It’s hot and also LOUD and runs the battery down quick.
But I’m having a lot of luck just running things when I’m away from the computer and can leave it plugged in.
It starts going weird (unreliable and slow) with context over 80k so you have to pick tasks one at a time and baby sit a lot more than Claude. But it really is very capable and feels like there’s an intelligence there to talk to. Maybe gpt-4 level clever?
I have an m5 max 64gb and I think anything slower would be quite painful.
I have an m3 max 64GB and it runs ok, I was playing around with it last night. Its reasoning is better, but the token/s still makes an MoE model (35B A3B) preferable so I switched back.
Try using Goose instead of Claude's harness? Goose doesn't load as much context in at the start, so it might be more usable. I've definitely been able to get it past 128K, although I typically don't go beyond 70-80k. MoE might also be better at this.
There was a lovely window of a few years when processors were fast enough and low-power enough that real development work could trivially happen on a Macbook Air in a lounge.
That window is waning as more and more memory and graphics processing power is being used locally.
I suppose the future is most likely going to involve farming out AI requests to your desktop machine, your company's compute farm, or a cloud endpoint, but if you're expecting to run an AI offline on your lap with a battery...your lap will get warm.
No idea what you are talking about. My battery lasts longer than ever while running vim and make and GCC. It’s amazing. Not sure why your windows are closing.
Because the local LLM, which you are not running, is running for much longer than gcc and is eating the battery. Different choices, different outcomes.
Bur the implication here in the parent comment is that for ‘real dev work’ to be done you must be running an LLM, and therefore you can’t do it on a MacBook anymore.
Turns out you can still accomplish stuff with a text editor and compiler.
Of course we can code without a LLM, but if we want to, as the sibling comment says, don't run a local LLM on a laptop. I'm offloading the LLM to the LLM vendor, like most of us IMO. The CPU is still busier then it used to be, because of the local harness.
> There was a lovely window of a few years when processors were fast enough and low-power enough that real development work could trivially happen on a Macbook Air in a lounge.
I was responding to this. I am appalled that anyone thinks (and is willing to say out loud in public) that they cannot do "real dev work" without an LLM.
Merely existing noticeably ages most battery chemistries.
You can take half an hour out of every week inconveniencing yourself to protect your battery, or you can spend an hour once a year just putting in a new one (or having it done for you in-store if you're using an Apple device) and save yourself the nagging worry.
If you're running it while idle and don't need the quickest results, reducing the clock speed improves energy efficiency (and in your case avoids overheating the battery). There will be some optimal speed that maximizes computations per joule that depends on the specific load and can only be found by measurement.
On Linux, you can cap CPU frequencies with "cpupower". Does MacOS have any equivalent?
My M5 Max has 3 modes, "low", "automatic", and "high". Automatic doesn't seem to simply switch between low and high -- it seems to sit in the middle and vary dynamically based on workload and temperature. With automatic I get more than half as many tokens per second as on high, but with much lower fan speed and temperature.
Macs fan control set to full blast and lifting the Mac above table for air circulation across bottom.
I use M1 Max with qwen3.8 27b mlx. The gpu temp can reach easily to 95°C as fan doesn’t kick in automatically until 90°C. With Macs fan control at full blast and MBP off table, temp usually hover around 85°C.
RAM 32GB, tokens are less than 20, may be about 12-15. Main issue is context window, I have set to 64k but practically beyond 32k, too much swap and slow down.
Yikes isn’t 95 well above the danger zone for most components or have things changed? I remember crypto mining in the early 2010’s and making sure to stay under 85C on my GPU.
Stuff runs so hot these days. I think one of the parameters pushed to keep Moores law from completely flattening was heat. My desktop regularly hits 95 just compiling code.
If you're going to try to use a mac laptop for long term inference you'll probably want to buy one of those $30 laptop fan cooling pads (sold for absurd sized 17 inch x86-64 gaming laptops) on Amazon. Otherwise you'll roast the battery over time and significantly hurt its cycle life and capacity. If you can get significant airflow under the laptop that'll help a lot.
I tried it on my M1 MacBook Pro. It's slow but surprisingly smart as a general purpose LLM. Maybe GPT-5.3 level. I gave it a bunch of tools and it can search the internet, make product recommendations, document, code, etc.
Awesome, I used Claude to write a small python script to do the same with Linode's API. The only difference is I setup a persistent drive, and with Linode you can boot off of it. So my biggest start up lag is ~ 2 minutes to deploy + boot, then maybe 2 more to warm the model.
I actually dislike LLMs. But I'm a realist, and on-demand compute like this is massive cost saving measure.
What GPU you end up with for that price? Vast.ai (https://cloud.vast.ai/?priceInstanceHourlyMax=2) has a bunch of setups available to reach 192GB VRAM under $2 :) Quick skim showed 4x48, 2x96 and 8x24, all for under 2 buckaroos or around there.
interested in this too. i suspect ppl dropping generic "its awesome" comments are not actually using in just managed to get it running for a prompt or two.
Yeah, that's my experience. It's a big "wow" factor to get a non-trivial LLM running on my Mac, but it's actually not that useful. Like trying to use Photoshop at 8 FPS.
Regarding this analogy, fps don't matter as much for Photoshop, since it's not an immediate mode GUI. 8 fps would be quite ok for comfortably getting feedback on live image filters and such.
It’s not because you didn’t find use cases for local LLMs that there are none.
