llama.cpp VS transformers

> 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

Hugging Face Transformers: 500,000+ lines

"casually dropping the most capable open weights on the planet" — @RyanMullins

Google folks do something really cool!

Gemma4 source: https://github.com/huggingface/transformers/pull/45192

It is a tokenizer artifact most likely (https://github.com/huggingface/transformers/issues/4786). So the output is not properly decoded in this case, it should just be a space.

Ready to build your own language AI? Explore Hugging Face's Transformers library and test your skills with our interactive coding challenges at AIAcademy.tech!

This time even Unsloth could not provide bitsandbytes 4-bit models. bitsandbytes does not support new models with MoE and linear attentions, and it's much less flexible than GGUF. Nowadays I think it's better to train lora over GGUF base model, see the discussion at https://github.com/huggingface/transformers/issues/40070

I'll find some time to do this and I hope someone can do this earlier than me.

Judging by the code in the HF transformers repo[1], smaller dense versions of this model will most likely be released at some point. Hopefully, soon.

[1]: https://github.com/huggingface/transformers/tree/main/src/tr...

Hugging Face Transformers is an open-source Python library that provides a unified API for working with Transformer-based models. The library handles model loading, tokenization, and inference, while pre-trained model checkpoints are hosted on the Hugging Face Hub and automatically downloaded when used.