Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

research.meta.ai

Post reply on HN
Print view View original

111–120 of 682 posts

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#111

Post by linguae »

cynicalsecurity wrote: post #69
sajithdilshan wrote: post #10

Still needs 32-64GB memory to run it locally. 64GB Macbook pro with an M5 chip costs more than 4k Euros in Germany. A more practical model would be a language specific (e.g Python or JVM language) and excellent at tool calling and reasoning. Maybe that way they can shrink it even more.

I don't understand the desire to run own AI models for programming locally. No laptop is ever going to be as powerful and energy efficient to run anything close to OpenAI, Anthropic or Google models. A model you can run on a loptop is simply not going to work as well as it's needed for programming. Small models for linguistic work fine, but anything more sophisticated simply won't provide enough resources or power. O…

I’m quite optimistic about the long-term future of local LLMs for privacy and cost control reasons. An LLM running on my own hardware, even if it’s not a laptop but a home server, is one where I don’t need to worry about token limits, token fees, privacy, and “rug-pulling” from the vendor.

In the short term, the big challenge is being able to afford hardware that can run a ~30B model. Last month I got to experiment with LLMs on a NVIDIA RTX 6000 Ada Generation as a visiting researcher during my summer break. I see the power of local LLMs for agentic coding; they’re no Claude, but they are quite useful. I wish I had gotten into local LLMs before hardware has gotten prohibitively expensive and in some cases unavailable; Apple discontinued certain Mac Minis and Mac Studios with high amounts of RAM due to the RAM shortage.

Hopefully high RAM prices don’t become a new normal, though the next year or two doesn’t look good.

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#113

Post by mirekrusin »

avaer wrote: post #101

I 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…

There is literally not a single comment like this, the only off topic comment like this is yours.

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#114

Post by Havoc »

cmrdporcupine wrote: post #38
Havoc wrote: post #13

The favourable comparisons to Gemma 4 and qwen3.6 look promising!

Those two offer MoE variants, this doesn't seem to. Dense model makes it dog slow on anything without HBM. Max 15tok/sec on decode on DDR5 systems like a Spark or a Strix Halo -- and that's at 4 bit quant.

The benchmark comparison is against the dense variants not MoE

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#116

Post by LogicFailsMe »

prplxd_nihilist wrote: post #33
zoobab wrote: post #23

Earlier quoted context omitted.

"With dirt cheap models like deepseek-v4-flash that will run "forever" on $10, the answer for me is clearly: no." When it's free, you are the product.

I see many people saying deepseek and other chinese providers have always been profitable. Also they show their training costs publicly. Can't say for sure since I have not used it personally, but I think they'll for sure outlive the western SOTAs.

OpenAI apparently runs a profitable inference business with 40% gross margin, but their advertising budget is nutso and their real costs are pretraining and research. I suspect Deepseek's comp is not predicated on capturing the lightcone of all future value, some googling insinuates their top pay is $212K US which would support that suspicion. Compare and contrast with the $1.35M and up at OpenAI.

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#117

Post by keybored »

avaer wrote: post #101

I 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…

Any retort to do this like “but why would they just openly release this”[1] pretty much answers itself. Public relations.

If a company can spend money to redeem itself then, well, it can (game theoretically or whatever) do whatever it wants in the future and then spend money to wipe the slate clean.

[1] By which I mean: the very act of being prompted to ask such a question, of planting a seed like hmm, Meta might have some aspects which are good for us. You don’t have to be convinced of it. Just the seed itself can pay for itself.

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#118

Post by xienze »

lostmsu wrote: post #92
IsTom wrote: post #59

Earlier quoted context omitted.

How is 30B smaller than 27B?

They say it is trained with quantization awareness, so it should only be 15GB or so. Qwen was only trained in FP8 with QAT. UPD, NVM, got misled by comments here. It is actually almost 60 GB so much larger

You're mixing up sizes of different quants. The 60GB is unquantized, and Qwen's unquantized size is around 54GB. Their sizes as like quantization levels are similar.

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#119

Post by monster_truck »

avaer wrote: post #101

I 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…

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 blackball me. This by extention taints OAI, the other company I'll never work for.

After a few hours with Glimmer I'm pretty impressed. It's better than the benchmark scores seem to indicate compared to Qwen 3.6 27B. I'm very excited for 3.8

Re: Muse Glimmer: 30B-parameter model optimized for always-on local agent workflows

#120

Post by skinfaxi »

avaer wrote: post #101

I 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…

How is this non-sequitor the top comment?
Post reply on HN