I use local LLMs on my Mac Mini M4 Pro with 48G to review text messages tone, act as a text correction tool, act as a code review tool, to do code agent work, generate code snippets, etc
I feel like it's 50/50 between people doing that, and people that have spent a lot of time tuning a system they are pointing at focused and well specified problems.
Seems threads about local LLMs on Apple hardware feature comments listing M3/4/5 at 48GB 64GB and not 128GB.
That is, users with M-series hardware that have less-than-max RAM share results whereas users with max RAM do not.
Speculating (not extrapolating), maybe users with machine that have max RAM are less interested in running local LLMs and are less averse to paying services for compute?
Personally, I’d love to see what output max RAM M-series Apple hardware in these threads.
Qwen3.8-27B runs at 59.5 tok/s on my M4 Max, 40-core GPU, 128 GB
I use it occasionally for classification and other tasks but I wouldn't trust those smaller models with the real work and for larger data processing it's too slow, e.g. a dataset I wanted to classify would've taken 56 days on my laptop vs just paying the cheap Luna prices to openai and getting it done in a few hours.
I've been using Qwen3.6-37B-A3B on an M1 Max w/ llama.cpp and for my practical uses I prefer it to qwen3.8. When 3.8 does answer its slower and, qualitatively, marginally better than qwen3.6, but 3.8 often ends up in unresolved thought loops and runs slower. The Moe 3.6 on my setup is much faster, 500t/s peaks, 30t/s typical, vs 3.8 150 peak, 4-9 t/s typical.
While I've spend a little time tuning, I'm assuming there will be deeper tuning for 3.8 that might close the gap.
My M4 Pro 48GB gets about 13tok/s, in both 3.6 and 3.8 27b Qwens. Qwen A3B and Gemma get closer to 100tok/s from memory but the results are pretty poor for coding tasks.
Edited to add: for agentic workflow I’m running omlx which tells me it has about a 90% cache hit rate (tradeoff is some disk and mem space) - that noticeably changes the felt speed.
I got the qwen 3.8 abliterated model running on my MacBook Pro M5 48GB and it's pretty nice having a local model that can do a lot of experimentation without rails.
It was actually great. I have like a non-AI box so to speak 8GB VRAM, co-incidentally from a gaming PC ...
All the previous models that were "frontier level, just try it!" but wouldn't run at all in agentic mode, including previous Qwens, just disappointed, period.
Then I ran then Qwen 3.8 27b and while it was super slow (4t/s) it literally one-shotted creating a usable "web search/pull" skill for `pi.dev`. while any other model previously just entirely failed to create anything usable even with actual guidance.
Since then I have actually gotten a gemma-4 12B qat 4bit quantized with a ~250MB MTP from unsloth to work with a 32k context "working" on this setup at 80-120 t/s. That's usable for private stuff on a co-incidental box!
It's still only 32k context and it's entirely dumb vs. our API paid at-work Claude Opus. But for entirely private local stuff it's totally workable without breaking the bank even after all these AI price hikes!. I bought this rig literally just for gaming a month ago.
Have you tried a mixture of experts model? Dense models have been quite slow for me, as I have only 6 GB VRAM. But with llama.cpp and --cpu-moe I get 200 t/s input and almost 30 t/s output with Gemma 4 26B A3B, which feels ok to use. Would be interested about your mileage there.
If I use the 12B Unified (dense) model I mentioned without MTP, then I get 37t/s, input ~700t/s.
It's all still quite frustrating in the end, like a Claude from a very long time ago by now but usable. If I want 64k context, I can't use MTP. I still haven't decided whether I'd rather have 37t/s but it's "less dumb" or I want MTP speed but it's going off the rails more. All of this is also with `-ctv q4_0 -ctk q4_)`, which is not ideal. I'm actually right now contending with 35k context but using q8_0 KV quantization. Still the 37t/s coz with those settings I can't use MTP.
But I'm not ready to go back to 4t/s. It's not interactive enough for me. That said, I had tried to use the Gemma E4B for example to have it build itself that websearch/fetch skill. It utterly failed, as did previous qwens.
I don't see a Gemma 4 26B A3B GGUF for download, but there is a gemma-4-26B-A4B-it-MXFP4_MOE.gguf that should fit into my overall RAM and then use lots of CPU like the Qwen 3.8. I guess I'll give it a try just to see the difference in speed though I don't expect anything "usable" out of that tbh.
Haha, OK, so it processes input tokens at 55t/s if started like the 12B. That's so slow that I don't even have the patience to wait for the context of the session I aborted mid-processing to test this, to load. Heck loading the system prompt alone is not fun.
That said, after that it seems to run at the same ~35t/s speed but, I was able to start it up with a 128k context (and no KV quantization), humm humm. I am now playing more with all the parameters like I did to get the 12B going e.g. I'm already up to ~125t/s for inputs now and with an 8bit MTP it's running about 45-85t/s output when its coding.
I run with `--spec-type draft-mtp --spec-draft-p-min 0.8 --spec-draft-n-max 8` and when it was that fast (during actual code generation) got some `draft acceptance = 0.91801 ( 3482 accepted / 3793 generated), mean len = 7.92`. I also tried 16 with no real improvement for those phases but it seemed slower, wasting too much with low 50% acceptances during other phases like thinking.
