Mizan v3 needs about 3.1 GB of memory at Q4_K_M (a 1.9 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 2 quants, from 1.9 GB (Q4_K_M) to 3.1 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 2.7 GB.
Mizan v3 – a 3B Persian‑first decision model
Mizan v3 is a 2.9B parameter decision model that returns calibrated probabilities for typed questions. It runs in a single forward pass and follows user‑defined rules. The model is text‑only and licensed under Apache‑2.0.
Good at
- Calibrated probability scoring
- Fast inference (≈55 ms on entry‑level GPU)
- Supports Persian and English questions
Limits
- No vision support – use v2 for images
- Requires 8 k context for memory estimates
- No text generation, only decision output
Best for
- Decision‑making APIs
- Rule‑based routing systems
- Calibrated scoring of structured records
Who it’s for. Developers building decision engines that need fast, calibrated outputs for Persian or English records. Suitable for systems that require a Jev/SystemOne‑compatible endpoint and commercial use.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_M | 1.9 GB | 3.1 GB | Fits | Get file → | |
| Q8_0 | 3.1 GB | 4.3 GB | Fits | Get file → |
The Q4_K_M quant (1.93 GB) uses about 3.1 GB of VRAM at 8 k context and is suitable for 8 GB GPUs or CPUs. The Q8_0 quant (3.09 GB) uses about 4.3 GB of VRAM and offers slightly higher accuracy. Choose Q4 for lower memory or Q8 when accuracy is critical.
What GPU do you need for Mizan v3?
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 Q4_K_M (1.9 GB file, about 3.1 GB in use).
- 6 GB VRAM: the best quant that fits is Q8_0 (3.1 GB file, about 4.3 GB in use).
Which Mizan v3 quant should you choose?
Q4_K_M is the usual starting point for Mizan v3: it balances file size and quality well. Here it is a 1.9 GB file.
The Q8_0 file is 3.1 GB, about 1.6 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Mizan v3
How much VRAM does Mizan v3 need?
About 3.1 GB at Q4_K_M with an 8K context window. That is the 1.9 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 Mizan v3 on an 8 GB GPU?
Yes, with Q8_0: the 3.1 GB file needs about 4.3 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Mizan v3 GGUF?
It depends on the quant: Q4_K_M is 1.9 GB, Q8_0 is 3.1 GB.
What context length does Mizan v3 support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 0.7 GB at 8K and 2.7 GB at 32K.
Where can I download Mizan v3 GGUF?
The files are on Hugging Face in the Nimruz/Mizan-v3-3B-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 Mizan v3 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.
Does Mizan v3 support image inputs?
No, Mizan v3 is text‑only. Use Mizan‑v2‑4B‑GGUF for image support.