---

license: apache-2.0

base_model:

- meta-models/Muse-Glimmer-30B

library_name: llama.cpp

pipeline_tag: image-text-to-text

tags:

- muse-glimmer

- muse-glimmer-30b

- gguf

- llama-cpp

- multimodal

- vision-language-model

- agentic-ai

- reasoning

- coding

- long-context

- local-llm

- image-understanding

- tool-use

- iq3-m

- iq4-xs

- iq4-nl

- q4-k-m

- q5-k-m

- q6-k

- q8-0

language:

- multilingual

---

Muse-Glimmer-30B-GGUF

High-quality GGUF quantizations of Meta's Muse Glimmer 30B.

This repository provides production-ready GGUF files converted directly from the official Muse Glimmer 30B Hugging Face release and quantized using the latest llama.cpp tooling.

Quick Download

| Quant | File |

|---------|---------|

| IQ3_M | Muse-Glimmer-30B-IQ3_M.gguf |

| IQ4_XS | Muse-Glimmer-30B-IQ4_XS.gguf |

| IQ4_NL | Muse-Glimmer-30B-IQ4_NL.gguf |

| Q4_K_M | Muse-Glimmer-30B-Q4_K_M.gguf |

| Q5_K_M | Muse-Glimmer-30B-Q5_K_M.gguf |

| Q6_K | Muse-Glimmer-30B-Q6_K.gguf |

| Q8_0 | Muse-Glimmer-30B-Q8_0.gguf |

| F16 | Muse-Glimmer-30B-F16.gguf |

Model Details

• Model: Muse Glimmer 30B
• Architecture: MuseGlimmerForConditionalGeneration
• Parameters: 30B
• Context Length: 131,072
• Modalities: Text + Image
• Base Model: meta-models/Muse-Glimmer-30B
• GGUF Conversion: llama.cpp
• License: Apache 2.0

Available Quantizations

IQ3_M

Excellent low-memory option with good quality retention.

Approximate size:

14-15 GB

Recommended for:

• 16 GB Macs
• 24 GB GPUs
• Memory-constrained deployments

IQ4_XS

High efficiency quantization with strong quality-per-GB.

Approximate size:

16-17 GB

Recommended for:

• Apple Silicon
• RTX 4090
• RTX 5090
• LM Studio users

IQ4_NL

Premium IQ quant focused on maintaining model quality at reduced size.

Approximate size:

17-18 GB

Recommended for:

• Users seeking maximum quality under 20 GB
• Apple Silicon deployments
• Long-context workflows

Q4_K_M

Recommended for most users.

Approximate size:

17-19 GB

Q5_K_M

Higher quality with moderate memory increase.

Approximate size:

20-22 GB

Q6_K

Near-lossless quantization.

Approximate size:

24-26 GB

Q8_0

Maximum quantized quality.

Approximate size:

31-33 GB

F16

Full precision GGUF.

Approximate size:

55-58 GB

Recommended Downloads

| Use Case | Recommended Quant |

|-----------|-----------|

| Lowest Memory Usage | IQ3_M |

| Best Efficiency | IQ4_XS |

| Best Quality Below 20 GB | IQ4_NL |

| General Purpose | Q4_K_M |

| High Quality | Q5_K_M |

| Near Lossless | Q6_K |

| Maximum Quantized Quality | Q8_0 |

| Full Precision | F16 |

Features

Muse Glimmer is designed for:

• Agentic task execution
• Coding and software engineering
• Long-context reasoning
• Function calling
• Tool use
• Screenshot understanding
• Document understanding
• Multimodal reasoning
• Vision-language tasks
• Local AI deployment

Runtime Compatibility

Tested or intended for:

• llama.cpp
• LM Studio
• Open WebUI
• Jan
• KoboldCpp
• Text Generation WebUI

llama.cpp Example

./llama-cli \
  -m Muse-Glimmer-30B-Q4_K_M.gguf \
  -c 131072

Hardware Recommendations

IQ3_M

• Apple Silicon 16 GB+
• RTX 4080 / 4090
• 24 GB GPUs

IQ4_XS

• Apple Silicon 24 GB+
• RTX 4090
• RTX 5090

IQ4_NL

• Apple Silicon 24 GB+
• RTX 4090
• RTX 5090

Q4_K_M

• Apple Silicon 32 GB+
• RTX 4090
• RTX 5090

Q5_K_M

• 24 GB+ VRAM
• Apple Silicon 48 GB+

Q6_K

• 32 GB+ available memory

Q8_0

• 40 GB+ available memory

F16

• 58 GB+ available memory

Conversion Information

Generated from:

meta-models/Muse-Glimmer-30B

Conversion pipeline:

HF Safetensors
→ GGUF F16
→ IQ / Q Quantization

Generated using native Muse Glimmer support in llama.cpp.

No fine-tuning, retraining, merging, alignment modifications, or architecture changes have been applied.

Available Files

Muse-Glimmer-30B-IQ3_M.gguf
Muse-Glimmer-30B-IQ4_XS.gguf
Muse-Glimmer-30B-IQ4_NL.gguf
Muse-Glimmer-30B-Q4_K_M.gguf
Muse-Glimmer-30B-Q5_K_M.gguf
Muse-Glimmer-30B-Q6_K.gguf
Muse-Glimmer-30B-Q8_0.gguf
Muse-Glimmer-30B-F16.gguf

SEO Keywords

Muse Glimmer GGUF, Muse Glimmer 30B GGUF, Muse Glimmer IQ3_M, Muse Glimmer IQ4_XS, Muse Glimmer IQ4_NL, Muse Glimmer Q4_K_M, Muse Glimmer Q5_K_M, Muse Glimmer Q6_K, Muse Glimmer Q8_0, Muse Glimmer llama.cpp, Muse Glimmer LM Studio, Muse Glimmer Open WebUI, Muse Glimmer local AI, Muse Glimmer multimodal, Muse Glimmer vision model, Muse Glimmer Apple Silicon, Muse Glimmer Mac.

Credits

• Meta Superintelligence Labs for Muse Glimmer.
• ggml-org for llama.cpp.
• Hugging Face for model hosting.

License

Apache 2.0.

Please also follow the original model license and usage policy provided with the base model.

Base model:

meta-models/Muse-Glimmer-30B