Ministral 8B

CONTEXT SCALING: null is a VERIFIED value, checked against config rather than left blank. config.json (mistralai/Ministral-8B-Instruct-2410) declares rope_scaling null, rope_theta 100000000, max_position_embeddings 32768. Verified native. PARAMS: 8.02B is the safetensors count. KV: 36 * 8 * 128 * 2 * 2 bytes = 147,456 bytes/token = 144 KiB/token. SLIDING WINDOW THAT NEVER BINDS: sliding_window is set to 32768, exactly equal to max_position_embeddings, so at any usable context the window covers the entire sequence and behaves as standard full attention. Recording it as 'sliding-window' would be misleading; it is a full-attention model with the mechanism present but inert. LICENCE IS NOT PRODUCTION: the Ministral release is under Mistral's research licence, unlike the Apache-2.0 Mixtral line. That is a deployment blocker independent of architecture. 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. CONTEXT_SCALING_METHOD (native): rope_scaling null at theta 1e8. LAYER LAYOUT (ministral-8b): layer_types read verbatim from config, one entry per layer; attention_layer_indices are the 0-based positions of the full-attention entries.