kf_config.json

{
  "library_name": "kerasformers",
  "kerasformers_version": "1.2.4",
  "model_module": "kerasformers.models.glm4v_moe",
  "model_class": "Glm4vMoeConditionalGenerate",
  "variant": "glm-4.6v",
  "weights": "model.weights.json",
  "schema_version": 2,
  "weight_dtype": "bfloat16",
  "weight_dtype_overrides": {
    "e_score_correction_bias": "float32"
  },
  "model_type": "glm4v_moe",
  "text_config": {
    "vocab_size": 151552,
    "embed_dim": 4096,
    "mlp_dim": 10944,
    "moe_mlp_dim": 1408,
    "num_layers": 46,
    "num_heads": 96,
    "num_kv_heads": 8,
    "head_dim": 128,
    "num_experts": 128,
    "num_experts_per_tok": 8,
    "n_shared_experts": 1,
    "n_group": 1,
    "topk_group": 1,
    "norm_topk_prob": true,
    "routed_scaling_factor": 1.0,
    "first_k_dense": 1,
    "partial_rotary_factor": 0.5,
    "norm_eps": 1e-05,
    "rope_theta": 500000,
    "mrope_section": [
      8,
      12,
      12
    ],
    "tie_embeddings": false,
    "vision_depth": 24,
    "vision_embed_dim": 1536,
    "vision_num_heads": 12,
    "vision_mlp_dim": 10944,
    "vision_out_dim": 4096,
    "image_size": 336,
    "patch_size": 14,
    "spatial_merge_size": 2,
    "temporal_patch_size": 2,
    "in_channels": 3,
    "vision_norm_eps": 1e-05,
    "image_token_id": 151363,
    "video_token_id": 151364,
    "image_start_token_id": 151339,
    "image_end_token_id": 151340,
    "video_start_token_id": 151341,
    "video_end_token_id": 151342
  }
}