I ran Gemma 4 as a local model in Codex CLI

https://blog.danielvaughan.com/i-ran-gemma-4-as-a-local-model-in-codex-cli-7fda754dc0d4

by dvaughan • 3 months ago

285 116 3 months ago

blog.danielvaughan.com

anactofgod• 3 months ago

Amazing. Thanks for your detailed posts on the bake-off between the Mac and GB10, Daniel, and on your learnings. I had trying similar on both compute platforms on my to-do list. Your post should save me a lot of debugs, sweat, and tears.

fortyseven4• 3 months ago

I've been VERY impressed with Gemma4 (26B at the moment). It's the first time I've been able to use OpenCode via a llamacpp server reliably and actually get shit done.

In fact, I started using it as a coding partner while learning how to use the Godot game engine (and some custom 'skills' I pulled together from the official docs). I purposely avoided Claude and friends entirely, and just used Gemma4 locally this week... and it's really helped me figure out not just coding issues I was encountering, but also helped me sift through the documentation quite readily. I never felt like I needed to give in and use Claude.

Very, very pleased.

logicallee• 3 months ago

Thanks for sharing that. What kind of hardware are you running this on?

fortyseven4• 3 months ago

4090, 128gb of ram (long before you'd have to take out a loan). I'm fairly sure it would run just as fine on a 3090.

Thanks to the settings suggestions in the article, I was able to squeeze in the 31b model. Still testing, but it's real tight in 24gb of vram. A bit slower, too, but usable. Not sure I'm seeing much of a quality boost yet, but I'm still testing.

blackmanta• 3 months ago

With a nvidia spark or 128gb+ memory machine, you can get a good speed up on the 31B model if you use the 26B MoE as a draft model. It uses more memory but I’ve seen acceptance rate at around 70%+ using Q8 on both models

foobar10000• 3 months ago

1 token ahead or 2?

It's interesting - imo we'll soon have draft models specifically post-trained for denser, more complicated models. Wouldn't be surprised if diffusion models made a comeback for this - they can draft many tokens at once, and learning curves seem to top out at 90+% match for auto-regressive ones so quite interesting..

electroglyph• 3 months ago

flow matching is making some strides right now, too

ehtbanton• 3 months ago

This is genuinely very helpful. I'm planning a MacBook pro purchase with local inference in mind and now see I'll have to aim for a slightly higher memory option because the Gemma A4 26B MoE is not all that!

tomr75• 3 months ago

pretty sure Nvidia GPU is better bang for buck because of usable inference speed..

egorfine6• 3 months ago

I have upgraded my M4 Pro 24GB to M5 Pro 48GB yesterday. The same Gemma 4 MoE model (4bit, don't remember which version) runs about 8x faster on M5 Pro and loads 2x times faster in memory.

So yes, do purchase that new MacBook Pro.

croemer• 3 months ago

You don't know if it's the newer model or the increase in RAM. If someone has already got 48GB it they might not benefit much. You changed 2 things at once.

59nadir2• 3 months ago

If you're doing it specifically for inference (or in most other situations) a Mac(book) represents very low RoE.

59nadir2• 3 months ago

s/RoE/RoI

brcmthrowaway• 3 months ago

Nothing about omlx?

vsrinivas• 3 months ago

Hey - I use the same, w/ both gemma4 and gpt-oss-*; some things I have to do for a good experience:

1) Pin to an earlier version of codex (sorry) - 0.55 is the best experience IME, but YMMV (see https://github.com/openai/codex/issues/11940, https://github.com/openai/codex/issues/8272).

2) Use the older completions endpoint (llama.cpp's responses support is incomplete - https://github.com/ggml-org/llama.cpp/issues/19138)

tuzemec2• 3 months ago

I'm currently experimenting with running google/gemma-4-26b-a4b with lm studio (https://lmstudio.ai/) and Opencode on a M3 Ultra with 48Gb RAM. And it seems to be working. I had to increase the context size to 65536 so the prompts from Opencode would work, but no other problems so far.

