PaliGemma 3B mix 224 text-model Q4_K_M GGUF header (abetlen conversion)

Official google/paligemma-3b-mix-224 config.json is gated (HTTP 401, gated manual). This public GGUF converts that checkpoint; general.name is paligemma-3b-mix-224 and the repo card license is gemma. First 200KB of the text-model Q4_K_M fetched and GGUF v3 header parsed. TEXT tower: architecture gemma, context_length 8192, embedding_length 2048, block_count 18, feed_forward_length 16384, attention.head_count 8, attention.head_count_kv 1, attention.key_length 256, attention.value_length 256, file_type 15 (Q4_K_M). Same repo ships the text model at f16 and q8_0 plus a separate mmproj-f16 vision tower, so llama.cpp runs the checkpoint. Third-party conversion, not Google own config.json.