Moon Beam 16B A3B needs about 9.0 GB of memory at IQ3_XXS (a 8.1 GB file) with an 8K context window, so it fits on a 12 GB GPU. This repository has 3 quants, from 6.1 GB (IQ2_XXS) to 9.7 GB (IQ4_XS).
Moon Beam 16B A3B – 16B parameter LLM fine‑tuned for Tablinum engram retrieval
Moon Beam 16B A3B is a 16B parameter model with 3B active experts. It is trained to read retrieved passages and answer only when the answer is present. The model is released under the MIT license.
Good at
- Retrieval‑augmented answering
- Calibrated response confidence
- Tablinum engram integration
Limits
- No vision support
- Requires 8k context for full performance
- Large memory footprint for higher quants
Best for
- Knowledge‑base Q&A
- Document‑centric search assistants
- Entrainment of retrieval pipelines
Who it’s for. Developers building retrieval‑augmented assistants that need calibrated answers. Researchers testing mixture‑of‑experts fine‑tuning on a 16B backbone.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ2_XXS | 6.1 GB | 7.1 GB | Fits | Get file → | |
| IQ3_XXS | 8.1 GB | 9.0 GB | Too big | Get file → | |
| IQ4_XS | 9.7 GB | 10.6 GB | Too big | Get file → |
For 8 GB VRAM, use IQ3_XXS (file size 8.06 GB, estimated 9 GB memory at 8k context). For 10 GB VRAM, IQ4_XS (9.67 GB, ~10.6 GB memory). IQ2_XXS (6.08 GB, ~7.1 GB memory) is suitable for 6 GB VRAM setups.
What GPU do you need for Moon Beam 16B A3B?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: the best quant that fits is IQ2_XXS (6.1 GB file, about 7.1 GB in use).
- 12 GB VRAM: the best quant that fits is IQ4_XS (9.7 GB file, about 10.6 GB in use).
Which Moon Beam 16B A3B quant should you choose?
IQ3_XXS (8.1 GB) is the middle option in this repository. A reasonable choice when a 4-bit file will not fit, mostly on larger models. Quality loss is noticeable but often acceptable.
If memory is tight, the smallest file is IQ2_XXS at 6.1 GB, with the quality trade-offs that come with fewer bits. Use it only when memory is very tight. Expect a clear drop in quality, which is usually worse on small models and easier to live with on large ones.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Moon Beam 16B A3B
How much VRAM does Moon Beam 16B A3B need?
About 9.0 GB at IQ3_XXS with an 8K context window. That is the 8.1 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run Moon Beam 16B A3B on an 8 GB GPU?
Yes, with IQ2_XXS: the 6.1 GB file needs about 7.1 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Moon Beam 16B A3B GGUF?
It depends on the quant: IQ2_XXS is 6.1 GB, IQ3_XXS is 8.1 GB, IQ4_XS is 9.7 GB.
What context length does Moon Beam 16B A3B support?
The GGUF header lists a trained context length of 8,192 tokens (8K). The KV cache grows with the context you use: about 0.5 GB at 8K and 0.5 GB at 8K.
Where can I download Moon Beam 16B A3B GGUF?
The files are on Hugging Face in the tablinum/Moon-Beam-16B-A3B-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Moon Beam 16B A3B use?
Hugging Face lists the license as mit. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run Moon Beam 16B A3B on 8 GB VRAM?
Yes, the IQ3_XXS quant uses 8.06 GB file size and requires about 9 GB memory at 8k context.
Does the model support vision tasks?
No, the model has no vision capabilities.
What context length does it support?
It was trained with 8192 context tokens and performs best at that length.