meshllm/dflash-pairings

DFlash pairing index for mesh-llm

index.json maps a target model's GGUF identity (architecture, executable layer count, width, expert count, optional family) to the DFlash draft mesh-llm uses by default, pinned by Hugging Face revision and SHA-256, with the default draft cap and the backends it is the default on.

Each mesh-llm release trusts one exact revision of this repo. Every pairing here was chosen from measured decode speed on DGX Spark (CUDA) and Apple M3 Ultra / M6 (Metal): a pairing is the default on a backend only where it beat decoding without speculation.

Layer packages can instead declare their draft in generation.speculative_decoding.proposers (kind: dflash); a package declaration takes precedence over this index.