gemma-4-26B-A4B-it config.json

text_config: hidden 2816, 30 layers, 16 q heads, 8 kv heads, head_dim 256, global_head_dim 512, num_global_key_value_heads 2, sliding_window 1024, 5 full_attention layers, num_experts 128, top_k_experts 8, moe_intermediate_size 704, attention_k_eq_v true, max_position_embeddings 262144, final_logit_softcapping 30.0. | SOURCE AUDIT 2026-10-03: kind was 'spec-sheet' but the URL is a raw Hugging Face config.json served as text/plain, not a vendor spec sheet; set to 'repo'. The schema field it fills (layer/window/head counts) is machine-readable model metadata and should be read as such.