EDIT: Oh look, I can now use larger batch sizes again and that gets prompt_eval reads back to 700+
Regarding how good it is: Probably, maybe, better than the 12B? It immediately noticed an error the 12B made, where it hadn't followed a skill's format description. But it's also annoyingly not following instructions the other one was and instead just gives me its opinion lol! And it couldn't figure out how to read a simple pi SKILL.md and run an included script. It complained that it wasn't registered as a tool. But like all other model changes, will get used to it and adjust I suppose :)
Thank you, very interesting to read. I haven't considered batch sizes at all up to now. Seems I should look into it.
How did you get the prediction that good? The default spec-draft-max-n is 3 and if I go beyond, the output rate stays the same or gets lower any magic spell you could share?
Regarding the MTP, if you haven't found out why yet, the `--spec-draft-p-min 0.8`. Something to play with and of course it's non-deterministic and depends on both the model itself and whatever you're actually throwing at it.
I'm having the same constraint, 8GB VRAM, and was trying the past months to get a local model running to drive my Home Assistant setup via voice, so many tool calls, little room for errors. I've tried multiple models, different temperature and top p values, different context windows... turns out that a quantized gemma4 is the best pick so far but qwen3.5:9b isn't far off. Surprisingly newer models are not necessarily performing better.
That's funny, I downloaded the same model on my 48GB M4 Pro and gave it a problem to solve in an existing codebase, it spun its wheels for twenty minutes and then fell over dead. This was using LMStudio and pi as a harness; I never use pi for anything else, so maybe I'm holding it wrong.
We don’t know what quantization level was used for the weights or the kv cache for you or for parent poster, so this is probably an apples to oranges comparison.
They made a kind of strange decision with Qwen3.8 27B, the template defaults the reasoning_effort to xhigh. I found if you set it to medium it doesn’t just sit there churning forever.
I have seen xhigh go down several rabbit holes, dwell on edge cases and write worse code as a result; it literally distracted itself into writing a complex chain of functions ignoring my prompt, when on “low” reasoning it gets it right on a prompt that requires a few lines of code in the right places.
Simon Willison’s blog has another example (SVG of a circle).
It’s a bit like how giving LLMs access to web search tools can cause them to go down a blind alley based on their first “reasoning” output that then leaves them unable to solve a puzzle correctly that they can fully solve on their own.
With qwen 27b, setting the right reasoning effort for the specific task is important. With xhigh it has a chance at hard problems that bigger models may even fail. But for many everyday tasks, I have found that no reasoning and a system prompt instructing it to be brief is good enough. Note that even with thinking disabled, it may still get into long "chain of thought" reasoning state (out of thinking blocks) if the task is hard and you do not give further instructions, esp with access to tools etc.
Was there a more recent refresh or is this the model from a year ago? The frontier models were barely functional and almost useless a year ago (gpt oss was pre opus 4.5!) - I would be very surprised if the original drop is anything more than totally obsolete/irrelevant at this point
I find that I remember models being a lot better than they were, even when I remember them being not very good - because of a novelty factor ("whoa it can do that now?") mostly. And then I go back and look at them and its like, what how did I find this impressive.
A funny example - I remember thinking "yeah sonnet 3.5 is a really good coding model"
Which exact model are you running? With only 48GB of RAM, by the time I got a model small enough, it was pretty bad in performance both in speed and reasoning.
I've been comparing against TextGen and llama.cpp while I port to LocalAI and have been surprised by what's happening over the API, even with the defaults and jinja. It's been a fair reminder not to eschew familiarizing myself with the repos.
Nothing beats the classic of figuring out something yourself with your brain, but I also like dictating to LLMs a stream of consciousness with what I'm interested in (while forcing it to NOT give any answers or opinions), and getting back file names it suggests I look at and explore.
Modern frontier LLMs can still be used as rubber ducks, and it's a great.
From what I've heard, Ollama has a bad reputation because it's a thin wrapper around llama.cpp without attributing it properly, thereby stealing recognition from the maintainers doing most of the work
It seems, and that seems is entirely my unvalidated impression, that Ollama lags in features, as they're integrating after the fact those changes. But (seriously) an LLM told me that, when some aspects of MoE models were better supported with the latest llama.
And it did in that case make a significant difference.
It's very far behind llama.cpp, vLLM and SGLang in features yes. In part because of that but also due to some poor default settings it generally performs a lot worse as well.
People who use Ollama generally (not everyone obviously) don't always clearly understand what quantization they use when running models, so people end up saying "I tried running Qwen 3.8 27b locally and it was dumb" while Ollama would default to a Q4 version of the model, which has very different results from the BF16 weights, doesn't really speak to the model itself because it's been so quantized in that case.
Sure, makes things easier, but tons of people misunderstand what they're using, then base and share their experiences on that, without really specifying what exact weights they use too.
For a single local user, using llama.cpp directly shouldn't be a problem if you're already using Ollama's CLI, it works basically the same except you manage weights yourself, and if you put your favorite agent to make sense of the faux "registry + image layers" Ollama has prepared locally for you, you can reuse the files you've already downloaded with Ollama.
In my limited testing of ollama (in a Docker container because I don't want Meta touching bare metal disk), I noticed that first of all the model is abandoned. Last updated in 2023. Secondly, it basically can't do anything. Anything I asked of it resulted in a lot of "I can't do that" so it's just stupid.