I tried running the same on an M3 Max with less memory, but couldn't increase the context size enough to be useful with Opencode.

It's also easy to integrate it with Zed via ACP. For now it's mostly simple code review tasks and generating small front-end related code snippets.

jwr2• 3 months ago

I do the same thing on a MacBook Pro with an M4 Max and 64GB. I had problems until the most recent LM Studio update (0.4.11+1), tool calling didn't work correctly.

Now both codex and opencode seem to work.

declan_roberts2• 3 months ago

Which do you prefer? And what lmstudio api works best for these tools?

usagisushi• 3 months ago

I have a similar setup. It might be worth checking out pi-coding-agent [0].

The system prompt and tools have very little overhead (<2k tokens), making the prefill latency feel noticeably snappier compared to Opencode.

[0] https://www.npmjs.com/package/@mariozechner/pi-coding-agent#...

tuzemec2• 3 months ago

Thanks! I just ran a quick test with pi, and it's working a bit faster.

jtbaker• 3 months ago

Just set up Pi after listening to Marios talk at AIE Europe[0] and have solid initial impressions! Especially on limited hardware like a MB Air, seems a lot more resource efficient

[0] https://www.youtube.com/live/_zdroS0Hc74?t=3633s

theshrike79• 3 months ago

Pi is _really_ good for personal stuff, but since it lacks every single safety imaginable, it's not realy something one can deploy in a corporate environment :D

smrtinsert• 3 months ago

gguf or mlx? edit, just tried a community mlx and lm studio said it didn't support loading it yet.

davidwritesbugs• 3 months ago

I did the same using the mlx version on an M1 Macbook using LMStudio integrated into XCode. I had to up the context size I ran it a against a very modest iOS codebase and it didn't do well, just petered out at one point. Odd. Pretty good chatbot and maybe against other code it'll work but not useful with XCode for me

rsolva• 3 months ago

I run this model on my AMD RX7900XTX with 24GB VRAM with up to 4 concurrent chats and 512K context window in total. It is very fast (~100 t/s) and feels instant and very capable, and I have used Claude Code less and less these days.

ozgrakkurt• 3 months ago

Not sure if you already tried but both GLM Flash and Qwen models are much better than Gemma for that in my experience.

I am using a 24GB GPU so it might be different in your case, but I doubt it.

qingcharles2• 3 months ago

I spun up a GPU on Runpod and tried the 31b full res and it was really impressive. I'm now using it via the Google API which gives you 1500 requests a day for free, IIRC.

hak8or• 3 months ago

Be very careful about using googles apis as a consumer, they have poor rate limiting and ineffective anomoly protection.

I (a hobbyist running a small side project for a dollar or two a month in normal usage, so my account is marked as "individual") got hit with a ~$17,000 bill from Google cloud because some combination of key got leaked or my homelab got compromised, and the attacker consumed tens of thousands in gemini usage in only a few hours. It wasn't even the same Google project as for my project, it was another that hasn't seen activity in a year+.

Google refuses to apply any adjustments, their billing specialist even mixed up my account with someone else, refuses to provide further information for why adjustments are being rejected, refuses any escalation, etc. I already filed a complaint with the FTC and NYS attorney General but the rep couldn't care any less.

My gripe is not that the key was potentially leaked or compromised or similar and then I have to pay as a very expensive "you messed up" mistake, it's that they let an api key rack up tens of thousands in maybe 4 hours or so with usage patterns (model selection, generating text vs image, volume of calls, likely different IP and user agent and whatnot). That's just predatory behavior on an account marked as individual/consumer (not a business).

zihotki• 3 months ago

For coding it makes no sense to use any quantization worse than Q6_K, from my experience. More quantized models make more mistakes and if for text processing it still can be fine, for coding it's not.

segmondy2• 3 months ago

I don't think most people realize that. Quality of tokens beats quantity of token. I always tell folks to go as high a quant as you can only go lower if you just don't have the memory capacity.

hmokiguess3• 3 months ago

what do you mean with that, I’m not sure I understood what you said

shaz0x• 3 months ago

On mobile the Q4 vs Q6 tradeoff flips. Gemma 4 E2B at Q4_K_M barely fits in RAM on a 6GB Android, so Q6 isn't on the table. In practice the Q4 hit shows up in tool-call reliability more than general reasoning, which is usually fine for a constrained skill surface.

mhitza• 3 months ago

The finding I did not expect: model quality matters more than token speed for agentic coding.

