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

81–90 of 682 posts

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

#81

Post by Manfrednotfunny »

mapontosevenths wrote: post #75
Manfrednotfunny wrote: post #25

Earlier quoted context omitted.

I would love to see any good research projects about it but i have the feeling that Frontier with MoE is making too fast of a progress so that a customized model would always be worse and that the MoE part is actually going somehow in this direction. On the other hand, at the GTC was a talk about coding in different lanugage (like spanish) and explaining that the quality between spanish and english is relevant differ…

Capability in LLM's is distributed throughout the manifold in subspaces. Even worse, the subspaces exist in superposition. That is to say, there is no single 'python' part of the model. The python bit is spread throughout the entire model and overlaps with other pieces that have similar, but unrelated, capabilities. For example the python subpspace might be partially in superposition with cupcake recipes, Esperanto,…

My current image from a MoE is that the base/core might be the more generic thing and that things like python are part of one expert though.

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

#82

Post by grim_io »

nutjob2 wrote: post #14

The more open weight models get released the greater the market for personal and small business oriented hardware to run these models. This will drive lower cost hardware, which has stagnated in recent years due to most software not needing the performance and capacity.

Higher demand for 5090's did not make them cheaper, because Nvidia got much higher margin products to focus on.

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

#83

Post by michimagdesign »

mycall wrote: post #64
pu_pe wrote: post #46

Earlier quoted context omitted.

Based on the benchmarks, it seems that Muse Glimmer barely edges out against Qwen3.6 27B, except for tool-calling skills (MCP, etc.). I wouldn't be surprised if they released it now because they are afraid they wouldn't beat Qwen3.8 27B.

Do AI companies make release plans based on upcoming other models like this? I would think all the processes that go into the repository and weight infrastructure pre-training, checkpointing, knowledge distillation, model compression, post training pipeline, ecosystem integrations, inference API, benchmarking, human eval/safety/alignment, docs, etc... all that dictates the release schedule.

Yes, not every model release is reactionary to other labs. Either they had hints for the release of other models or they cut efforts in late stage testing of the models to hit these earlier release dates. There’s always some flexibility. And there’s certainly the incentive to cannibalize the news cycles for competitor models.

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

#84

Post by polymorph1sm »

Some interesting findings from the chat template designs:

1. The template name is Onyx ATEM as found in the tool call exception message

2. It appears to be following a harmony-style chat template. But the tool use seems to be a xml like : / /

3. atem: a internal joke of meta in reverse?

https://huggingface.co/meta-models/Muse-Glimmer-30B/blob/mai...

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

#85

Post by Sabinus »

echelon wrote: post #73
mycall wrote: post #64

Earlier quoted context omitted.

Do AI companies make release plans based on upcoming other models like this? I would think all the processes that go into the repository and weight infrastructure pre-training, checkpointing, knowledge distillation, model compression, post training pipeline, ecosystem integrations, inference API, benchmarking, human eval/safety/alignment, docs, etc... all that dictates the release schedule.

There has been a long history of AI model releases made shortly before or after a major planned release by another company. Almost always to upstage or steal thunder. Just recently, Minimax H3 released as open weights on the eve of Seedance 2.5 global availability. It's not as good, but it's good enough and it's completely open. Flux 3, which is nowhere near as good as either, suddenly announced their release once ne…

I've seen it here on HN (it's particularly noticeable via the /active page) multiple times. If Google, OpenAI or Anthropic release something significant, odds are good you'll see a headline from one of the others.

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

#86

Post by stogot »

mycall wrote: post #64
pu_pe wrote: post #46

Earlier quoted context omitted.

Based on the benchmarks, it seems that Muse Glimmer barely edges out against Qwen3.6 27B, except for tool-calling skills (MCP, etc.). I wouldn't be surprised if they released it now because they are afraid they wouldn't beat Qwen3.8 27B.

Do AI companies make release plans based on upcoming other models like this? I would think all the processes that go into the repository and weight infrastructure pre-training, checkpointing, knowledge distillation, model compression, post training pipeline, ecosystem integrations, inference API, benchmarking, human eval/safety/alignment, docs, etc... all that dictates the release schedule.

the last few items there (benchmarking, human evaluation, docs) can be rushed or skipped by leadership if they want to beat comp. they probably spend a few weeks on those things normally

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

#87

Post by jurgenburgen »

neuroticnews25 wrote: post #71
dbbk wrote: post #58

Earlier quoted context omitted.

Well if you're spending thousands on API tokens already, you could just drop the same amount on a 128GB MacBook Pro and that's a one time cost.

Don't forget about energy usage, you'll probably never break even vs same model on openrouter.

If you can’t do it cheaper on your own hardware it does make you wonder how much of the cost of inference those large LLM providers are eating? Datacenter hardware isn’t magic.

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

#88

Post by ComputerPerson »

ignoramous wrote: post #48
gunalx wrote: post #4

Meta did not abandon opensource. I would love to see a smaller distill, or a moe of this size but the benchmarks seems competetive as long as it isnt benchmaxed witch i would not be suprosed if it is.

> Meta did not abandon opensource Open weights* I don't think outside of the Big 3 (Ant, OAI, GDM), given the strong competition from China, any other Lab has a chance at capturing the coding market if they aren't open weights (save for xAI whose latest Grok looks every bit good & will probably rely on Cursor for distribution instead of going open weights). There's literally no other selling point, as the capabilitie…

There was a good discussion yesterday on the DeepSeek Flash release thread about this.

There's a large market, very large, who want the best regardless of what it costs. Probably a large enough market to keep that domain of research afloat (as opposed to shifting research manpower to cost cutting).

The reasoning is just that the marginal cost of AI is very secondary to fixed costs of the businesses themselves; it's not an excuse to sacrifice performance.

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

#89

Post by amrit3128 »

zoobab wrote: post #23
sparkling wrote: post #18

Earlier quoted context omitted.

Even if you had a 64GB machine: Are you willing to reserve 90% of your memory to run a LLM? With dirt cheap models like deepseek-v4-flash that will run "forever" on $10, the answer for me is clearly: no.

"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.

Ah yes, I'm sure Trovalds and Stallman are harvesting my data through free software, aren't they? This argument is used by boomers who were fed cold war era propoganda that surely everybody is selfish, and you're always at fault.

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

#90

Post by dist-epoch »

mapontosevenths wrote: post #75
Manfrednotfunny wrote: post #25

Earlier quoted context omitted.

I would love to see any good research projects about it but i have the feeling that Frontier with MoE is making too fast of a progress so that a customized model would always be worse and that the MoE part is actually going somehow in this direction. On the other hand, at the GTC was a talk about coding in different lanugage (like spanish) and explaining that the quality between spanish and english is relevant differ…

Capability in LLM's is distributed throughout the manifold in subspaces. Even worse, the subspaces exist in superposition. That is to say, there is no single 'python' part of the model. The python bit is spread throughout the entire model and overlaps with other pieces that have similar, but unrelated, capabilities. For example the python subpspace might be partially in superposition with cupcake recipes, Esperanto,…

There was some paper about routing at training bio-knowledge into a particular region of the model, which you then can cutoff when serving. But you probably lose some efficiency since maybe you sized that region too small/too big.
Post reply on HN