Are you confusing llama (the model from meta that popularised a lot of open AI stuff but is quite dated now) and ollama (the set of tools for running models, originally llama but no longer limited to it, not maintained by meta)
1) Shipping with 2k default context window for the longest time, w/o any warning and being not easy to change (like any other setting). Totally made a lot of people think local LLMs are dumb as rocks. Just checked, still not fixed -- defaults to 4K if less than 24GB VRAM: https://docs.ollama.com/context-length
2) Registry mistrust due to Deepseek R1 naming. What model do you download/start with `ollama run deepseek-r1`? Not Deepseek R1, but this "for research purposes" thinking finetune of Llama 3 released alongside R1 paper: https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama... . For actual R1 you have to pull `deepseek-r1:671b`
3) Can't reuse existing ggufs you have. If you point it to gguf, it would create a copy of it in ollama folder.
4) Doesn't accept engine parameters from CLI args. Only env variables or modelfile. But some things are only in env variables (e.g. KV cache quantization). And even then not even close to what's available in llama.cpp.
5) Often meaningfully slower than llama.cpp
6) VC-funded wrapper for llama.cpp. For long time had questionable attribution to it. Now looks good.
> (and also because VLLM is not super Windows friendly)
llama.cpp is Windows friendly and fairly easy. Not GUI app with installer, but easy. Zip with binaries are on github, run
llama-server.exe -hf unsloth/gemma-4-12B-it-qat-GGUF:UD-Q4_K_XL
will pull the model you expect from hugging face (or -m for manually downloaded .gguf) and start web server with API and HTML chat app.
Thank you for your responses. They've been very helpful.
I'm going to test the llama.cpp binaries directly. That seems to me to be the closest to what Ollama is trying to achieve, after all.
I’m running Qwen3.8 aggressive uncensored Q4_K_M on a 4090 in a loop against the 2026 CrackMe CTF challenges.
Using oh-my-pi in a prebuilt environment that I let Qwen build too.
Codex wouldn’t even look at the files - literally, as soon as it read something with CTF it shut down. Didn’t even offer to fall back to a dumber model.
This is similar to my use case as well. I’ll use a cloud model to identify security issues in a codebase and then use a report/spec output as input to the local model to build tests or verify the issue is real.
b) Don't run quantizations of the LLM that are worse than the best available Q8 (the largest possible file size unsloth GGUF for a given model like qwen 3.8 27B as an example). I would rather things go slowly but I have confidence that it's doing things more accurately.
Even a 4-bit quant of Qwen3.8 27b is indistinguishable from Gemini 3.7 flash in our internal tests. With an RTX5090 card and ninfer, you can get ~800 TPS token generation (c=8) and ~140 Tokens per second single stream.
Local models running on your GTX 5090 or Macbook M99 are never going to reach the levels of the currently hosted "frontier models" with hundreds/thousands of CPU's and GPU's. Get a grip.
Anybody who's tested this over the last 12-18 months knows this. Don't waste our time because you're late to the party.
The section on system prompts and context window management is spot on; most people don't realize how much the default quantization in popular runners degrades logic compared to full FP16. I'd be curious to see if the author has benchmarked the impact of KV cache compression on longer context reasoning, as that usually seems to be where my local Llama 3 setup starts to fall apart.
> Both the NVFP4 and AWQ W4A16 failed to properly close their tool calls ...
If I understand correctly, this failure mode is just not possible with llama.cpp / ik_llama.cpp, which enforces token generation to follow the grammar once a tool call is detected.
> ... and botched Cisco command line syntax (the correct command was ‘show arp’, while they executed ‘show run’)
But this failure mode can still happen.
Anyway, NVFP4 and AWQ W4A16 are generally regarded as low quality quants. IQK/Trellis quants from ik_llama.cpp and EXL3 quants from exllama should work better.
So, perhaps the lesson here is "don't use vllm at home"?
No favorite in single spark usage really. Glimmer is interesting because it scales well under load. Qwen3.6 MoE was fast enough on low concurrency. Qwen3.8 works but slower than the DS flash on the two sparks.
I saw some guy streaming how he was deploying qwen3.8 37B on his local setup. Well, he was asking Claude to do it. It took him two hours of passing errors to Claude for the endpoint to start working, he then started testing it against DS4 Flash when Qwen had thinking disabled and Claude messed up sampling parameters, it was an absolute pain to watch
Your mileage may vary. I tried this a couple months ago and spent a full day on it just not working before giving up. Anything I sent, it wouldn't run.
My experience with Claude is that it suffers badly from “not invented here” syndrome. So probably it rebuilt something like llama from scratch and then 2 hours suddenly seems reasonable (if you don’t question the approach). And that’s the thing, someone with no experience isn’t going to question it.
Not necessarily rebuild llama from scratch, build attempting to build it without cmake and manually invoking all the build commands would be quite in character.
I wonder if this is an artifact of RL, where the training heavily emphasizes codegen. It may be that the model is just better at generating code than reusing libraries, so it prefers the lowest cost approach.
I also wonder if this manifests much less in contexts where the libraries/frameworks are a large part of the training set. It may be that the model doesn’t generalize well so it’s always better to use knowledge in its training set vs attempting to understand how to use a new, potentially never before seen (from the model perspective) api
I'm not saying this is the case in the particular example, but these days a lot of people aren't bothering to read a 1 page README, and are instead letting claude "just fucking do it already". I see it at work.
Claude has got better at "just fucking doing it" by asking if it's ok to go read the latest github issues and pull the README, which means that people will likely get lazier and lazier.
My ability to detect sarcasm is not good. From looking at funlang's profile and other comments the profile looks like a LLM generated bot.
Forums with full no verification pseudonyms seem like they have a real challenge ahead. How long until we need humanhackernews.com with public pseudonyms and a private trusted verification?