I'm really surprised how that was not obvious.

Also, instead of limiting context size to something like 32k, at the cost of ~halving token generation speed, you can offload MoE stuff to the CPU with --cpu-moe.

kingstnap• 3 months ago

It's even more strange how its not obvious to someone who uses codex extensively daily.

The rate limiting step is the LLM going down stupid rabbit holes or overthinking hard and getting decision paralysis.

The only time raw speed really matters is if you are trying to add many many lines of new code. But if you are doing that at token limiting rates you are going to be approaching the singularity of AI slop codebase in no time.

dangoodmanUT• 3 months ago

Yeah, it’s like drinking coffee when being really tired. You’re still tired, just “faster”, it’s a weird sensation.

triceratops• 3 months ago

Why would token speed matter for anything other than getting work done faster? It's in the name - "speed".

dminik• 3 months ago

This would be true if the models were capable of always completing the tasks. But, since their failure rate is fairly high, going in a wrong direction for longer could mean that you take more time than a faster model, where you can spot it going wrong earlier.

Havoc• 3 months ago

You can also try speculative decoding with the E2B model. Under some conditions it can result in a decent speed up

danilop• 3 months ago

Nice walkthrough and interesting findings! The difference between the MoE and the dense models seems to be bigger than what benchmarks report. It makes sense because a small gain in toll planning and handling can have a large influence on results.

egorfine6• 3 months ago

Related: I have upgraded my M4 Pro 24GB to M5 Pro 48GB yesterday. The same Gemma 4 MoE model (Q4) runs about 8x more t/s on M5 Pro and loads 2x times faster from disk to memory.

Gonna run some more tests later today.

Confiks3• 3 months ago

The same Gemma 4 MoE model (Q4)

As you have so much RAM I would suggest running Q8_0 directly. It's not slower (perhaps except for the initial model load), and might even be faster, while being almost identical in quality to the original model.

And just to be sure: you're are running the MLX version, right? The mlx-community quantization seemed to be broken when I tried it last week (it spit out garbage), so I downloaded the unsloth version instead. That too was broken in mlx-lm (it crashed), but has since been fixed on the main branch of https://github.com/ml-explore/mlx-lm.

I unfortunately only have 16 GiB of RAM on a Macbook M1, but I just tried to run the Q8_0 GGUF version on a 2023 AMD Framework 13 with 64 GiB RAM just using the CPU, and that works surprisingly well with tokens/s much faster than I can read the output. The prompt cache is also very useful to quickly insert a large system prompt or file to datamine although there are probably better ways to do that instead of manually through a script.

egorfine6• 3 months ago

As you have so much RAM I would suggest running Q8_0 directly

On the 48GB mac - absolutely. The 24GB one cannot run Q8, hence why the comparison.

And just to be sure: you're are running the MLX version, right?

Nah, not yet. I have only tested in LM Studio and they don't have MLX versions recommended yet.

but has since been fixed on the main branch

That's good to know, I will play around with it.

minimaxir• 3 months ago

Gemma 4 is not supported by the MLX engine yet.

egorfine6• 3 months ago

That too was broken in mlx-lm (it crashed), but has since been fixed on the main branch

Unfortunately I have got zero success running gemma with mlx-lm main branch. Can you point me out what is the right way? I have zero experience with mlx-lm.

karpetrosyan• 3 months ago

I think local models are not yet that good or fast for complex things, so I am just using local Gemma 4 for some dummy refactorings or something really simple.

dajonker• 3 months ago

I don't really have the hardware to try it out, but I'm curious to see how Qwen3.5 stacks up against Gemma 4 in a comparison like this. Especially this model that was fine tuned to be good at tool calling that has more than 500k downloads as of this moment: https://huggingface.co/Jackrong/Qwen3.5-27B-Claude-4.6-Opus-...

anana_2• 3 months ago

It's rather surprising that a solo dev can squeeze more performance out of a model with rather humble resources vs a frontier lab. I'm skeptical of claims that such a fine-tuned model is "better" -- maybe on certain benchmarks, but overall?

