model_config.json

{
  "model_name": "gemma4-e4b",
  "architecture": "gemma4",
  "hidden_size": 2560,
  "num_hidden_layers": 42,
  "num_attention_heads": 8,
  "num_key_value_heads": 2,
  "head_dim": 256,
  "global_head_dim": 512,
  "vocab_size": 262144,
  "context_length": 2048,
  "sliding_window": 512,
  "per_layer_dim": 256,
  "num_layers": 42,
  "embed_scale": 50.59644256269407,
  "per_layer_embed_scale": 16.0,
  "per_layer_model_projection_scale": 0.01976423537605237,
  "per_layer_input_scale": 0.7071067811865475,
  "rms_norm_eps": 1e-06,
  "bos_token_id": 2,
  "eos_token_id": 1,
  "final_logit_softcapping": 30.0,
  "quantization": "int4",
  "compute_units": "CPU_AND_NE",
  "tokenizer_repo": "google/gemma-4-E4B-it"
}