llama.cpp VS alpaca.cpp

> The benchmark prompt was:

> Write a compact Python function that parses a unified diff and returns the changed file paths. Then explain two edge cases.

> Each benchmark generated about 128 tokens.

Generating 128 tokens is probably not enough for good benchmark results. MTP speedup depends on how often the predicted tokens are accepted. In my experience, the very early output has a higher acceptance rate, so short testing can give false positive speedups.

Also llama.cpp includes a tool specifically for benchmarking:

https://github.com/ggml-org/llama.cpp/blob/master/tools/llam...

In my build, MTP came from mainline llama.cpp, not ik_llama. ik_llama got me to ~47 (engine + quant), but I couldn't get MTP running there — my build rejected the -mtp flags and ignored the model's nextn tensors. Mainline llama.cpp added MTP fairly recently (PR #22673, merged 2026-05-16), and that's where it worked for me. (There may well be an ik_llama path I missed — this is just what got it going on my box.)

That script grew up. Today I'm releasing LlamaStash, the first public release of a fast, cross-platform, terminal-native launcher for llama.cpp with zero overhead.

LlamaStash spawns the unmodified upstream llama-server. So three different questions follow from that, and there is a benchmark suite for each.

llama.cpp includes a benchmarking tool called llama-bench https://github.com/ggml-org/llama.cpp/blob/master/tools/llam...

ik_llama includes llama-sweep-bench https://github.com/ikawrakow/ik_llama.cpp/blob/main/examples...

When comparing hardware, the output of these tools is very helpful to let others put it into context. The post says the output is "reading speed" but knowing the prefill and token generation speeds would be a lot more helpful.

lol the same way we implement all of the reduced precision fp8, fp4 types today: by storing them in the corresponding uint:

https://github.com/ggml-org/llama.cpp/discussions/15095

# Build llama.cpp with Metal backend git clone https://github.com/ggml-org/llama.cpp cd llama.cpp && cmake -B build -DGGML_METAL=ON && cmake --build build -j # Community-quantized GGUFs (Google ships safetensors; unsloth ships GGUF) huggingface-cli download unsloth/gemma-4-31B-it-GGUF \ gemma-4-31B-it-Q4_K_M.gguf --local-dir . huggingface-cli download unsloth/gemma-4-26B-A4B-it-GGUF \ gemma-4-26B-A4B-it-Q4_K_M.gguf --local-dir . # Benchmark: 200 generations of 512 tokens, log per-call timing ./build/bin/llama-bench -m gemma-4-31B-it-Q4_K_M.gguf -n 512 -r 200 -o json > dense.json ./build/bin/llama-bench -m gemma-4-26B-A4B-it-Q4_K_M.gguf -n 512 -r 200 -o json > moe.json

Where's the 30B-in-6GB claim? ^FGB in your GH link finds [0] which is neither by jart nor by ggerganov but by another user who promptly gets told to look at [1] where Justine denies that claim.

  [0] https://github.com/antimatter15/alpaca.cpp/issues/182

You can just download the language model, dude!!! Everyone doesn’t need to make their own and the open source models literally get better every day.

Alpaca

Square could have also trained a more specific data source for their NLP, very similar to Alpaca. Alpaca was trained from interactions from a larger dataset. So while it isn't as smart, it's still able to understand instructions and act upon them.

The model I found was in 8 parts. The alpaca.cpp chat client (chat.cpp) needs to be modified to run the 8 part model, documented here: https://github.com/antimatter15/alpaca.cpp/issues/149

try the instructions on this github repo https://github.com/antimatter15/alpaca.cpp, its not the best one but I was able to run this model on my linux machine with 16GB memory, I think its a good starting point.

Doesn't run offline. If you're running something without a beefy-ish GPU, there's https://github.com/antimatter15/alpaca.cpp .

Ye, atm. toying with alpaca 7B/13B in a local install.