FYI the latest iteration of that finetune is here: https://huggingface.co/Jackrong/Qwopus3.5-27B-v3

1dom2• 3 months ago

I feel that's a little bit misleading.

That link doesn't have much affiliation with Qwen or anyone who produces/trained the Qwen models. That doesn't mean it's not good or safe, but it seems quite subjective to suggest it's the latest latest or greatest Qwen iteration.

I can see huggingface turning into the same poisoned watering-hole as NPM if people fall into the same habits of dropping links and context like that.

NitpickLawyer• 3 months ago

can squeeze more performance out of a model with rather humble resources vs a frontier lab.

That's the idea behind distillation. They are finetuning it on traces produced by opus. This is poor man's distillation (and the least efficient) and it still works unreasonably well for what it costs.

mapontosevenths2• 3 months ago

I'm just some guy on hackernews, but I actually did try this on my DGX Spark. I went back to Gemma 4 after a few rounds. My orchestration model kept having to send the Qwen model back to fix mistakes that Gemma wouldn't have made. I wound up with less working code per hour due to the mistakes.

Technically, I use OpenWebUI with Ollama, so I used the weights below, but it should be the same.

https://ollama.com/kwangsuklee/Qwen3.5-27B-Claude-4.6-Opus-R...

estimator7292• 3 months ago

I'd be super interested to hear about your workflow with OpenWebUI. I haven't figured out how to use it for anything other than the basic chatbot UI. I haven't been able to hook anything else into it

notpublic• 3 months ago

Jackrong has published the finetuning steps here. It seems to be quite thorough with notebooks etc. I am going through it myself now...

https://github.com/R6410418/Jackrong-llm-finetuning-guide

OutOfHere4• 3 months ago

Gemma 4 is a strongly censored model, so much so that it refused to answer medical and health related questions, even basic ones. No one should be using it, and if this is the best that Google can do, it should stop now. Other models do not have such ridiculous self-imposed problems.

tgv2• 3 months ago

I don't quite get why you feel so strongly about it that this should be a deal breaker for everyone. It's really much better than a wrong answer, for everyone.

OutOfHere4• 3 months ago

It's really much better than a wrong answer

That is a bad premise and a false dichotomy, because most medical questions are simple, with well-known standard answers. ChatGPT and Gemini answer such questions correctly, also finding glaring omissions by doctors, even without having to look up information.

As for the medical questions that are not simple, the ones that require looking up information, the model should in principle be able to respond that it does not know the answer when this is truthfully the case, implying that the answer, or a simple extrapoloation thereof, was not in its training data.

mft_2• 3 months ago

I suspect a possible future of local models is extreme specialisation - you load a Python-expert model for Python coding, do your shopping with a model focused just on this task, have a model specialised in speech-to-text plus automation to run your smart home, and so on. This makes sense: running a huge model for a task that only uses a small fraction of its ability is wasteful, and home hardware especially isn't suited to this wastefulness. I'd rather have multiple models with a deep narrow ability in particular areas, than a general wide shallow uncertain ability.

Anyway, is it possible that this may be what lies behind Gemma 4's "censoring"? As in, Google took a deliberate choice to focus its training on certain domains, and incorporated the censor to prevent it answering about topics it hasn't been trained on?

Or maybe they're just being sensibly cautious: asking even the top models for critical health advice is risky; asking a 32B model probably orders of magnitude moreso.

