gemma-2-9b config.json

hidden 3584, 42 layers, 16 q heads, 8 kv heads, head_dim 256, sliding_window 4096, attn_logit_softcapping 50.0, final_logit_softcapping 30.0, query_pre_attn_scalar 256, max_position_embeddings 8192. | 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. | SOURCE AUDIT 2026-10-03: CONSOLIDATED: A second source record (hf-edge-gemma-2-9b) pointed at this same config.json URL and was cited by nothing; its notes recorded that google/gemma-2-9b is gated so the values must be read from the unsloth mirror, and that the param count 9241705984 was confirmed against google/gemma-2-9b safetensors metadata. Both facts are retained here; the redundant record was deleted.