this post was submitted on 10 Aug 2026
25 points (100.0% liked)

LocalLLaMA

5046 readers
8 users here now

Welcome to LocalLLaMA! Here we discuss running and developing machine learning models at home. Lets explore cutting edge open source neural network technology together.

Get support from the community! Ask questions, share prompts, discuss benchmarks, get hyped at the latest and greatest model releases! Enjoy talking about our awesome hobby.

As ambassadors of the self-hosting machine learning community, we strive to support each other and share our enthusiasm in a positive constructive way.

Rules:

Rule 1 - No harassment or personal character attacks of community members. I.E no namecalling, no generalizing entire groups of people that make up our community, no baseless personal insults.

Rule 2 - No comparing artificial intelligence/machine learning models to cryptocurrency. I.E no comparing the usefulness of models to that of NFTs, no comparing the resource usage required to train a model is anything close to maintaining a blockchain/ mining for crypto, no implying its just a fad/bubble that will leave people with nothing of value when it burst.

Rule 3 - No comparing artificial intelligence/machine learning to simple text prediction algorithms. I.E statements such as "llms are basically just simple text predictions like what your phone keyboard autocorrect uses, and they're still using the same algorithms since <over 10 years ago>.

Rule 4 - No implying that models are devoid of purpose or potential for enriching peoples lives.

founded 3 years ago
MODERATORS
 

Meta released a ~30B parameter open weight dense model today called Muse Glimmer.

The main link I submitted goes to their official GGUF release for 24GB and 32GB discrete GPUs.

If you want the full sized safetensors, they're here: https://huggingface.co/meta-models/Muse-Glimmer-30B

Also, Meta's announcement post is here: https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model

Since the model is dense it's slow on Strix Halo (~2 tok/s) but I'm getting usable speeds on a discrete GPU (~25 tok/s on my hardware with the official GGUFs).

If you get complaints about unsupported model type in llama.cpp pull the latest code from git and rebuild from source. I've got it working with version: 10355 (dd1ea5243) and I'm testing it out now.

Edit: with the drafter, I'm usually getting more like 30~40 tok/s.

top 3 comments
sorted by: hot top controversial new old
[–] melfie@lemmy.zip 1 points 5 days ago

Glad to see more competition for this model size. I don’t like how terse it is, but the code quality doesn’t seem to far off from Qwen 3.6.

[–] ag10n@lemmy.world 4 points 1 week ago

Going to try this out too, context is listed at 128k so curious if it really compares to Qwen3.6-27B

[–] avidamoeba@lemmy.ca 4 points 1 week ago

Glimmer of hope? 🤭