model.yaml

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: 0.1
  response_residual_dropout: 0.0
  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: true
  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: uber_model
  message_format: role
  system_prompt: demo_or_style
  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: 2
  max_multi_image_crops: 8
  pooling_w: 2
  pooling_h: 2
  overlap_margins:
  - 4
  - 4
  use_col_tokens: true
  loss_token_weighting: root_subsegments
  legacy_image_mask: false
  max_answer_len: null
  img_aug: true
bi_directional_attn: null
lora_enable: true
lora_rank: 32
lora_alpha: 16
lora_dropout: 0.0
lora_bias: none
n_action_bins: 256
norm_stats:
  libero_spatial_no_noops_modified:
    action:
      mean:
      - 0.15312479436397552
      - 0.13707277178764343
      - -0.15526802837848663
      - -0.005176450591534376
      - -0.01120874285697937
      - -0.020194264128804207
      - 0.4578818082809448
      std:
      - 0.41272708773612976
      - 0.34724321961402893
      - 0.50869220495224
      - 0.037266165018081665
      - 0.07244449853897095
      - 0.05762382969260216
      - 0.49827873706817627
      max:
      - 0.9375
      - 0.9375
      - 0.9375
      - 0.1971428543329239
      - 0.33642858266830444
      - 0.375
      - 1.0
      min:
      - -0.9375
      - -0.9375
      - -0.9375
      - -0.1875
      - -0.3675000071525574
      - -0.36000001430511475
      - 0.0
      q01:
      - -0.7454732114076613
      - -0.6616071462631226
      - -0.9375
      - -0.1071428582072258
      - -0.20678570866584778
      - -0.1842857152223587
      - 0.0
      q99:
      - 0.9375
      - 0.8758928775787354
      - 0.9321428537368774
      - 0.1039285734295845
      - 0.17678570747375488
      - 0.14571428298950195
      - 1.0
    proprio:
      mean:
      - -0.024462558329105377
      - 0.106529600918293
      - 1.0580483675003052
      - 3.0628468990325928
      - -0.10464039444923401
      - 0.08307311683893204
      - 0.0
      - 0.01995457336306572
      - -0.020162804052233696
      std:
      - 0.1101478561758995
      - 0.13784688711166382
      - 0.1044282391667366
      - 0.10451053828001022
      - 0.4112098217010498
      - 0.2176690548658371
      - 0.0
      - 0.017260896041989326
      - 0.0171116404235363
      max:
      - 0.1759040206670761
      - 0.3904820382595062
      - 1.3290715217590332
      - 3.4566118717193604
      - 1.2268599271774292
      - 1.0429412126541138
      - 0.0
      - 0.041053611785173416
      - 0.000775813648942858
      min:
      - -0.3095473051071167
      - -0.29250794649124146
      - 0.9095591306686401
      - 2.497488260269165
      - -1.8006486892700195
      - -0.7207611203193665
      - 0.0
      - -0.0004703797458205372
      - -0.041536275297403336
      q01:
      - -0.2727657300233841
      - -0.23721413239836692
      - 0.9160063165426254
      - 2.77949666261673
      - -1.3187511622905732
      - -0.41989982962608335
      - 0.0
      - 0.001503719249740243
      - -0.03989770736545324
      q99:
      - 0.13529365032911292
      - 0.3629165390133857
      - 1.2862326657772063
      - 3.2829698753356933
      - 0.9332760351896285
      - 0.6325724506378171
      - 0.0
      - 0.039933966137468815
      - -0.001671919699292631
    num_transitions: 52970
    num_trajectories: 432