tandem.cpp
tandem.cpp Boosts Long-Prompt Speed 2–6x, but Still Falls Short
tandem.cpp v0.3.0 uses expert caching and 4,096-token batches to speed up local Qwen3.8 and DeepSeek-V4.1 prompts by 2–6x.
1 article tagged with this topic
tandem.cpp v0.3.0 uses expert caching and 4,096-token batches to speed up local Qwen3.8 and DeepSeek-V4.1 prompts by 2–6x.