57 comments

[ 2.8 ms ] story [ 79.9 ms ] thread
Is there a link to the video? It did not render when I went to the page. Curious about the real-time feel of this
Not sure you really need huggingface-cli to download anything if you're just using llama.cpp. You can pass `-hf ...` and it will download the models for you. Set `LLAMA_CACHE` to change where the downloads go:

  LLAMA_CACHE="models" ./llama-server \
    -hf unsloth/gemma-4-31B-it-GGUF:UD-Q4_K_XL \
    ...
I wrote a similar post some time ago just used ollama and opencode https://blog.kulman.sk/running-local-llm-coding-server/
Ollama is not a good choice - https://sleepingrobots.com/dreams/stop-using-ollama/

As for oprncode, doesn't the system prompt eat too much of the context? Local models are really constraint in regards contex, and opencode AFAIR uses a 10k of it or some thing close.

Ollama seems fine to me, technically. it works. why wouldn't someone use it? Because someone else doesn't like it? if it cost money, I'd pay more attention to the people behind it, but I don't, so I don't.

I wish people would stop wasting their outrage budget on things like this and pay more attention to politics.

Focusing outrage on politics is pointless.

Technical people are rather good at learning new things, and ollama situation is a good learning experience.

llama.cpp gets you more tokens/s even if you ignore ollama team bad behavior.

Political outrage is only pointless (in the cases when it is actually pointless), because not enough people get outraged.

Politicians count on your apathy so they can get away with their horseshit. You paying attention is kryptonite to crooked politicians.

Did you have any issue with tool calling inside opencode? I tried the same approach, but my models don't see any tool.
Useful stuff in here that I wish I'd seen a few days ago :-)

I am not convinced that the MTP setup for the QAT model adds very much in terms of speed on my M1 Max, but it is definitely worth experimenting with.

Fiddling about with local models has done so much for my conceptual understanding of what is going on.

FWIW and YMMV but I also found the Gemma 4 MTP head was occasionally breaking markup in Opencode, causing the thinking to display untidily and ultimately in some cases missing the stop token. So I've stopped using MTP there for now.

Recent Qwen 3.6 models have developer role support so it will occasionally surprise you with a structured multiple choice questionnaire.

oMLX (https://github.com/jundot/omlx) makes running the mlx inference server quite easy for those interested in UI-based hosting. oMLX also supports mtp or dflash drafting.
Whay about of the tons of caches that just pile up until you notice that you must delete them manually?
>64 GB

Thats the rub. I have an M4 with 48G. I wonder if it is worth testing this out.

My past attempts (with Ollama and various LLMs) were too slow to use.

i’m running m4 pro 48gb right now

omlx + gemma 12b 6 bit + pi

it’s feasible for sure

MoEs for speed (qwen 35b, cohere 30b, gemma 26b)

Dense for more methodical work (qwen 27b [reigning champ], gemma 31b, gemma 12b)

MoE i recommend 5bit+

Dense i think 4 bit is okay

Play with your context size, you don’t really need that much, have lazy loading for tools and mcps

my pi extensions for anyone looking for a skinny quick setup, i have use `--no-skills` right now too:

    "npm:pi-codex-goal",
    "npm:pi-simplify",
    "npm:pi-mcp-adapter",
    "git:github.com/elpapi42/pi-minimal-subagent",
    "npm:@wierdbytes/pi-statusline",
    "npm:@aliou/pi-guardrails",
    "npm:pi-lens",
    "npm:@juicesharp/rpiv-todo",
    "npm:pi-hashline-readmap",
    "npm:@mrclrchtr/supi-review",
    "npm:pi-cmux",
    "npm:@mrclrchtr/supi-context",
    "npm:pi-tool-search"

think of local models as "zero sugar" models and that's where we're at right now. I think it's crazy how good these models are compared to last year's frontier models
People are using 3090 (24GB) to run models, and it is the most cost effective way to run the. Yes, it is 2x faster, but memory wise you surely can spend 24gb on llm.

Also there are smaller, still usefull models that can run on 8GB or less.

