LL

Llama 3.2 4X3B MOE Ultra Instruct 10B

LlamaGGUF by mradermacherUpdated 2025-05-29
Download on Hugging Face
Parameters9.95B
Context128K
Layers28
Architecturellama
Smallest quant2.3 GB
Quick answer

Llama 3.2 4X3B MOE Ultra Instruct 10B needs about 7.5 GB of memory at Q4_K_M (a 6.1 GB file) with an 8K context window, so it fits on a 8 GB GPU. This repository has 22 quants, from 2.3 GB (IQ1_S) to 8.2 GB (Q6_K). Longer context needs more: at 32K tokens the KV cache alone adds about 3.8 GB.

Llama 3.2 4X3B MOE Ultra Instruct 10B – 9.95B parameter model

Llama 3.2 4X3B MOE Ultra Instruct 10B is a mixture of experts model with 9.95B parameters. It supports a context length of 131072 tokens and is optimized for instruction following. The model is available in multiple GGUF quantizations.

Good at

  • Instruction following
  • Tool usage
  • Large context handling

Limits

  • No vision support
  • License not specified
  • Requires high VRAM for larger quants

Best for

  • Chatbots needing long context
  • Tool‑integrated assistants
  • Research prototyping

Who it’s for. Developers and researchers who need a large‑context instruction model and can run it on GPU hardware with sufficient VRAM. The model is suitable for building chat applications that require tool integration and extended dialogue history.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ1_S2.3 GB 3.7 GB
Fits Get file →
IQ1_M2.5 GB 3.9 GB
Fits Get file →
IQ2_XXS2.8 GB 4.3 GB
Fits Get file →
IQ2_XS3.1 GB 4.5 GB
Fits Get file →
IQ2_S3.2 GB 4.6 GB
Fits Get file →
IQ2_M3.5 GB 4.9 GB
Fits Get file →
Q2_K_S3.5 GB 5.0 GB
Fits Get file →
Q2_K3.8 GB 5.2 GB
Fits Get file →
IQ3_XXS3.9 GB 5.4 GB
Fits Get file →
IQ3_XS4.2 GB 5.7 GB
Fits Get file →
IQ3_S4.4 GB 5.9 GB
Fits Get file →
Q3_K_S4.4 GB 5.9 GB
Fits Get file →
IQ3_M4.5 GB 6.0 GB
Fits Get file →
Q3_K_M4.9 GB 6.3 GB
Fits Get file →
Q3_K_L5.3 GB 6.7 GB
Fits Get file →
IQ4_XS5.4 GB 6.8 GB
Fits Get file →
Q4_05.7 GB 7.2 GB
Fits Get file →
Q4_K_S5.7 GB 7.2 GB
Fits Get file →
Q4_K_M6.1 GB 7.5 GB
Tight Get file →
Q5_K_S6.9 GB 8.3 GB
Too big Get file →
Q5_K_M7.1 GB 8.5 GB
Too big Get file →
Q6_K8.2 GB 9.6 GB
Too big Get file →

For 8 GB VRAM, use IQ1_S (file size 2.29 GB, memory 3.7 GB) or IQ1_M (2.49 GB, 3.9 GB). For 12 GB VRAM, IQ2_S (3.2 GB, 4.6 GB) or Q2_K_S (3.54 GB, 5 GB) are suitable. For 16 GB VRAM, Q4_K_M (6.08 GB, 7.5 GB) or Q5_K_S (6.9 GB, 8.3 GB) provide higher quality.

What GPU do you need for Llama 3.2 4X3B MOE Ultra Instruct 10B?

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 IQ1_M (2.5 GB file, about 3.9 GB in use).
  • 6 GB VRAM: the best quant that fits is IQ3_M (4.5 GB file, about 6.0 GB in use).
  • 8 GB VRAM: the best quant that fits is Q4_K_M (6.1 GB file, about 7.5 GB in use).
  • 12 GB VRAM: the best quant that fits is Q6_K (8.2 GB file, about 9.6 GB in use).

Which Llama 3.2 4X3B MOE Ultra Instruct 10B quant should you choose?

Q4_K_M is the usual starting point for Llama 3.2 4X3B MOE Ultra Instruct 10B: it balances file size and quality well. Here it is a 6.1 GB file.

If memory is tight, the smallest file is IQ1_S at 2.3 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

Architecturellama
Parameters9.95B (approximate, mixture of experts)
Layers28
Trained context131,072 tokens
KV cache (fp16)112 KB per token

Source

Quantized bymradermacher
FamilyLlama
Downloads (HF)1,187
Last updated2025-05-29

Frequently asked questions about Llama 3.2 4X3B MOE Ultra Instruct 10B

How much VRAM does Llama 3.2 4X3B MOE Ultra Instruct 10B need?

About 7.5 GB at Q4_K_M with an 8K context window. That is the 6.1 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 Llama 3.2 4X3B MOE Ultra Instruct 10B on an 8 GB GPU?

Yes, with Q4_K_M: the 6.1 GB file needs about 7.5 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of Llama 3.2 4X3B MOE Ultra Instruct 10B GGUF?

It depends on the quant: IQ1_S is 2.3 GB, IQ1_M is 2.5 GB, IQ2_XXS is 2.8 GB, IQ2_XS is 3.1 GB, IQ2_S is 3.2 GB, IQ2_M is 3.5 GB and 16 more.

What context length does Llama 3.2 4X3B MOE Ultra Instruct 10B support?

The GGUF header lists a trained context length of 131,072 tokens (128K). The KV cache grows with the context you use: about 0.9 GB at 8K and 3.8 GB at 32K.

Where can I download Llama 3.2 4X3B MOE Ultra Instruct 10B GGUF?

The files are on Hugging Face in the mradermacher/Llama-3.2-4X3B-MOE-Ultra-Instruct-10B-i1-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.

Can I run this model on 8 GB VRAM?

Yes, using IQ1_S or IQ1_M quantizations, which require about 3.7–3.9 GB of VRAM at 8k context.

Does this model support vision?

No, the model does not have vision capabilities.

What is the context length of this model?

The model supports a context length of 131072 tokens.

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More Llama models

All Llama →

Other medium models

All medium models →