Granite 3.3 8B Instruct

CONTEXT SCALING: null is a VERIFIED value here, checked against config rather than left blank. config.json (ibm-granite/granite-3.3-8b-instruct) declares rope_scaling null, rope_theta 10000000, max_position_embeddings 131072. Verified null. Architecture: GraniteForCausalLM, hidden 4096, 40 layers, 32 query heads / 8 KV heads, rope_theta 10000000, 131072 native context, vocab 49159 with tie_word_embeddings true. The serving-relevant shape is 40 layers at hidden 4096 - same width as Llama-3.1-8B but 25% deeper, and only 32 query heads instead of Llama's 32 at the same width, so this is a deeper-narrower-attention design with a large MLP (intermediate_size 12800) relative to the attention block. More layers at the same KV-head count means more KV per token than the 8B models it competes with, at identical context length. head_dim derived as 4096 / 32 = 128; gqa_ratio = 4.KV bytes/token at bf16 (2 bytes/element): 40 layers x 8 KV heads x 128 head_dim x 2 (K and V) x 2 bytes = 163840 B/token = 160.0 KiB/token. At 128k context that is 20.00 GiB of KV per sequence at bf16. context_scaling is null: context length is native to the pretrained config, no rope_scaling. SERVICEABILITY JOIN (added 2026-10-03; the architecture x engine matrix this record was missing, with a citation per cell in docs/13-model-compatibility.md): FAMILY=dense-gqa. REQUIRED KERNEL CLASS: ordinary causal attention plus GEMM/GEMV. This is the baseline every engine in this repo implements, so the join here is a floor rather than a filter: no engine record in data/engines/ states that dense GQA is unsupported. SUPPORTED BY (cited): vLLM LlamaForCausalLM / Qwen2 / Qwen3 / Mistral / Phi3ForCausalLM rows [[mdl2-vllm-supported-models-2026-10]]; TensorRT-LLM LlamaForCausalLM, Qwen2ForCausalLM, Qwen3ForCausalLM, Phi3ForCausalLM rows [[mdl2-trtllm-supported-models]]; SGLang generative table (Llama, Qwen, Mistral families) [[mdl2-sglang-generative-models]]; llama.cpp LLM_ARCH_LLAMA [[pap-arch-llamacpp-arch-registry]]. HARD CONSTRAINTS THAT DO APPLY even to dense GQA, and they are model-shape rather than architecture-class: tensor-parallel size must divide the attention head count or the engine refuses to start [[mgpu-tp-size-must-divide-attention-heads]] (affects vllm and sglang); and a GQA KV cache is the thing paged-attention machinery exists for, so an engine with no paged KV (llama.cpp's slot-based pool [[llama-cpp]], exllamav2 [[exllamav2]]) loses the concurrency lever rather than the ability to run. ONE HARDWARE-DEPENDENT KERNEL GATE THAT IS NOT ABOUT GQA: on RDNA, vLLM's custom paged-attention kernel is gated on head_size == 128 AND block_size == 16, so a dense GQA model with head_dim 256 falls to the Triton 2D fallback and decode falls off a cliff - measured 12.1 tok/s at 518 tokens of context to 4.2 tok/s at 32k, with kernel_paged_attention_2d growing 28.3x [[rdna-triton-paged-attn-decode-cliff]]. Check head_dim before blaming the card.