maple preview needs about 13.2 GB of memory at Q4_K_M (a 12.3 GB file) with an 8K context window, so it fits on a 16 GB GPU. This repository has 4 quants, from 5.5 GB (TQ2_0) to 40.5 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 1.6 GB.
20B-A1B ternary-weight Mixture-of-Experts reasoning LLM
Maple‑Preview is a 20.2B parameter model designed for efficient on‑device reasoning. It supports a 131,072 token context and runs on the PrismML llama.cpp fork. The model is released under the MIT license.
Good at
- High‑speed reasoning
- Large context handling
- Efficient on‑device inference
Limits
- No vision support
- Requires PrismML llama.cpp fork
- Limited agentic performance
Best for
- Solving complex reasoning tasks
- Educational problem solving
- Research prototyping
Who it’s for. Developers and researchers needing a fast, reasoning‑focused LLM that can run on GPUs with the PrismML fork. The model is suitable for local inference on devices with sufficient VRAM and for experimentation with ternary quantization.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| TQ2_0 | 5.5 GB | 6.4 GB | Fits | Get file → | |
| Q2_0 | 6.0 GB | 6.9 GB | Fits | Get file → | |
| Q4_K_M | 12.3 GB | 13.2 GB | Too big | Get file → | |
| F16 | 40.5 GB | 41.4 GB | Too big | Get file → |
For 8 GB VRAM use the TQ2_0 pack (5.45 GB file, ~6.4 GB memory at 8 k context). For 16 GB VRAM use Q2_0 (5.99 GB file, ~6.9 GB memory). For 24 GB VRAM use Q4_K_M (12.33 GB file, ~13.2 GB memory). The F16 pack requires ~40.46 GB file and ~41.4 GB memory.
What GPU do you need for maple preview?
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 Q2_0 (6.0 GB file, about 6.9 GB in use).
- 12 GB VRAM: the best quant that fits is Q2_0 (6.0 GB file, about 6.9 GB in use).
- 16 GB VRAM: the best quant that fits is Q4_K_M (12.3 GB file, about 13.2 GB in use).
- 24 GB VRAM: the best quant that fits is Q4_K_M (12.3 GB file, about 13.2 GB in use).
- 32 GB VRAM: the best quant that fits is Q4_K_M (12.3 GB file, about 13.2 GB in use).
- 48 GB VRAM: the best quant that fits is F16 (40.5 GB file, about 41.4 GB in use).
Which maple preview quant should you choose?
Q4_K_M is the usual starting point for maple preview: it balances file size and quality well. Here it is a 12.3 GB file.
If memory is tight, the smallest file is TQ2_0 at 5.5 GB, with the quality trade-offs that come with fewer bits. Lower bit counts mean smaller files and more quality loss.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about maple preview
How much VRAM does maple preview need?
About 13.2 GB at Q4_K_M with an 8K context window. That is the 12.3 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 maple preview on an 8 GB GPU?
Yes, with Q2_0: the 6.0 GB file needs about 6.9 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of maple preview GGUF?
It depends on the quant: TQ2_0 is 5.5 GB, Q2_0 is 6.0 GB, Q4_K_M is 12.3 GB, F16 is 40.5 GB.
What context length does maple preview support?
The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 0.4 GB at 8K and 1.6 GB at 32K.
Where can I download maple preview GGUF?
The files are on Hugging Face in the stamsam/maple-preview-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 maple preview 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 maple‑tq2_0 on an RTX 4000 Ada with 8 GB VRAM?
Yes, maple‑tq2_0 runs at ~97 tok/s on an RTX 4000 Ada with 8 GB VRAM, using the PrismML fork.
Does maple‑preview support vision tasks?
No, maple‑preview is a text‑only model with no vision capabilities.