OutOfHere4• 3 months ago

is it possible that this may be what lies behind Gemma 4's "censoring"

Your explanation would make sense if various other rare domains were also censored, but they aren't, so it doesn't.

asking even the top models for critical health advice is risky

Not asking, and living in ignorance, is riskier. For high-stakes questions, of course I'd want references that only an online model like ChatGPT or Gemini, etc. would be able to find. If I am asking a local model for health advice, odds are that it is because I am traveling and am temporarily offline, or am preparing off-grid infrastructure. In both cases I definitely require a best-effort answer. I also require the model to be able to tell when it doesn't know the answer.

If you would, ignore health advice for a moment, and switch to electrical advice. Imagine I am putting together electrical infrastructure, and the model gives me bad advice, risking electrocution and/or a serious fire. Why is electrical advice not censored, and what makes it not be high-stakes!? The logic is the same.

For the record, various open-source Asian models do not have any such problem, so I would rather use them.

stratos123• 3 months ago

I suspect a possible future of local models is extreme specialisation - you load a Python-expert model for Python coding, do your shopping with a model focused just on this task, have a model specialised in speech-to-text plus automation to run your smart home, and so on.

I'd find this very surprising, since a lot of cognitive skills are general. At least on the scale of "being trained on a lot of non-Python code improves a model's capabilities in Python", but maybe even "being trained on a lot of unrelated tasks that require perseverance improves a model's capabilities in agentic coding".

For this reason there are currently very few specialist models - training on specialized datasets just doesn't work all that well. For example, there are the tiny Jetbrains Mellum models meant for in-editor autocomplete, but even those are AFAIK merely fine-tuned on specific languages, while their pretraining dataset is mixed-language.

vorticalbox• 3 months ago

You can get abliterated versions that have no (or very limited) refusals.

I tend to use Huihuiai versions.

fortyseven4• 3 months ago

Weird. A great number of my medical or legal queries are actually answered, but come with a disclaimer, often at the end of the inference. (I'd offer up some examples, but I'm not at the desk.)

I also find that you can coerce a wide spectrum of otherwise declined queries by editing its initial rejection into the start of an answer. For example changing the "I'm sorry I can't answer that..." response to "Here's how..." And then resubmitting the inference, allowing it to continue from there. It's not perfect, sometimes it takes multiple attempts, but it does work. At least in my experience. (This isn't Gemma-specific tip, either. Nearly every model I've tried this with tends to bend quite a bit doing this.)

cbg0• 3 months ago

Since it's open weights there's nothing stopping you from grabbing one of the uncensored variants from huggingface.co

icedchai• 3 months ago

There are uncensored / "abliterated" / heretic versions available on Huggingface.

meander_water• 3 months ago

I would have liked to see quality results between the different quantization methods - Q4_K_M, Q_8_0, Q_6_K rather than tok/s

alvsilvao• 3 months ago

I also tried Gemma 4 on a M1 Macbook Pro. It worked but it was too slow. Great to know that it works on more advanced laptops!

magic_hamster2• 3 months ago

Ollama is the worst engine you could use for this. Since you are already running on an Nvidia stack for the dense model, you should serve this with vLLM. With 128GB you could try for the original safetensors even though you might need to be careful with caches and context length.

fortyseven4• 3 months ago

Strangely, I haven't had a lot of luck with vLLM; I finally ended up ditching Ollama and going straight to the tap with llama-serve in llamacpp. No regrets.

magic_hamster2• 3 months ago

Good job. llama.cpp is already much better.

taf2• 3 months ago

I did this with qwen 3.5 - tool calling was the biggest issue but for getting it to work with vllm and mlx I just asked codex to help. The bulk of my the time was waiting on download. For vllm it created a proxy service to translate some codex idioms to vllm and vice versa. In practice I got good results on my first prompt but followup questions usually would fail due to the models trouble with tool calling - I need to try again with gemma4

hacker_homie• 3 months ago

Using Gemma4-31B-q4_NL in open code with a 128k context and it’s been great.

declan_roberts2• 3 months ago

How successful is it at calling tools?

axjns• 3 months ago

I'm suprised folks are having such great coding experiences. Using Gemma-4 on a moderately complex code base, it utterly flailed and gave a half baked implementation.

antisthenes• 3 months ago

I mean OP had them code something called "parse_csv_summary", which sounds like something a junior dev can whip up in an hour or so, since CSV manipulation in Python is one of the most basic tasks you can encounter in the problem space.