To me laziness has got nothing to do with it. I'd rather focus on my limited time on things that actually matter, e.g. the design of an important product feature or the root cause of a bug and the proper fix. I delegate everything else to Claude. It's really nice to be able to say to Fable "see if you can get this running locally" and then come back 10-30 mins later and read the results.
I've discovered a lot of neat tooling this way that I otherwise would not have bothered tryin to set up because that can take up a lot of time. And even when it's fast, you're suffering from context-switching penalties. You framed it as "read the latest github issues and pull the README" but anyone who has worked in tech for an appreciable amount of time knows that that stuff can turn into a deep rabbit hole.
In the README there is usually a section where it explains which llama.cpp parameters to configure (sampling, penalties, context size, etc) to get better results.
Wow. I tried to get Qwen3.8 4B to parse song lyrics and analyse them. Getting ollama running was a minute or two.
However coming up with a prompt that didn't turn out total garbage was impossible. After wasting over an hour and I ended up getting Qwen side by side with Llama 3.2 3B, just to see if I was being stupid. Nope, it just looks like Llama is orders of magnitude better at this specific task for some reason).
If you think I'm doing it wrong, you're probably right, I don't know a ton about local LLMs. But I hand selected 50 songs, set up ollama with both LLMs, and for each iteration on the prompt text, ran both LLMs 10x times per song. Side-by-side comparisons showed that Qwen 3 4B was so bad that I actually downloaded Qwen again, thinking there must have been some mistake and I accidentally grabbed an old 1B model.
This is the way. It's also very important to automate as much of this verification as possible into the harness, rather than sit there and prod it in the chat.
Of course some things are not auto verifiable, and you'll have to give human judgement and input there, but you'll save a lot more time if you spend 1 week painstakingly writing checks for as many little things as possible and integrating them into the harness.
> What the big AI labs have over the smaller labs, is a huge amount of data and diverse set of tasks, hence they generalize better, but still not great.
Hehe, this kind of sounds like the opposite of generalization. As in it’s just specialization at scale.
Tbh the way you're treated depends on the hotness of the task.
A couple of years ago, people I know got paid OK for relatively simple programming and logic RLHF tasks. But very soon it turned dark because that sort of data was required less and less, and the number of feedbackers has grown.
Today, the type of data the model developers pay for requires actual domain experience. E.g in software engineering they have people work in simulated environments with other LLMs, grade them, feedback, PRs, Jira everything.
This pays well and they treat you well, entice you with more money/task/hour, etc, for now. In few years when this gets drilled into LLMs, these guys will face the same painful hours and bad pay and less work and so on too.
In physical tasks, we are still in the early stages where basic packing clothes (in a textile factory setting) etc is being recorded and data is only now being used for training. Due to the problems with translating human hand data to robotic hands, these people do the factory work holding robotic grippers and operating that, you should see a YouTube video. But this also means that it's much less sweeping than data collection in SWE. In many cases it's not practical to collect data given that you have to use the specific gripper, wear a big gopro type thing, etc. So I am expecting much slower of an impact on physical tasks (of this kind) compared to how quick the uptake was in SWE/math.
I have yet to hear back from them on how it's going for teamwork white collar tasks, it's been a few months. Everyone is paying for the end products of those it seems - grok bot, perplexity computer, claure cowork, chatgpt work etc,. Not as much as their coding agents of course.
That's funny, because I just went through the opposite.
I got llama.cpp working with qwen3.6 and qwen3.8 by Googling and manually adjusting things according to reddit posts and Google not-really-helpful AI suggestions.
I tried settings up per-model stuff in settings.json, but again Google got in my way, and llama.cpp having 2 different settings.json (and Google lying about where 1 goes) made it far too difficulty to figure out. I spent hours on it.
Then I got fed up and asked Claude.
Immediately, it told me that the winget version of llama.cpp is for Vulkan, and I needed a different one and pointed at it. It doubled my speed.
Then it figured out what I was doing wrong with settings.json (wrong spot, global settings can't go in the per-model file, etc etc) and fixed all that, and got it working.
Then it tuned it somewhat.
Then I showed it the official settings pages for both models, and it undid the tuning and all the damage I had done with my tinkering, and got everything working.
In 30 minutes.
It was absolutely amazing.
Every time I see people recommending Qwen locally with llama.cpp, they just say "download it" and act like anyone that can't get it running is an idiot. But if there's a "using this settings.json" tutorial somewhere, I didn't find it, and neither did Google over a week of searching.
But Claude got it done for me.
Now, I admit, I haven't played with it much. Just before all this, I ran out of Claude on the $20 plan and bumped up to $100, and It's been so amazing that it's really hard to work on the local. Especially since it feels like Qwen3.8 35b a3b is probably around the corner, and why mess with 3.6 when 3.8 will probably release soon?
Rate limiting on free LLM APIs is usually where the pain lies. I've seen 5 concurrent reqs hit 20K/day limit in under 2 hours. Does anyone know a free API that still allows some reasonable concurrent requests?
215 comments
[ 0.23 ms ] story [ 38.4 ms ] threadBut I’m having a lot of luck just running things when I’m away from the computer and can leave it plugged in.
It starts going weird (unreliable and slow) with context over 80k so you have to pick tasks one at a time and baby sit a lot more than Claude. But it really is very capable and feels like there’s an intelligence there to talk to. Maybe gpt-4 level clever?
I have an m5 max 64gb and I think anything slower would be quite painful.
