config.json

{
  "architectures": [
    "MiniMaxM3VLVisionModel"
  ],
  "attention_dropout": 0.0,
  "dtype": "bfloat16",
  "hidden_act": "gelu",
  "hidden_size": 1280,
  "image_size": 2016,
  "img_token_compression_config": {
    "image_token_compression_method": "patch_merge",
    "spatial_merge_size": 2,
    "temporal_patch_size": 2
  },
  "initializer_factor": 1.0,
  "initializer_range": 0.02,
  "intermediate_size": 5120,
  "layer_norm_eps": 1e-05,
  "model_type": "minimax_m3_vl_vision",
  "num_attention_heads": 16,
  "num_channels": 3,
  "num_hidden_layers": 32,
  "patch_size": 14,
  "position_embedding_type": "rope",
  "projection_dim": 6144,
  "rope_mode": "3d",
  "rope_parameters": {
    "rope_theta": 10000.0,
    "rope_type": "axial"
  },
  "spatial_merge_size": 2,
  "temporal_patch_size": 2,
  "transformers_version": "5.17.0",
  "vision_segment_max_frames": 4,
  "vocab_size": 32000
}