Banking, scientific data analysis, sales, etc. Everything uses and manipulates csvs.

So it was firmly in the bottom quartile of difficulty - and there LLMs actually do quite well.

segmondy2• 3 months ago

"The reason I had not done this before is that local models could not call tools. "

Rubbish, we have been calling tools locally for 2 years, and it's very false that gemma3 scored under 7% in tool calling. Hell, I was getting at least 75% tool calling with llama3.3

girvo• 3 months ago

This entire article reads like AI slop anyway.

I also recommend anyone with a GB10 device to go try out the spark-vllm-docker setup, and check the Nvidia GB10 forums for the recently released optimised Qwen 3.5 122B A10B setup: 50tk/s is quite impressive for a decent local model!

StrLght• 3 months ago

I was also surprised by this sentence. It sounds like this is the author's first attempt at running models locally.

Or maybe the author has been running heavily quantized small models all that time — Gemma 4 gguf he's using is Q4 and only 16 GB. In my experience quants like this tend to perform much worse.

nphard85• 3 months ago

To be fair, the author does mention the huge difference between Gemma 3 and Gemma 4 on Tau function calling benchmark.

bitwize• 3 months ago

I recently spun up Gemma 4 26B-A4B on my local box and pointed OpenCode at it, and it did reasonably well! My machine is 8 years old, though I had the foresight to double the RAM to 32 GiB before the RAMpocalypse, and I can get a little bit of GPU oomph but not a lot, not with a mere GTX 1070. So it's slow, and nowhere near frontier model quality, but it can generate reasonable code and is good for faffing with!

cjbgkagh• 3 months ago

I've been playing with this for the last few days. The model is fast, pretty smart, and I am hitting the same tool use issues. This blog post is unusually pertinent. The model speed isn't an issue on my dual 4090s, the productivity is mainly limited by the intelligence (while high it's still not high enough for some tasks) and getting stuck in loops.

What I would like is for it to be able detect when these things happen and to "Phone a Friend" to a smarter model to ask for advice.

I'm definitely moving into agent orchestration territory where I'll have an number of agents constantly running and working on things as I am not the bottleneck. I'll have a mix of on-prem and AI providers.

My role now is less coder and more designer / manager / architect as agents readily go off in tangents and mess that they're not smart enough to get out of.

adrian_b2• 3 months ago

Google has replaced chat_template.jinja and tokenizer_config.json a few days ago in gemma-4-31B-it, which is supposed to have solved some problems related to tool invocation.

So if you have not updated your model, you should do it.

OsrsNeedsf2P• 3 months ago

I laughed when I saw the .md table rendering as a service. Blows my mind what people will use

m3kw9• 3 months ago

The setup allots around 4k of context after system prompt lol

gertlabs3• 3 months ago

Gemma 4 26B really is an outlier in its weight class.

In our little known, difficult to game benchmarks, it scored about as well as GPT 5.2 and Gemini 3 Pro Preview on one-shot coding problems. It had me re-reviewing our entire benchmarking methodology.

But it struggled in the other two sections of our benchmark: agentic coding and non-coding decision making. Tool use, iterative refinement, managing large contexts, and reasoning outside of coding brought the scores back down to reality. It actually performed worse when it had to use tools and a custom harness to write code for an eval vs getting the chance to one-shot it. No doubt it's been overfit on common harnesses and agentic benchmarks. But the main problem is likely scaling context on small models.

