Mark-V3-1.1B needs about 1.4 GB of memory at IQ4_NL (a 0.7 GB file) with an 8K context window, so it fits on a 4 GB GPU. Longer context needs more: at 32K tokens the KV cache alone adds about 0.7 GB.
Mark‑V3‑1.1B: a sparse Transformer with 1.22B parameters and MoE
Mark‑V3‑1.1B is a third‑generation sparse Transformer with 1,222,464,960 total physical parameters and 330,651,072 active parameters per token across 18 layers. It uses Hadamard‑transformed Multi‑Head Latent Attention and Partitioned‑Balancing MoE with 32 routed experts and 2 shared experts. The model is quantized in IQ4_NL, occupying 0.69 GB.
Good at
- Sparse MoE architecture
- Low KV cache memory per token
- Large context of 32,768 tokens
Limits
- No vision support
- AGPL‑3.0 license requires source disclosure
- Requires 8,000‑token context for full performance
Best for
- Low‑memory inference on 8 GB GPUs
- Research on sparse Transformers
- Fine‑tuning with limited VRAM
Who it’s for. Developers and researchers who need efficient inference of a 1.22 B parameter model with large context and low memory footprint. The model is suitable for environments where GPU memory is constrained but a large context window is required.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| IQ4_NL | 0.7 GB | 1.4 GB | Fits | Get file → |
The IQ4_NL quantized file is 0.69 GB and requires approximately 1.4 GB of VRAM at an 8,000‑token context. This fits comfortably on 8 GB GPUs. For tighter budgets, consider lower‑precision quantizations if available, but only the IQ4_NL file is listed.
What GPU do you need for Mark-V3-1.1B?
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 IQ4_NL (0.7 GB file, about 1.4 GB in use).
Which Mark-V3-1.1B quant should you choose?
IQ4_NL (0.7 GB) is the middle option in this repository. The usual sweet spot between file size and quality, and the quant most people start with.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Mark-V3-1.1B
How much VRAM does Mark-V3-1.1B need?
About 1.4 GB at IQ4_NL with an 8K context window. That is the 0.7 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 Mark-V3-1.1B on an 8 GB GPU?
Yes, with IQ4_NL: the 0.7 GB file needs about 1.4 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Mark-V3-1.1B GGUF?
It depends on the quant: IQ4_NL is 0.7 GB.
What context length does Mark-V3-1.1B support?
The GGUF header lists a trained context length of 32,768 tokens (32K). The KV cache grows with the context you use: about 0.2 GB at 8K and 0.7 GB at 32K.
Where can I download Mark-V3-1.1B GGUF?
The files are on Hugging Face in the miceworks/Mark-V3-1.1B repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Mark-V3-1.1B use?
Hugging Face lists the license as agpl-3.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
What is the maximum context length?
The model supports up to 32,768 tokens, as trained with that context size.
What are the license implications?
The model is released under AGPL‑3.0, which requires that any derivative works be released under the same license and source be made available.