Try using Goose instead of Claude's harness? Goose doesn't load as much context in at the start, so it might be more usable. I've definitely been able to get it past 128K, although I typically don't go beyond 70-80k. MoE might also be better at this.
Who'd run this on battery?
Or do you mean kills the battery faster even when used while charging?
That window is waning as more and more memory and graphics processing power is being used locally.
I suppose the future is most likely going to involve farming out AI requests to your desktop machine, your company's compute farm, or a cloud endpoint, but if you're expecting to run an AI offline on your lap with a battery...your lap will get warm.
Turns out you can still accomplish stuff with a text editor and compiler.
I was responding to this. I am appalled that anyone thinks (and is willing to say out loud in public) that they cannot do "real dev work" without an LLM.
You can take half an hour out of every week inconveniencing yourself to protect your battery, or you can spend an hour once a year just putting in a new one (or having it done for you in-store if you're using an Apple device) and save yourself the nagging worry.
On Linux, you can cap CPU frequencies with "cpupower". Does MacOS have any equivalent?
I use M1 Max with qwen3.8 27b mlx. The gpu temp can reach easily to 95°C as fan doesn’t kick in automatically until 90°C. With Macs fan control at full blast and MBP off table, temp usually hover around 85°C.
- find a free GPU droplet on digital ocean
- fire it up
- pull in a snapshot of the model + extra files/packages etc
- set up a ssh tunnel so that the localhost:8000 routes to the above
Then I just configured OpenCode to use the above and was off to the races.
Works out to be about ~$2/hr all said and done which isn't bad as I only pay when I'm using it (but could get expensive with 24/7 running)
I actually dislike LLMs. But I'm a realist, and on-demand compute like this is massive cost saving measure.
Going to try this out vs the snapshot!
I also really like this experiment b/c it's a mix of LLMs and old school IaaC/DevOps.
What GPU you end up with for that price? Vast.ai (https://cloud.vast.ai/?priceInstanceHourlyMax=2) has a bunch of setups available to reach 192GB VRAM under $2 :) Quick skim showed 4x48, 2x96 and 8x24, all for under 2 buckaroos or around there.
And yeah, did a lot of work with Vast AI at a past job and it's pretty wild the variety of prices/hardware that they have.
That's cool, what actual GPU though? I'm still curious :P
I use local LLMs on my Mac Mini M4 Pro with 48G to review text messages tone, act as a text correction tool, act as a code review tool, to do code agent work, generate code snippets, etc
Gemma 4 26B A4B gives me steady 20 tps.
That is, users with M-series hardware that have less-than-max RAM share results whereas users with max RAM do not.
Speculating (not extrapolating), maybe users with machine that have max RAM are less interested in running local LLMs and are less averse to paying services for compute?
Personally, I’d love to see what output max RAM M-series Apple hardware in these threads.
I use it occasionally for classification and other tasks but I wouldn't trust those smaller models with the real work and for larger data processing it's too slow, e.g. a dataset I wanted to classify would've taken 56 days on my laptop vs just paying the cheap Luna prices to openai and getting it done in a few hours.
While I've spend a little time tuning, I'm assuming there will be deeper tuning for 3.8 that might close the gap.
Edited to add: for agentic workflow I’m running omlx which tells me it has about a 90% cache hit rate (tradeoff is some disk and mem space) - that noticeably changes the felt speed.
All the previous models that were "frontier level, just try it!" but wouldn't run at all in agentic mode, including previous Qwens, just disappointed, period.
Then I ran then Qwen 3.8 27b and while it was super slow (4t/s) it literally one-shotted creating a usable "web search/pull" skill for `pi.dev`. while any other model previously just entirely failed to create anything usable even with actual guidance.
Since then I have actually gotten a gemma-4 12B qat 4bit quantized with a ~250MB MTP from unsloth to work with a 32k context "working" on this setup at 80-120 t/s. That's usable for private stuff on a co-incidental box!
It's still only 32k context and it's entirely dumb vs. our API paid at-work Claude Opus. But for entirely private local stuff it's totally workable without breaking the bank even after all these AI price hikes!. I bought this rig literally just for gaming a month ago.
It's all still quite frustrating in the end, like a Claude from a very long time ago by now but usable. If I want 64k context, I can't use MTP. I still haven't decided whether I'd rather have 37t/s but it's "less dumb" or I want MTP speed but it's going off the rails more. All of this is also with `-ctv q4_0 -ctk q4_)`, which is not ideal. I'm actually right now contending with 35k context but using q8_0 KV quantization. Still the 37t/s coz with those settings I can't use MTP.
But I'm not ready to go back to 4t/s. It's not interactive enough for me. That said, I had tried to use the Gemma E4B for example to have it build itself that websearch/fetch skill. It utterly failed, as did previous qwens.
I don't see a Gemma 4 26B A3B GGUF for download, but there is a gemma-4-26B-A4B-it-MXFP4_MOE.gguf that should fit into my overall RAM and then use lots of CPU like the Qwen 3.8. I guess I'll give it a try just to see the difference in speed though I don't expect anything "usable" out of that tbh.
That said, after that it seems to run at the same ~35t/s speed but, I was able to start it up with a 128k context (and no KV quantization), humm humm. I am now playing more with all the parameters like I did to get the 12B going e.g. I'm already up to ~125t/s for inputs now and with an 8bit MTP it's running about 45-85t/s output when its coding.
