Expand description
Explicit architecture capability registry for the generic GGUF path.
Mirrors the pinned llama.cpp llm_arch / LLM_ARCH_NAMES inventory
(.scratch/llama.cpp/src/llama-arch.{h,cpp}) with Ferrox-side
classification into decoder families, memory kinds, and scope.
Unknown strings and detected-but-unimplemented features fail closed
(LoadError) instead of silently defaulting into fluent-but-wrong
logits.
Architecture names are registry keys only. Hot-path kernels never
branch on them; load-time resolution produces an ArchProfile
whose fields the decoder reads as plain data.
Structs§
- Arch
Profile - Load-time resolved profile for one GGUF
general.architecturestring.
Enums§
- Arch
Path - How the generic
Decoder/ModelConfig::from_ggufpath treats a GGUF architecture string. - Arch
Scope - How far this architecture is in Ferrox’s delivery scope (plan: text-generation parity; encoder/multimodal/diffusion/audio deferred).
- Decoder
Family - Shared execution family (maps many GGUF strings onto one engine path).
- Memory
Kind - Memory / KV backend selected once at load (llama.cpp
create_memory). - QkNorm
Style - How
attn_q_norm/attn_k_normweights are applied (when present).
Functions§
- architecture_
catalog - Full inventory keyed by GGUF
general.architecturestring. Kept in sync with.scratch/llama.cpp/src/llama-arch.cppLLM_ARCH_NAMES. - coverage_
report_ markdown - Markdown coverage table for docs / CI drift checks.
- default_
swa_ pattern - The alternating sliding-window period an architecture uses when its
GGUF carries
{arch}.attention.sliding_windowbut not{arch}.attention.sliding_window_pattern. - resolve_
architecture - Resolve a GGUF
general.architecturevalue.Nonemeans the string is not in the registry — callers must fail closed rather than guess. - resolve_
profile - Resolve a GGUF
general.architecturevalue to its profile. - swa_
rope_ base_ follows_ model - True when this architecture’s SWA layers use the model’s own RoPE
base rather than llama.cpp’s
rope_freq_base_train_swadefault of10000. - unsupported_
feature_ keys - Metadata keys that, when present with a nonzero value, require math ferrox’s generic decoder does not implement unless the architecture profile opts into those features (Gemma family).