Just asking, what is the recommended models for M3 MacBook with 18G memory? Seems modern local models are not available.
Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows
331–340 of 682 posts
Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows
#332Earlier quoted context omitted.
It’s rather amusing to me to read comments like this, and then simultaneously whenever a Chinese company or team releases open-weight models or whatever there is a giant round of applause, America is so behind, and there’s nothing but positive things to say about the intelligent, creative, and well-intentioned Chinese engineers (which is true, America certainly doesn’t have a monopoly on great people). Don’t you know…
The GP is claiming Meta is an awful company for what they have done and how they continue to treat their employees. That’s a perfectly ok opinion to hold, and many seem to agree. Have DeepSeek, Moonshot, or the other Chinese AI companies done such things that attract moral outrage?
The Chinese government owns half of every Chinese don't they? They've done some pretty horrific things.
Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows
#333Next step: Burn the weights of these local models into an asic that ships cheap on a laptop (AMD/taalas looking at you), and I will be a happy camper. Make it pluggable so I can select a model I want. I use qwen3.5-coder currently on my laptop, and while it works well enough for me, it is somewhat slow processing tokens.
I would hazard a guess that fast small models with a smart agent harness can do quite well compared to large models which cant be run locally.
Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows
#334From twitter Alexandr Wang
> 3/ muse glimmer was developed with its own architecture and recipe, optimized for its size and agentic performance requirements.
This means we're in the endgame does it not? If the architecture was NOT optimized for intelligence ...
Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows
#335Earlier quoted context omitted.
Meta can never be redeemed, but it's still valid to admit that FB at one point had a very badass engineering culture. They're one of 2 companies I would absolutely never work for (weapons etc aside). FB's recruiters hounded me so often I requested that they blackball me. The day they became Meta, I learned this by checking my email to see that they started trying to reach out again. I once again requested that they b…
> It's better than the benchmark scores seem to indicate compared to Qwen 3.6 27B. I'm very excited for 3.8 Is it worth considering if it's only marginally better than Qwen 3.6 though? Qwen 3.8 27B is almost there, and will probably be better suited as drop-in replacement for 3.6. Not even considering there's probably going to be a 3.8-35B-A3B too - which will have even better performance.
Qwen3.6 is very token inefficient with it's thinking. Quantized versions often get into loops.
Glimmer is trained with 4 effort levels, not just thinking on/off. Maybe it's more token efficient in general. There's official 4 bit quantizations with reported 1% loss across 15 benchmarks -- so quants probably work good.
IMO that alone is worth trying for, even if they're otherwise equal.
Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows
#336I lament the comments saying this in any way redeems Meta (the company). The researchers releasing this stuff have almost nothing to do with Meta other than being bankrolled by the slaughterhouse. You aren't the customer, you are the pawn in big tech's game of thrones. Your good will is a commodity to be traded, almost literally. It will be used against you the moment it's convenient. This is open weights because Met…
It’s rather amusing to me to read comments like this, and then simultaneously whenever a Chinese company or team releases open-weight models or whatever there is a giant round of applause, America is so behind, and there’s nothing but positive things to say about the intelligent, creative, and well-intentioned Chinese engineers (which is true, America certainly doesn’t have a monopoly on great people). Don’t you know…
These sentiments aren’t formed in a vacuum. America is becoming increasingly oligarchic and corrupt with decades of experience of companies profiting off harming people and lying through their teeth and Meta is like one of the worst offenders. They are pissing on every ally they have and once again started a war and both have material impacts on other countries.
Americans seem to take the US’ geeat reputation for granted and don’t realize how it has slipped and what that means. They also take for granted that China BAD is truth when this sentiment basically just sprang out of nowhere when the west realized it was their geopolitical rival. But to the rest of the worlds citizens, China is not starting any wars and is the source of cheap goods and innovation to other countries. EV batteries recently. The US is now directly causing high oil prices with their war and exports their rapacious companies like “prediction markets” which are 90% sports gambling now to the rest of the world. Meanwhile the classic American move to these kinds of comments is to claim that negative sentiment MUST be part of some bot campaign because surely no one could actually dislike the great America??
Those factors are all rightly part of the sentiment.
Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows
#337Earlier quoted context omitted.
It’s rather amusing to me to read comments like this, and then simultaneously whenever a Chinese company or team releases open-weight models or whatever there is a giant round of applause, America is so behind, and there’s nothing but positive things to say about the intelligent, creative, and well-intentioned Chinese engineers (which is true, America certainly doesn’t have a monopoly on great people). Don’t you know…
Apparently Americans haven’t got the memo yet that the world is moving closer to China.
I don't think that's entirely true. As an example you can look at how the European Union is increasingly putting up trade barriers with China and identifying opportunities for new rules and regulations to prevent Chinese dumping as America is also trying to do.
Japan, South Korea, the Philippines, and other countries participate in freedom of navigation and combined arms exercises because they perceive China to be a threat to their countries. [2]
I think it's more of a mixed bag. You see a lot of public talking points, and in Europe specifically a lot more healthy discussion about not being militarily as dependent on the United States as it has in the past and looking itself to follow Trump's lead to onshore capabilities (cloud for example) but I wouldn't read such moves as moving closer to China so much as they are hedging their bets a bit more.
Here's an article [1] that was reporting on this topic that I found interesting along with a select quote from the article:
> Luxembourg’s Prime Minister Luc Frieden said Thursday at an EU summit that China is “an existential threat for our industries.”
> Even Germany, whose economy has long relied on exports to China, is alarmed. Chancellor Friedrich Merz said this week that Beijing keeps its currency up to 30% undervalued, calling this “a massive competitive disadvantage.”
[2] https://www.msn.com/en-us/news/world/u-s-and-9-allies-just-h... - this isn't the "best" article but I just grabbed one to illustrate the point.
Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows
#338[dead]
Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows
#339Meta is rocking AI. As of last week I have been using their excellent muse coding harness with their model Muse Spark 1.2. Starting this morning I am running their new local 30B model muse-glimmer on my old MacMini 32G using Ollama (remember to increase the context size!) and pi coding harness. I am getting good results with muse-glimmer running locally, with the caveat that everything runs slowly (e.g., give it a ta…
Newb question but I’m curious what would help it to run faster? Would it need more vRAM or just system memory?
It runs very quickly on my RTX 5090 fwiw. Whole thing is loading entirely into vRAM with a ~130k context size (the max) fitting as well.
Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows
#340Earlier quoted context omitted.
It’s really interesting timing, Qwen over thinking is what kills it for me. I’m just glad we have more options in this size class now.
Just to play devil’s advocate: you can’t compare Qwen to a (proprietary/closed source) hosted model and deduce that Qwen is overthinking, as Qwen gives you the full reasoning/thinking trace while all the proprietary models now give you only a summary “to prevent distillation”, making it hard to properly compare apples to apples here.
People say Qwen overthinks because they analyzed the thinking traces, and Qwen finds the answer relatively quickly but then second guesses itself multiple times for another 20,000+ tokens. Regardless of what other models do, that's clearly overthinking.