Meta-Llama-3.1-8B-Instruct config.json

Meta's own repo is gated; this is a mirror of the same weights. hidden 4096, 32 layers, 32 q heads, 8 kv heads, no explicit head_dim (implied 128), rope_theta 500000, rope_scaling llama3 factor 8 from an 8192 base, max_position_embeddings 131072. | 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.