Expert caching that greatly increases performance. Self contained, off by default, use -ehs N to activate by miltos22 · Pull Request #26563 · ggml-org/llama.cpp
github.com
Overview I want to make this as short as possible. I made a CUDA only for now feature where it tracks a heat map of all expert usage and the hottest experts are cached on GPU and computed there, wh...
git remote add miltos22 https://github.com/miltos22/llama.cpp-wackMall-merge-request.git
git checkout miltos22/master
cmake -B build -DGGML_CUDA=ON
cmake --build build --config Release -j $(nproc)
GGML_CUDA_ENABLE_UNIFIED_MEMORY=1 LLAMA_API_KEY=myapikeyhere ./build/bin/llama-server --models-preset ~/models.ini --models-max 1

In my models.ini file I’m using:

  • expert-hot-s = 10 for Gemma 4 26b a4b Q4_K_M QAT
    • [56187] GPU hot store allocated: CUDA0, 1300296448 bytes (1240 MiB) for 10+1 slots (10 expert + 1 sentinel)
    • From about 17 tps up to about 23 (+35%)
  • expert-hot-s = 32 for Qwen 3.6 35b a3b Q4_K_S
    • [33181] GPU hot store allocated: CUDA0, 2422292480 bytes (2310 MiB) for 32+1 slots (32 expert + 1 sentinel)
    • From about 25 tps up to about 29 (+16%)

This is basically a smarter version of n-cpu-moe. You could also set it to -1 for auto fill, but I don’t think that works with GGML_CUDA_ENABLE_UNIFIED_MEMORY=1. I could probably push these numbers a bit higher, but I like having Chrome open.

Performance varies, and it takes time for it to adjust which experts are loaded into VRAM. You’ll see lines in the log output like:

[40117] === Expert hot store: re-sync swapped 5 expert slots ===

You’ll want to keep llama open for multiple prompts for it to learn which are the hot experts, and it can vary by task. It doesn’t seem to ever hurt performance. There might be a bug in the Gemma handling, I saw it fail tool calls a couple times and I don’t think it ever did that before, I’ll revisit when this is merged.

In the pull request there are more knobs you can play with looking in common/arg.cpp but I haven’t tested them:

--expert-heat-decay F                   expert heatmap decay rate per update (default: 0.999)
                                        (env: LLAMA_ARG_EXPERT_HEAT_DECAY)
--expert-heat-log-period N              expert heatmap log interval in updates (default: 100, 0 = off)
                                        (env: LLAMA_ARG_EXPERT_HEAT_LOG_PERIOD)
--expert-hyst F                         expert hot store hysteresis ratio (default: 1.3, 0 = off)
                                        (env: LLAMA_ARG_EXPERT_HYST)
--expert-dwell N                        expert hot store minimum dwell updates before swap (default: 0 = off)
                                        (env: LLAMA_ARG_EXPERT_DWELL)
-ehs,  --expert-hot-s N                 -1 = autofit slots from free VRAM, 0 = disabled, N = manual top-N
                                        slots
                                        (env: LLAMA_ARG_EXPERT_HOT_S)

Right now this is only for CUDA. Someone try this for DeepSeek V4 Flash 0731

  • BeefAndPoultry@lemmus.orgOPEnglish
    1·
    12 hours ago

    (Oops I got my Gemma and Qwen speeds mixed up, edited the post to fix it.)

    But now with the new commits they added, with the same number of hot experts, Qwen is up to about 34. If I increase hot experts to 48 then I get around 37.

    Gemma is still around 23 with just 10 hot experts. With 16 hot experts I get about 26 TPS. If I overprovision my VRAM (thanks to GGML_CUDA_ENABLE_UNIFIED_MEMORY=1) then 24 hot experts can give me 29 TPS, and 32 hot experts 34 TPS.