llama.cpp VS whisper.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

whisper.cpp is Georgi Gerganov's port of OpenAI's Whisper model to plain C++. It runs on CPU, no GPU needed, and the smallest model (ggml-tiny.bin) is only about 75MB.

That's it. The plugin handles format conversion, chunking, and transcription automatically. For fully offline transcription, install whisper.cpp instead.

The first issue is that running Whisper eats up some serious computational resources, so it's not ideal to have it running constantly. In the Whisper WASM port, there is an example of a wake word detection implementation command.wasm, which partially addresses this with VAD, only running transcription once voice activity is detected.

Thank you for taking the time to build something and share it. However what is the advantage of using this over whisper.cpp stream that can also do real time conversion?

https://github.com/ggml-org/whisper.cpp/tree/master/examples...

Whisper and the many variants. Here's a good implementation.

https://github.com/ggml-org/whisper.cpp

Whisper large v3 from openai, but we host it ourselves on Modal.com. It's easy, fast, no rate limits, and cheap as well.

If you want to run it locally, I'd still go with whisper, then I'd look at something like whisper.cpp https://github.com/ggml-org/whisper.cpp. Runs quite well.

not the gp but found this https://github.com/ggml-org/whisper.cpp/blob/master/models/c...

Actually that's a really good question, I hadn't considered that the comparison here is just CPU vs using Metal (CPU+GPU).

To answer the question though - I think this would be used for cases where you are building an app that wants to utilize a small AI model while at the same time having the GPU free to do graphics related things, which I'm guessing is why Apple stuck these into their hardware in the first place.

Here is an interesting comparison between the two from a whisper.cpp thread - ignoring startup times - the CPU+ANE seems about on par with CPU+GPU: https://github.com/ggml-org/whisper.cpp/pull/566#issuecommen...

A great thread with the type of info your looking for lives here: https://github.com/ggerganov/whisper.cpp/issues/89

But you can likely find similar threads for the llama.cpp benchmark here: https://github.com/ggerganov/llama.cpp/tree/master/examples/...

These are good examples because the llama.cpp and whisper.cpp benchmarks take full advantage of the Apple hardware but also take full advantage of non-Apple hardware with GPU support, AVX support etc.

It’s been true for a while now that the memory bandwidth of modern Apple systems in tandem with the neural cores and gpu has made them very competitive Nvidia for local inference and even training.