> 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:
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.
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:
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
jumploopstoday at 12:09 AM
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.
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.
reddit_cloneyesterday at 6:59 PM
>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.
mark_l_watsonyesterday at 9:30 PM
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.
jmkniyesterday at 7:17 PM
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
anigbrowlyesterday at 11:58 PM
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.
smetanniktoday at 12:21 AM
I wonder why something like LM Studio didn't work for the author?
One way or another local AI is the future. I actually find weaker models more interesting because it keeps me sharp (at the cost of velocity of course).
reenorapyesterday at 9:07 PM
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?
bicepjaiyesterday at 10:21 PM
I assumed lmstudio is the obvious choice after ollama. Is there a reason lmstudio is not used widely ?
deletedyesterday at 9:13 PM
everlieryesterday at 10:36 PM
You can also install Harbor and then it's:
harbor up omlx opencode
cdolanyesterday at 6:14 PM
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
rectangyesterday at 7:58 PM
Does anybody run a local agent on a Mac using an outboard GPU?
attogramyesterday at 7:02 PM
8b max on a std 16gb macbook. Anything more and your mac is toast
metadaemonyesterday at 7:21 PM
Has anyone compared a setup like this to just using LM Studio?
namnnumbryesterday at 6:36 PM
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.
LoganDarkyesterday at 8:02 PM
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.
sleepybrettyesterday at 7:32 PM
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