model_name: molmo
llm:
  d_model: 3584
  n_heads: 28
  n_kv_heads: 4
  head_dim: null
  qkv_bias: true
  clip_qkv: null
  n_layers: 28
  mlp_ratio: 4
  mlp_hidden_size: 37888
  activation_type: swiglu
  block_type: sequential
  rope: true
  rope_full_precision: true
  rope_theta: 1000000.0
  rope_type: default
  rope_factor: null
  rope_high_freq_factor: null
  rope_low_freq_factor: null
  rope_original_max_position_embeddings: null
  attention_type: sdpa
  float32_attention: true
  attention_dropout: 0.0
  attention_layer_norm: false
  attention_layer_norm_type: olmo
  residual_dropout: 0.0
  response_residual_dropout: 0.1
  layer_norm_type: rms
  layer_norm_with_affine: true
  layer_norm_eps: 1.0e-06
  attention_layer_norm_with_affine: true
  max_sequence_length: 4096
  max_position_embeddings: null
  include_bias: false
  bias_for_layer_norm: null
  norm_after: false
  moe_num_experts: 8
  moe_top_k: 2
  moe_mlp_impl: sparse
  moe_log_expert_assignment: false
  moe_shared_expert: false
  moe_lbl_in_fp32: false
  moe_interleave: false
  moe_loss_weight: 0.1
  moe_zloss_weight: null
  moe_dropless: true
  moe_capacity_factor: 1.25
  embedding_dropout: 0.0
  scale_logits: false
  vocab_size: 152064
  additional_vocab_size: 128
  weight_tying: false
  embedding_size: 152064
  use_position_ids: true
  tokenizer:
    identifier: Qwen/Qwen2.5-7B
    tokenizer_dir: null
    depth_tokens: false
  init_path: gs://mm-olmo/pretrained_llms/qwen2.5-7b.pt
  init_incremental: null
  new_embedding_init_range: 0.02
  initializer_range: 0.02
  normalize_input_embeds: false
  activation_checkpoint: whole_layer
  compile: blocks
  fix_pad_tokenizer: false
  resize_vocab: false
  init_std: 0.02
  init_fn: normal
  init_cutoff_factor: null
vision_backbone:
  vit:
    image_model_type: siglip
    image_default_input_size:
    - 378
    - 378
    image_patch_size: 14
    image_pos_patch_size: 14
    image_emb_dim: 1152
    image_num_heads: 16
    image_num_key_value_heads: 16
    image_num_layers: 27
    image_head_dim: 72
    image_mlp_dim: 4304
    image_mlp_activations: gelu_pytorch_tanh
    image_dropout_rate: 0.0
    image_num_pos: 729
    image_norm_eps: 1.0e-06
    attention_dropout: 0.0
    residual_dropout: 0.0
    initializer_range: 0.02
    float32_attention: true
    attention_type: sdpa
    activation_checkpointing: true
    init_path: gs://mm-olmo/pretrained_image_encoders/siglip2-so400m-14-384.pt
    resize_mode: siglip
    pad_value: 0.0
    normalize: siglip
  image_pooling_2d: attention_meanq
  pooling_attention_mask: false
  image_projector: mlp
  image_padding_embed: null
  vit_layers:
  - -3
  - -9
  skip_unused_layers: true
  image_feature_dropout: 0.0
  connector_activation_checkpointing: true
  compile_vit: blocks
data_formatter:
  prompt_templates: none
  message_format: none
  system_prompt: style_and_length
  always_start_with_space: false
  default_inference_len: 65
  select_answer: best
  debug: false
  image_last: false
  format_message_list: null
  p_one_message: 0.0
mm_preprocessor:
  crop_mode: overlap-and-resize-c2
  max_crops: 8
  max_images: null
  max_multi_image_crops: 6
  pooling_w: 2
  pooling_h: 2
  overlap_margins:
  - 4
  - 4
  use_col_tokens: true
  loss_token_weighting: null
  legacy_image_mask: false
  max_answer_len: null
  img_aug: false
bi_directional_attn: null
lora_enable: null
lora_rank: null
lora_alpha: null
lora_dropout: null
lora_bias: null
n_action_bins: 256
norm_stats: null