I've an M1 Pro with 32GB ram and it's running pretty well
8b max on a std 16gb macbook. Anything more and your mac is toast
> 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.

llama.cpp includes a tool specifically for benchmarking that will sweep the arguments for you so you don't have to restart the server and send it prompts:

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

EDIT: Also the section about downloading the models should have mentioned that llama.cpp has a "-hf" argument that will download the models for you. I appreciate the author for sharing their experience, but for beginners this might not be the best guide to use.

> I appreciate the author for sharing their experience, but for beginners this might not be the best guide to use.

Yeah, I didn't write this as a proper developer guide. My screen recording started getting loads of favourites and I started getting messages asking about how I set it up, so just through up a quick rundown of how I setup this test.

I little just saw the Unclothe announcement about "Double the speed" and thought "Ha. I wonder if that will get it fast enough I'd actually be prepared to use it" and had a go at setting it up.

I'd done tests before last year with things like Devstral, but they were always both so slow and dumb, I didn't want to bother.

This finally hit the "wow, this is useable" level of both speed and intelligence.

I have used omlx.ai with great success to both download multiple mlx models (including gemma and qwen) suited for my hardware AND to be able to automagically launch both open-source and close-source (claude code, codex) harnesses using these models. All from a web or desktop UI

You would not need to follow a blog post with omlx IMHO

FYI you can open Claude code in the terminal, point it at this article and just tell it to "do it", if you're feeling extra lazy
Has anyone compared a setup like this to just using LM Studio?
(comment deleted)
or you can just load up ollama, have it load a local model and point claude or opencode at it...

is this article old? It's not. I'm not sure why he went through all the bother of llama.cpp

ollama is a wrapper on top of llama.cpp, and it makes llama.cpp slower, why use it?

Also Ollama has other issues (like forgetting what it really is - a wrapper).

Does anybody run a local agent on a Mac using an outboard GPU?
I poured a couple days into custom Burn inference for Qwen3-Coder-Next only to find it doesn't come with a speculative decoder, so on my M4 Max I can't push it much further than 120t/s. That's still kinda slow, though still faster than llama.cpp's 70.9t/s and MLX's 80.6t/s with the same model. Claude Fable 5 is recommending I use the Qwen3 MTP -- I worry that will compromise the quality somewhat, but might give it a try to see if I can get more usable speeds.
My biggest pet peeve with all these articles on local AI is the only thing they talk about is tokens per second. No one mentions the quality of the answers. No one. I don't mind waiting a little longer if the quality is better. Quickly serving me slop doesn't make it more useful. Are people really only looking at tokens per second?
The quality is obviously much worse, but still useful as a reference if you generally know what you are doing

It solve the "I'm coding on the plane and need to look up this thing I've forgotten" problem, for me at least

(comment deleted)
Nice writeup, thanks.

I run something very similar except for directly using pi as the agentic harness I use little-coder that wraps pi with reasonable defaults for running local models. Even though my local setup is a bit slow, it is a thrill to do real work completely locally.

I assumed lmstudio is the obvious choice after ollama. Is there a reason lmstudio is not used widely ?
Why would anyone use Ollama at all (aside from obvious reasons one can look up online) - llama.cpp used directly, without this wrapper is faster.

Basically one has two real choices for local LLMs: llama.cpp (if single user) or vLLM (if multi-user/enterprise).

You can also install Harbor and then it's:

harbor up omlx opencode

This video is realtime. And shows the agent responding at a perfectly usable speed.

Alas, this video appears not have been linked to the text that describes it. Perhaps I should ask an AI to generate an artistic rendering of the author's description.

I've been quite impressed with DeepSeek v4 Flash running via antirez's ds4[0].

It feels like a GPT-4 class model in terms of "stored knowledge" but is better at long-horizon tool calling than any of the GPT-4 class models.

Running on a 128GB MBP M4 Max, I'm getting ~24 t/s on generation and ~200 t/s on prefill. I was expecting it to feel slow, and it certainly does when e.g. generating code, but it's surprisingly useful as a "machine orchestrator" for simple tasks.

For non-agentic usecases, it's a decent enough model to converse with, and has the benefit of being entirely self-contained/private.

[0]https://github.com/antirez/ds4