Marco Nano Instruct 1-bit (ternary, TQ1_0) needs about 4.6 GB of memory at IQ2_XXS (a 3.1 GB file) with an 8K context window, so it fits on a 6 GB GPU. This repository has 5 quants, from 2.2 GB (TQ1_0) to 4.0 GB (IQ3_S). Longer context needs more: at 32K tokens the KV cache alone adds about 3.8 GB.
Marco Nano Instruct 1-bit (ternary, TQ1_0) – 8B sparse MoE model
Marco Nano Instruct is an 8B parameter multilingual mixture‑of‑experts model with only 0.6B active per token. It is post‑trained on instruction data and distilled into low‑bit GGUF files. The model supports tool usage and runs in llama.cpp.
Good at
- Low‑bit inference on mobile GPUs
- Sparse MoE efficiency
- Tool‑aware instruction following
Limits
- No vision support
- Requires at least 8k context for memory estimates
- Apache‑2.0 license
Best for
- Mobile LLM inference
- Low‑memory instruction tasks
- Tool‑enabled chatbots
Who it’s for. Developers needing a lightweight multilingual LLM for mobile or edge devices. The model is suitable for applications that can tolerate a 0.6B active parameter budget and benefit from sparse MoE efficiency.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| TQ1_0 | 2.2 GB | 3.6 GB | Fits | Get file → | |
| Q2_0 | 2.5 GB | 4.0 GB | Fits | Get file → | |
| IQ2_XXS | 3.1 GB | 4.6 GB | Fits | Get file → | |
| IQ2_S | 3.5 GB | 4.9 GB | Fits | Get file → | |
| IQ3_S | 4.0 GB | 5.5 GB | Fits | Get file → |
For 8 GB VRAM use the TQ1_0 file (2.18 GB, 3.6 GB memory at 8k context). If more VRAM is available, Q2_0 (2.54 GB, 4 GB) or IQ2_XXS (3.14 GB, 4.6 GB) can be used. IQ2_S and IQ3_S provide higher quality at 3.46 GB/4.9 GB and 4.02 GB/5.5 GB respectively.
What GPU do you need for Marco Nano Instruct 1-bit (ternary, TQ1_0)?
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: the best quant that fits is Q2_0 (2.5 GB file, about 4.0 GB in use).
- 6 GB VRAM: the best quant that fits is IQ3_S (4.0 GB file, about 5.5 GB in use).
Which Marco Nano Instruct 1-bit (ternary, TQ1_0) quant should you choose?
IQ2_XXS (3.1 GB) is the middle option in this repository. 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.
If memory is tight, the smallest file is TQ1_0 at 2.2 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 Marco Nano Instruct 1-bit (ternary, TQ1_0)
How much VRAM does Marco Nano Instruct 1-bit (ternary, TQ1_0) need?
About 4.6 GB at IQ2_XXS with an 8K context window. That is the 3.1 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 Marco Nano Instruct 1-bit (ternary, TQ1_0) on an 8 GB GPU?
Yes, with IQ3_S: the 4.0 GB file needs about 5.5 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Marco Nano Instruct 1-bit (ternary, TQ1_0) GGUF?
It depends on the quant: TQ1_0 is 2.2 GB, Q2_0 is 2.5 GB, IQ2_XXS is 3.1 GB, IQ2_S is 3.5 GB, IQ3_S is 4.0 GB.
What context length does Marco Nano Instruct 1-bit (ternary, TQ1_0) support?
The GGUF header lists a trained context length of 32,768 tokens (32K). The KV cache grows with the context you use: about 0.9 GB at 8K and 3.8 GB at 32K.
Where can I download Marco Nano Instruct 1-bit (ternary, TQ1_0) GGUF?
The files are on Hugging Face in the RegalInvader/Marco-Nano-Instruct-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 Marco Nano Instruct 1-bit (ternary, TQ1_0) use?
Hugging Face lists the license as apache-2.0. 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 Marco Nano Instruct on 8 GB VRAM?
Yes, the TQ1_0 quantization requires about 3.6 GB memory at 8k context, well within 8 GB VRAM.
Does the model support vision?
No, the model has no vision capabilities.
What license applies to this model?
The model is released under the apache-2.0 license.