vllm.models.deepseek_v4.cpu.mtp ¶

CPU DeepSeek-V4 MTP (speculative decoding) — deferred, not implemented.

DeepSeekV4MTP is only instantiated when vllm/model_executor/models/registry.py's speculative-decoding registry resolves it by name for a speculative-decoding config; a plain (non-spec) DeepSeek-V4 CPU model never constructs this class. This stub exists solely so vllm.models.deepseek_v4.cpu is import-clean.