config.json

{
  "auto_map": {
      "AutoConfig": "config_molmoe.MolmoConfig",
      "AutoModelForCausalLM": "modeling_molmoe.MolmoForCausalLM"
  },
  "activation_type": "swiglu",
  "additional_vocab_size": 128,
  "alibi": false,
  "alibi_bias_max": 8.0,
  "always_start_with_space": true,
  "architectures": [
    "OLMoForCausalLM"
  ],
  "attention_dropout": 0.0,
  "attention_layer_norm": true,
  "attention_layer_norm_with_affine": true,
  "attention_type": "sdpa",
  "attn_logit_softcapping": null,
  "bias_for_layer_norm": false,
  "block_group_size": 1,
  "block_type": "moe",
  "clip_qkv": null,
  "crop_mode": "overlap-and-resize-c2",
  "d_model": 2048,
  "default_inference_len": 65,
  "do_random_scale": false,
  "embedding_dropout": 0.0,
  "embedding_size": 50304,
  "final_logit_softcapping": null,
  "fix_image_input_idx": 2,
  "float32_attention": true,
  "gin_bindings": null,
  "head_dim": null,
  "image_feature_dropout": 0.0,
  "image_padding_embed": "pad_and_partial_pad",
  "image_pooling_2d": "attention-meanq",
  "image_pooling_h": 2,
  "image_pooling_w": 2,
  "image_projector": "mlp",
  "include_bias": false,
  "init_cutoff_factor": 3.0,
  "init_device": "meta",
  "init_fn": "normal",
  "init_std": 0.02,
  "initializer_range": 0.02,
  "layer_norm_eps": 1e-05,
  "layer_norm_type": "rms",
  "layer_norm_with_affine": true,
  "llm_load_path": null,
  "loss_token_weighting": "root_subsegments",
  "low_cpu_fsdp": true,
  "max_crops": 12,
  "max_position_embeddings": 32768,
  "max_sequence_length": 4096,
  "message_formatting": "role",
  "mlp_hidden_size": null,
  "mlp_ratio": 1,
  "model_type": "molmo",
  "moe_capacity_factor": 1.25,
  "moe_dropless": true,
  "moe_interleave": false,
  "moe_lbl_in_fp32": false,
  "moe_log_expert_assignment": false,
  "moe_loss_weight": 0.0,
  "moe_mlp_impl": "sparse",
  "moe_num_experts": 64,
  "moe_shared_expert": false,
  "moe_top_k": 8,
  "moe_zloss_weight": 0.0,
  "multi_query_attention": null,
  "n_heads": 16,
  "n_kv_heads": null,
  "n_layers": 16,
  "new_embedding_init_range": 0.02,
  "norm_after": false,
  "normalize_input_embeds": false,
  "overlap_margins": [
    4,
    4
  ],
  "pad_to": null,
  "pad_token_id": 1,
  "pad_tokenizer": false,
  "precision": "amp_bf16",
  "prompt_override": null,
  "prompt_type": "uber_model",
  "qkv_bias": false,
  "query_pre_attn_scalar": 224,
  "residual_dropout": 0.1,
  "response_attention_dropout": 0.0,
  "response_residual_dropout": 0.0,
  "rope": true,
  "rope_full_precision": true,
  "rope_impl": "llama",
  "rope_theta": 10000.0,
  "scale_logits": false,
  "system_prompt_kind": "demo_or_style",
  "tokenizer": {
    "identifier": "allenai/gpt-neox-olmo-dolma-v1_5",
    "olmo_bos_token_id": null,
    "olmo_eos_token_id": null,
    "tokenizer_adds_space": false,
    "tokenizer_dir": null,
    "truncate_direction": "right"
  },
  "transformers_version": "4.45.0.dev0",
  "unconditioned": false,
  "use_cache": true,
  "use_cls_feature": false,
  "use_col_tokens": true,
  "use_position_ids": true,
  "vision_backbone": {
    "attention_dropout": 0.0,
    "fsdp_wrap": false,
    "image_default_input_size": [
      336,
      336
    ],
    "image_dropout_rate": 0.0,
    "image_emb_dim": 1024,
    "image_head_dim": 64,
    "image_mlp_activations": "quick_gelu",
    "image_mlp_dim": 4096,
    "image_model_type": "openai",
    "image_norm_eps": 1e-05,
    "image_num_heads": 16,
    "image_num_key_value_heads": 16,
    "image_num_layers": 23,
    "image_num_pos": 577,
    "image_patch_size": 14,
    "image_pos_patch_size": 14,
    "initializer_range": 0.02,
    "residual_dropout": 0.0,
    "resize_mode": "default"
  },
  "vit_layers": [
    -2,
    -9
  ],
  "vit_load_path": null,
  "vocab_size": 50280,
  "weight_tying": false
}