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

61–70 of 682 posts

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

#61

Post by kzrdude »

QuadmasterXLII wrote: post #57
petcat wrote: post #49

As an industry, I wish we would stop calling these things "open weight" because it is too easy to confuse with actual "open source", which they are not. Photoshop source code+ OSI license = open source Photoshop binary you can run on your own computer = open weight Photoshop SaaS web app = closed, proprietary (Opus, GPT, etc.) "Open weight" models are still just binary blobs that are completely inscrutable. It's like…

Given an open weights model trained to sometimes bite kids, we can’t train it to not bite kids, even though billions of dollars of research have been thrown at this open problem. Given an open weights model trained to never bite kids, you can get it to bite kids with 10 prompts and a linear projection, the known simple algorithm doesn’t even need a backwards pass. yay asymmetry!

Any pointers to more info about that? Sounds interesting.

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

#63

Post by ignoramous »

scrlk wrote: post #5

Will be interesting to see how Qwen3.8 27B compares against this once it releases this week. Seems like dense 30B is back in fashion? EDIT: An open weight version of Muse Spark 1.2 is going to be released as well: https://x.com/alexandr_wang/status/2086756152034066792 https://xcancel.com/alexandr_wang/status/2086756152034066792

> Seems like dense 30B is back in fashion?

Surprising that Meta don't host this model, even as rate-limited free-tier.

> open weight version of Muse Spark 1.2

Wait. Is this "version" different from what Meta serves?

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

#64

Post by mycall »

pu_pe wrote: post #46
scrlk wrote: post #5

Will be interesting to see how Qwen3.8 27B compares against this once it releases this week. Seems like dense 30B is back in fashion? EDIT: An open weight version of Muse Spark 1.2 is going to be released as well: https://x.com/alexandr_wang/status/2086756152034066792 https://xcancel.com/alexandr_wang/status/2086756152034066792

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.

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

#67

Post by craigmart »

petcat wrote: post #49

As an industry, I wish we would stop calling these things "open weight" because it is too easy to confuse with actual "open source", which they are not. Photoshop source code+ OSI license = open source Photoshop binary you can run on your own computer = open weight Photoshop SaaS web app = closed, proprietary (Opus, GPT, etc.) "Open weight" models are still just binary blobs that are completely inscrutable. It's like…

I believe that comparing LLMs with traditional deterministic software is fundamentally misleading. It is extremely difficult to truly interpret what LLMs do internally, and as of now, nobody fully understands it. Even if you trained the LLM yourself, there is no source code you can simply read and learn from.

Sure, having information about how these models were trained is helpful for reproducibility, but it is basically impossible for anyone without substantial capital and access to the same (likely copyrighted) data to reproduce the model. For normal users, owning the model weights essentially means owning 100% of the model, you can inspect and study the weights in much the same way as the lab that produced the model can, you can modify the weights, and you can use and distribute them if the license allows you to

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

#68

Post by Mistletoe »

formerly_proven wrote: post #60
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.

4K bucks buys you around 180 months of with zero upfront cost.

Haha wow. I’m trying to even imagine the AI landscape in 15 years and I can’t.

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

#69

Post by cynicalsecurity »

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. Or models would need to be significantly dumbed down - then why use them at all? So far the idea of carrying a "thin" or "thin"-like device looks more reasonable to me, while running AI on your own server.

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

#70

Post by plufz »

Manfrednotfunny wrote: post #27
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.

I'm waiting for the speed/quality per dollar metric to go down a little bit further and then I will def run it at home. Its not just that you send a sentence to an API endpoint, you always send EVERYTHING to that agent as a context. You want to analyse your spending history? You now send everything to someone. Either no one cares but understands this implication on how easy it is to really capture you or no one reall…

Yeah, it does feel a bit silly with my encrypted disks, encrypted backups, unique passwords, advanced router, etc, while I send everything I do in plain text to anthropic.
Post reply on HN