I run with `--spec-type draft-mtp --spec-draft-p-min 0.8 --spec-draft-n-max 8` and when it was that fast (during actual code generation) got some `draft acceptance = 0.91801 ( 3482 accepted / 3793 generated), mean len = 7.92`. I also tried 16 with no real improvement for those phases but it seemed slower, wasting too much with low 50% acceptances during other phases like thinking.
Regarding how good it is: Probably, maybe, better than the 12B? It immediately noticed an error the 12B made, where it hadn't followed a skill's format description. But it's also annoyingly not following instructions the other one was and instead just gives me its opinion lol! And it couldn't figure out how to read a simple pi SKILL.md and run an included script. It complained that it wasn't registered as a tool. But like all other model changes, will get used to it and adjust I suppose :)How did you get the prediction that good? The default spec-draft-max-n is 3 and if I go beyond, the output rate stays the same or gets lower any magic spell you could share?
This is what I read before I went to try it: https://github.com/ggml-org/llama.cpp/discussions/25198
I have seen xhigh go down several rabbit holes, dwell on edge cases and write worse code as a result; it literally distracted itself into writing a complex chain of functions ignoring my prompt, when on “low” reasoning it gets it right on a prompt that requires a few lines of code in the right places.
Simon Willison’s blog has another example (SVG of a circle).
It’s a bit like how giving LLMs access to web search tools can cause them to go down a blind alley based on their first “reasoning” output that then leaves them unable to solve a puzzle correctly that they can fully solve on their own.
A funny example - I remember thinking "yeah sonnet 3.5 is a really good coding model"
https://stack.convex.dev/using-cursor-claude-and-convex-to-b...
>Prompting Cursor to Scaffold my App: FAIL This was my first hurdle.
>It became immediately apparent that I would not be able to prompt my way through the entire process.
>While the tooling we have is undeniably powerful, it's not yet capable of completing most nontrivial tasks
It couldn't run pnpm install lmao. Opus 4.5 was a crazy jump
Modern frontier LLMs can still be used as rubber ducks, and it's a great.
Genuine question : is there something fundamentally wrong with Ollama ?
I use Ollama because it is easy to set up and manage (and also because VLLM is not super Windows friendly).
I thought the main advantage of VLLM was better concurrency management (better batching).
But if the quality of the interference itself is an issue, then maybe I should reconsider my choice.
And it did in that case make a significant difference.
Sure, makes things easier, but tons of people misunderstand what they're using, then base and share their experiences on that, without really specifying what exact weights they use too.
For a single local user, using llama.cpp directly shouldn't be a problem if you're already using Ollama's CLI, it works basically the same except you manage weights yourself, and if you put your favorite agent to make sense of the faux "registry + image layers" Ollama has prepared locally for you, you can reuse the files you've already downloaded with Ollama.
Yes https://news.ycombinator.com/item?id=47788385
1) Shipping with 2k default context window for the longest time, w/o any warning and being not easy to change (like any other setting). Totally made a lot of people think local LLMs are dumb as rocks. Just checked, still not fixed -- defaults to 4K if less than 24GB VRAM: https://docs.ollama.com/context-length
2) Registry mistrust due to Deepseek R1 naming. What model do you download/start with `ollama run deepseek-r1`? Not Deepseek R1, but this "for research purposes" thinking finetune of Llama 3 released alongside R1 paper: https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Llama... . For actual R1 you have to pull `deepseek-r1:671b`
3) Can't reuse existing ggufs you have. If you point it to gguf, it would create a copy of it in ollama folder.
4) Doesn't accept engine parameters from CLI args. Only env variables or modelfile. But some things are only in env variables (e.g. KV cache quantization). And even then not even close to what's available in llama.cpp.
5) Often meaningfully slower than llama.cpp
6) VC-funded wrapper for llama.cpp. For long time had questionable attribution to it. Now looks good.
> (and also because VLLM is not super Windows friendly)
llama.cpp is Windows friendly and fairly easy. Not GUI app with installer, but easy. Zip with binaries are on github, run llama-server.exe -hf unsloth/gemma-4-12B-it-qat-GGUF:UD-Q4_K_XL
will pull the model you expect from hugging face (or -m for manually downloaded .gguf) and start web server with API and HTML chat app.
Using oh-my-pi in a prebuilt environment that I let Qwen build too.
Codex wouldn’t even look at the files - literally, as soon as it read something with CTF it shut down. Didn’t even offer to fall back to a dumber model.
Using opencode and it built a old fashioned arcade vertical shooter with no issues.
Images are ok'ish, just had grok create updated images, and it came out great.
a) Don't quantize your KV cache
b) Don't run quantizations of the LLM that are worse than the best available Q8 (the largest possible file size unsloth GGUF for a given model like qwen 3.8 27B as an example). I would rather things go slowly but I have confidence that it's doing things more accurately.
The written text is good lol
Local models running on your GTX 5090 or Macbook M99 are never going to reach the levels of the currently hosted "frontier models" with hundreds/thousands of CPU's and GPU's. Get a grip.
Anybody who's tested this over the last 12-18 months knows this. Don't waste our time because you're late to the party.
If I understand correctly, this failure mode is just not possible with llama.cpp / ik_llama.cpp, which enforces token generation to follow the grammar once a tool call is detected.
> ... and botched Cisco command line syntax (the correct command was ‘show arp’, while they executed ‘show run’)
But this failure mode can still happen.
Anyway, NVFP4 and AWQ W4A16 are generally regarded as low quality quants. IQK/Trellis quants from ik_llama.cpp and EXL3 quants from exllama should work better.