Still, incredible model, and incredible speed on an M-series Macbook. Benchmarks at https://gertlabs.com

seemaze• 3 months ago

Thats funny, it failed my usual ‘hello world’ benchmark for LLM’s:

“Write a single file web page that implements a 1 dimensional bin fitting calculator using the best fit decreasing algorithm. Allow the user to input bin size, item size, and item quantity.”

Qwen3.5, Nematron, Step 3.5, gpt-oss all passed first go..

iknowstuff• 3 months ago

Gemma 31B scoring below 26B-A4B?

gertlabs3• 3 months ago

In one shot coding, surprisingly, yes, by a decent amount. And it isn't a sample size issue. In agentic, no: https://gertlabs.com/?agentic=agentic

My early takeaway is that Gemma 26B-A4B is the best tuned out of the bunch, but being small and with few active params, it's severely constrained by context (large inputs and tasks with large required outputs tank Gemma 26B's performance). We're working on a clean visualization for this; the data is there.

It's not uncommon for a sub-release of a model to show improvements across the board on its model card, but actually have mixed real performance compared to its predecessor (sometimes even being worse on average).

datadrivenangel• 3 months ago

Overall it's a very good open weights model! Notably I thought it makes more dumb coding mistakes than GPT-OSS on my M5, but it's fairly close overall.

neonstatic• 3 months ago

I have very mixed feelings about that model. I want to like it. It's very fast and seems to be fit for many uses. I strongly dislike its "personality", but it responds well to system prompts.

Unfortunately, my experience with it as a coding assistant is very poor. It doesn't understand libraries it seems to know about, it doesn't see root causes of problems I want it to solve, and it refuses to use MCP tools even when asked. It has a very strong fixation on the concept of time. Anything past January 2025, which I think is its knowledge cutoff, the model will label as "science fiction" or "their fantasy" and role play from there.

prettyblocks• 3 months ago

For me the vision/OCR is much better than other models in weights class.

mudkipdev• 3 months ago

Does the large system prompt work fine for this model? If needed, you could use a lightweight CLI like Pi, which only comes with 4 tools by default

2001zhaozhao• 3 months ago

I think it might be a good idea to make some kind of local-first harness that is designed to fully saturate some local hardware churning experiments on Gemma 4 (or another local model) 24/7 and only occasionally calls Claude Opus for big architectural decisions and hard-to-fix bugs.

Something like:

* Human + Claude Opus sets up project direction and identifies research experiments that can be performed by a local model

* Gemma 4 on local hardware autonomously performs smaller research experiments / POCs, including autonomous testing and validation steps that burn a lot of tokens but can convincingly prove that the POC works. This is automatically scheduled to fully utilize the local hardware. There might even be a prioritization system to make these POC experiments only run when there's no more urgent request on the local hardware. The local model has an option to call Opus if it's truly stuck on a task.

* Once an approach is proven through the experimentation, human works with Opus to implement into main project from scratch

If you can get a complex harness to work on models of this weight-class paired with the right local hardware (maybe your old gaming GPU plus 32gb of RAM), you can churn through millions of output tokens a day (and probably like ~100 million input tokens though the vast majority are cached). The main cost advantage compared to cloud models is actually that you have total control over prompt caching locally which makes it basically free, whereas most API providers for small LLM models ask for full price for input tokens even if the prompt is exactly repeated across every request.

rahimnathwani• 3 months ago

Nico Bailon (author of many extensions for Mario's Pi coding agent) has something that might be a good starting point for this:

https://x.com/i/status/2043054831947198640

https://github.com/nicobailon/pi-model-switch

flux3125• 3 months ago

In my experience if you're coding or doing something that requires precision, quantizing the kv cache is definitely not worth it.

If you're just chatting or doing less precise things it's 1000% worth it going down to Q8 or sometimes even Q4

Avlin67• 3 months ago

but my dual 5090 is way faster your mac

dpoloncsak• 3 months ago

Is it really fair to benchmark LLMs on a Mac using ollama and not MLX? Does ollama make proper use of the M-series yet?

I ran Gemma 4 as a local model in Codex CLI

Related Stories