Local artifact finder
Filter by the capability you need
Showing every official artifact. No JavaScript? All files remain visible.
K-Quant Dynamic
muse-glimmer-30B-kquant-dynamic.gguf
- Exact bytes
- 19,653,957,984
- Binary size
- 18.304 GiB
- Meta target
- 32 GiB
- Meta average degradation
- 0.2%
K-Quant 17 GB
muse-glimmer-30B-kquant-17gb.gguf
- Exact bytes
- 16,756,681,056
- Binary size
- 15.606 GiB
- Meta target
- 24 GiB
- Meta average degradation
- 1%
Vision projector
mmproj-kquant.gguf
- Exact bytes
- 1,400,328,928
- Binary size
- 1.304 GiB
- Use
- Image input with a K-Quant main file
DFlash drafter
dflash-kquant.gguf
- Exact bytes
- 1,631,205,312
- Binary size
- 1.519 GiB
- Use
- DFlash speculative decoding with a K-Quant main file
Selection logic
Choose the main file before the companions
Prefer more quality headroom
K-Quant Dynamic is the larger official main file. Meta reports0.2% average degradation and places it in the32 GiB target tier.
Prefer the smaller target tier
K-Quant 17 GB is the smaller official main file. Meta reports1% average degradation and places it in the24 GiB target tier.
Average degradation is not a per-task promise
Meta's published percentages summarize an average across its evaluation set. A particular coding, vision, agentic, reasoning, or safety result may move differently, and the figures are not an independent reproduction.
File manifest
Verify names and byte counts against the official tree
Wide table: swipe or use Shift + mouse wheel to inspect every column.
| Role | File | Bytes | GiB |
|---|---|---|---|
| Main model | muse-glimmer-30B-kquant-dynamic.gguf | 19,653,957,984 | 18.304 GiB |
| Main model | muse-glimmer-30B-kquant-17gb.gguf | 16,756,681,056 | 15.606 GiB |
| Vision projector | mmproj-kquant.gguf | 1,400,328,928 | 1.304 GiB |
| DFlash drafter | dflash-kquant.gguf | 1,631,205,312 | 1.519 GiB |
Limitations
- Official target tiers are planning references, not minimum memory or guaranteed compatibility.
- Main-file size does not include selected companions, KV cache, runtime buffers, or system usage.
- This page does not infer support in a runtime merely because the file format is GGUF.
- The filter only changes what is visible in this browser; it sends no query and stores no selection.
Continue from artifact choice
- Estimate hardware fit
Add the selected main file, companions, KV range, and planning reserve.
- Review system requirements
Understand the difference between an official target and a local capacity estimate.
- Verify llama.cpp support
Confirm that a runtime build includes the merged architecture support.
- Diagnose loading failures
Separate build recognition, memory, integrity, and companion issues.