So, perhaps the lesson here is "don't use vllm at home"?
What? It's literally three actions and you're good: download llama.cpp, download the model on Huggingface, and run it with.
I have no idea how it's supposed to take two hours (unless you have a slow connection and the model download takes this much time, that is).
I also wonder if this manifests much less in contexts where the libraries/frameworks are a large part of the training set. It may be that the model doesn’t generalize well so it’s always better to use knowledge in its training set vs attempting to understand how to use a new, potentially never before seen (from the model perspective) api
Claude has got better at "just fucking doing it" by asking if it's ok to go read the latest github issues and pull the README, which means that people will likely get lazier and lazier.
Forums with full no verification pseudonyms seem like they have a real challenge ahead. How long until we need humanhackernews.com with public pseudonyms and a private trusted verification?
I've discovered a lot of neat tooling this way that I otherwise would not have bothered tryin to set up because that can take up a lot of time. And even when it's fast, you're suffering from context-switching penalties. You framed it as "read the latest github issues and pull the README" but anyone who has worked in tech for an appreciable amount of time knows that that stuff can turn into a deep rabbit hole.
However coming up with a prompt that didn't turn out total garbage was impossible. After wasting over an hour and I ended up getting Qwen side by side with Llama 3.2 3B, just to see if I was being stupid. Nope, it just looks like Llama is orders of magnitude better at this specific task for some reason).
If you think I'm doing it wrong, you're probably right, I don't know a ton about local LLMs. But I hand selected 50 songs, set up ollama with both LLMs, and for each iteration on the prompt text, ran both LLMs 10x times per song. Side-by-side comparisons showed that Qwen 3 4B was so bad that I actually downloaded Qwen again, thinking there must have been some mistake and I accidentally grabbed an old 1B model.
Probably this:
https://huggingface.co/empero-ai/Qwen3.8-4B-Distill
> Qwen3.8-4B is a full-parameter distillation of Qwen3.8 2.4T A95B into the Qwen3.5-4B architecture.
https://huggingface.co/Qwen/Qwen3-4B/blob/main/README.md
SO, if that isn't official it explains the results I got. They were dreadful.
> commited on May 21, 2025, over 1 year ago
Fairly old update to the README.md of (instead of Qwen3.8), should have raised some flags?
https://huggingface.co/Qwen/Qwen3.5-4B
Of course some things are not auto verifiable, and you'll have to give human judgement and input there, but you'll save a lot more time if you spend 1 week painstakingly writing checks for as many little things as possible and integrating them into the harness.
Hehe, this kind of sounds like the opposite of generalization. As in it’s just specialization at scale.
It’s paying hundreds of thousands of RLHF’ers from every subject and through some dystopian income stream.
It’s decent money don’t get me wrong, but you aren’t paid at if a task isn’t completed in time for example.
A couple of years ago, people I know got paid OK for relatively simple programming and logic RLHF tasks. But very soon it turned dark because that sort of data was required less and less, and the number of feedbackers has grown.
Today, the type of data the model developers pay for requires actual domain experience. E.g in software engineering they have people work in simulated environments with other LLMs, grade them, feedback, PRs, Jira everything.
This pays well and they treat you well, entice you with more money/task/hour, etc, for now. In few years when this gets drilled into LLMs, these guys will face the same painful hours and bad pay and less work and so on too.
In physical tasks, we are still in the early stages where basic packing clothes (in a textile factory setting) etc is being recorded and data is only now being used for training. Due to the problems with translating human hand data to robotic hands, these people do the factory work holding robotic grippers and operating that, you should see a YouTube video. But this also means that it's much less sweeping than data collection in SWE. In many cases it's not practical to collect data given that you have to use the specific gripper, wear a big gopro type thing, etc. So I am expecting much slower of an impact on physical tasks (of this kind) compared to how quick the uptake was in SWE/math.
I have yet to hear back from them on how it's going for teamwork white collar tasks, it's been a few months. Everyone is paying for the end products of those it seems - grok bot, perplexity computer, claure cowork, chatgpt work etc,. Not as much as their coding agents of course.
I got llama.cpp working with qwen3.6 and qwen3.8 by Googling and manually adjusting things according to reddit posts and Google not-really-helpful AI suggestions.
I tried settings up per-model stuff in settings.json, but again Google got in my way, and llama.cpp having 2 different settings.json (and Google lying about where 1 goes) made it far too difficulty to figure out. I spent hours on it.
Then I got fed up and asked Claude.
Immediately, it told me that the winget version of llama.cpp is for Vulkan, and I needed a different one and pointed at it. It doubled my speed.
Then it figured out what I was doing wrong with settings.json (wrong spot, global settings can't go in the per-model file, etc etc) and fixed all that, and got it working.
Then it tuned it somewhat.
Then I showed it the official settings pages for both models, and it undid the tuning and all the damage I had done with my tinkering, and got everything working.
In 30 minutes.
It was absolutely amazing.
Every time I see people recommending Qwen locally with llama.cpp, they just say "download it" and act like anyone that can't get it running is an idiot. But if there's a "using this settings.json" tutorial somewhere, I didn't find it, and neither did Google over a week of searching.
But Claude got it done for me.
Now, I admit, I haven't played with it much. Just before all this, I ran out of Claude on the $20 plan and bumped up to $100, and It's been so amazing that it's really hard to work on the local. Especially since it feels like Qwen3.8 35b a3b is probably around the corner, and why mess with 3.6 when 3.8 will probably release soon?