---
license: apache-2.0
base_model: meta-models/Muse-Glimmer-30B
library_name: basert
pipeline_tag: image-text-to-text
tags:
  - basert
  - apple-silicon
  - quantized
---

# Muse-Glimmer-30B

[BaseRT](https://github.com/basecompute/baseRT) `.base` builds of [`meta-models/Muse-Glimmer-30B`](https://huggingface.co/meta-models/Muse-Glimmer-30B) for fast local inference on Apple Silicon (Metal).

These are the **official GGUF k-quants** from
[`meta-models/Muse-Glimmer-30B-GGUF`](https://huggingface.co/meta-models/Muse-Glimmer-30B-GGUF),
repackaged into `.base` — the super-block bytes are copied through verbatim, so
the weights are bit-identical to the upstream GGUFs rather than a
re-quantization of them. File names match the upstream ones.

Both builds are **text + vision**: the `mmproj-kquant.gguf` perception tower is
folded into the same bundle, so there is no separate projector file to manage.

## Files

| File | Source GGUF | Size |
|------|-------------|------|
| `muse-glimmer-30B-kquant-dynamic.base` | `muse-glimmer-30B-kquant-dynamic.gguf` | 21.0 GB |
| `muse-glimmer-30B-kquant-17gb.base` | `muse-glimmer-30B-kquant-17gb.gguf` | 18.1 GB |

Each `.base` is larger than its source GGUF because the perception tower is
folded in (the upstream `mmproj-kquant.gguf` is a separate 1.4 GB file).

`kquant-dynamic` is the default: a per-tensor Q4_K/Q5_K/Q6_K mix
(`Q4_K_M`-class). `kquant-17gb` is the smaller fixed-size build.

## Usage

```bash
curl -LsSf https://basecompute.co/install.sh | sh
basert pull basecompute/Muse-Glimmer-30B
basert chat basecompute/Muse-Glimmer-30B
```

Images use the model's own `<|patch|>` placeholder:

```bash
basert complete basecompute/Muse-Glimmer-30B --chat \
  --image photo.png --prompt "<|patch|>Describe this image."
```

Released under the apache-2.0 license, inherited from the base model.