -2
mm_projector_type: mlp2x_gelu
mm_image_folder: ./llava/
mm_use_im_patch_token: false

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: ./data/blip_laion_cc_sbu_558k.json
dataset_prepared_path:
val_set_size: 0.0
output_dir: ./out

sequence_len: 2048
sample_packing: false
pad_to_sequence_len: true

wandb_project: axolotl-llava
wandb_entity: oaaic
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 2
micro_batch_size: 64
num_epochs: 1
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 0.002

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:
save_steps: 0.1
debug:
deepspeed: deepspeed/zero2.json
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  pad_token: "<unk>"