[bigyabai]

We have to get real, here - most people are not replacing GPT or Claude with local inference, even on M5. If you can afford to do that (RAM shortage or not), then you are in the minority of customers.

Alleviating the memory constraint would only really make Nvidia a danger to cloud margins, and their consumer sales are neutered while they focus on the datacenter segment. It's feels facetious to insinuate that people would be doing inference on their Macbook Neo or Wintel laptop if they only had a gorbillion gigabytes of memory and a 400W accelerator card plugged into the wall outlet.

[kamranjon]

You’re out of the loop if you don’t think m series chips with unified memory aren’t one of the best platforms for running local inference

[bigyabai]

They aren't. Apple Silicon is unusable for interactive prefill and decode speeds in agentic workflows and SOTA LLMs.

[kamranjon]

You’re just out of the loop, and that’s fine but it’s worth learning about.

There is a pretty large and growing community of us using entirely local models for our agentic flows. From GLM 4.7 flash on 32gb machines with >60tok/s to Gemma and Qwen dense and MOE models on 64gb machines all the way up to Deepseek V4 flash on 128gb machines with 450tok/s prefill and 25-30tok/s decode.

I use DS4 on the daily - it’s become my main model.

I know it’s in fashion to talk trash about Apple but their hardware outperforms other options like DGX Sparc when it comes to local inference, they got the unified memory, memory bandwidth and the GPU cores to actually be useful in a way that most other hardware just isn’t.

[bigyabai]

> From GLM 4.7 flash

GLM 4.7 Flash is a 30b model that was far behind SOTA at launch, and I know that because I pay for z.ai inference and have run the model locally. Qwen and Deepseek V4 Flash have the same issue, and beg the question; are you really going to process a 64k agentic context at 450tok/s? That's 2+ minutes that you spend waiting for the first token to generate! Of course nobody can sell that as competitive inference, and it only gets worse with larger models. We're talking about non-interactive speeds, here.

If you're satisfied with small local models, more power to you. It puts you in the same barrel as Strix Halo enthusiasts or the guys that bought 2x3090s on Reddit. You are completely ignoring the market if you think that any of those SOCs are unprecedented or unparalleled for inference workloads, though. The free DS4 API is faster at prefill and decode, you could not give away Mac inference at zero cost and compete with what China provides for free. That's how far behind Macs are for local inference, to put things into perspective.

[kamranjon]

I think you’re confused - nobody running local models is concerned about SOTA - that’s just marketing hype from large providers - we are interested in data governance, security, control and freedom. You can’t compare hosted services to local inference, these are two very different things, out of principle we aren’t interested in handing our code bases over to untrustworthy third parties.

On your first point, nobody is pasting 64k tokens at once as context, if you are you’ll experience very similar wait times even with hosted providers - context is built up piecemeal and by virtue of being context does not need to be replaced constantly - this is how all agents work.

100% local models are not SOTA, but they are good enough to be incredibly useful if you are a skilled engineer, and I understand the industry is pushing engineers to offload more and more of their work to incentivize higher token spending, but talented engineers can absolutely be just as productive using local models today - it’s just a different style of working that folks who have become accustomed to large providers can’t really comprehend at this point. They’ve vendor-locked themselves into a delusion that they absolutely need SOTA for everything and as a result see everything as black and white.

[Danox]

You sound like IBM in the mainframe era...

[aroman]

My hardware isn't powerful enough to try, so I'm asking out of genuine curiosity, not to push back: what do you use DS4 for? Did it replace e.g Claude Code with Opus for you, or was it replacing something else?

[kamranjon]

I use it as my main coding agent - so its running DS4 server on my 128gb mbp and I run the pi coding agent on my other machine which calls out to it. Mostly Go and Typescript work.

I also use it in local agent mode if im coding directly on the machine which is nice cause you can save sessions and resume them, and so for personal projects and training related stuff it's been great.

Even got an autoresearch loop going where the agent looks at the previous run, adjusts parameters and code if needed, and then hands off training to another script (so full system resources are available for training), ad infinitum - it works really well - what antirez has done with that project is pretty incredible.

[johncalvinyoung]

Isn't Deepseek V4 Flash still like 150+ GB even at Q4?

[kamranjon]

I think most people running deepseek locally are using DS4: https://github.com/antirez/ds4

Which provides a 2-bit quant and a mixed 2-bit/4-bit quant - which range from 80gb to 97gb

Deepseek is particularly well suited for quantization and the quality of the 2bit quants antirez has provided have been validated by folks like ggernov of llama.cpp project