LL

Llama3.2 DeepHermes 3 3B Preview Reasoning MAX NEO Imatrix

LlamaGGUF by DavidAUUpdated 2025-03-17
Download on Hugging Face
Parameters3.21B
Context128K
Layers28
Architecturellama
Smallest quant1.4 GB
Quick answer

Llama3.2 DeepHermes 3 3B Preview Reasoning MAX NEO Imatrix needs about 3.9 GB of memory at Q4_K_M (a 2.5 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 25 quants, from 1.4 GB (IQ1_S) to 6.4 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 3.8 GB.

Llama3.2 DeepHermes 3 3B Preview Reasoning MAX NEO Imatrix – a 3.21B parameter hybrid reasoning model

Llama3.2 DeepHermes 3 3B Preview Reasoning MAX NEO Imatrix is a 3.21B parameter Llama model that supports long‑chain reasoning and normal LLM response modes. It is released under the apache‑2.0 license and can be run with llama.cpp or vLLM. The model uses a 128k token context window.

Good at

  • Long‑chain reasoning
  • Role‑play and instruction following
  • Tool use and function calling

Limits

  • Requires at least 8k context window
  • No vision support
  • Large memory for higher quants

Best for

  • Complex problem solving
  • Interactive chat agents
  • Script generation

Who it’s for. Developers and researchers who need a compact reasoning model for local inference. The model is suitable for applications that require both intuitive responses and deep reasoning within a single inference pipeline.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ1_S1.4 GB 2.8 GB
Fits Get file →
IQ1_M1.4 GB 2.9 GB
Fits Get file →
IQ2_XXS1.5 GB 3.0 GB
Fits Get file →
IQ2_XS1.6 GB 3.1 GB
Fits Get file →
IQ2_S1.7 GB 3.1 GB
Fits Get file →
IQ2_M1.7 GB 3.2 GB
Fits Get file →
Q2_K1.8 GB 3.3 GB
Fits Get file →
IQ3_XXS1.9 GB 3.3 GB
Fits Get file →
IQ3_XS1.9 GB 3.4 GB
Fits Get file →
IQ3_S2.0 GB 3.4 GB
Fits Get file →
Q3_K_S2.0 GB 3.4 GB
Fits Get file →
IQ3_M2.1 GB 3.5 GB
Fits Get file →
Q3_K_M2.2 GB 3.6 GB
Fits Get file →
Q3_K_L2.3 GB 3.7 GB
Fits Get file →
IQ4_XS2.3 GB 3.7 GB
Fits Get file →
IQ4_NL2.4 GB 3.8 GB
Fits Get file →
Q4_02.4 GB 3.8 GB
Fits Get file →
Q4_K_S2.4 GB 3.8 GB
Fits Get file →
Q4_K_M2.5 GB 3.9 GB
Fits Get file →
Q5_K_S2.7 GB 4.2 GB
Fits Get file →
Q5_02.7 GB 4.2 GB
Fits Get file →
Q5_K_M2.8 GB 4.2 GB
Fits Get file →
Q6_K3.1 GB 4.5 GB
Fits Get file →
Q8_03.8 GB 5.2 GB
Fits Get file →
F166.4 GB 7.9 GB
Tight Get file →

For 8–12 GB VRAM, use IQ4_XS (file size 2.29 GB, memory 3.7 GB at 8k context) or Q4_0 (2.39 GB, 3.8 GB). If 16 GB VRAM is available, IQ4_NL (2.38 GB, 3.8 GB) or Q4_K_M (2.48 GB, 3.9 GB) give better performance. For 24 GB+ VRAM, Q5_K_M (2.79 GB, 4.2 GB) or Q6_K (3.11 GB, 4.5 GB) are recommended.

What GPU do you need for Llama3.2 DeepHermes 3 3B Preview Reasoning MAX NEO Imatrix?

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 (2.5 GB file, about 3.9 GB in use).
  • 6 GB VRAM: the best quant that fits is Q8_0 (3.8 GB file, about 5.2 GB in use).
  • 8 GB VRAM: the best quant that fits is F16 (6.4 GB file, about 7.9 GB in use).

Which Llama3.2 DeepHermes 3 3B Preview Reasoning MAX NEO Imatrix quant should you choose?

Q4_K_M is the usual starting point for Llama3.2 DeepHermes 3 3B Preview Reasoning MAX NEO Imatrix: it balances file size and quality well. Here it is a 2.5 GB file.

The Q8_0 file is 3.8 GB, about 1.5 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.

If memory is tight, the smallest file is IQ1_S at 1.4 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
Parameters3.21B
Layers28
Trained context131,072 tokens
KV cache (fp16)112 KB per token

Source

Quantized byDavidAU
FamilyLlama
Licenseapache-2.0
Downloads (HF)3,398
Last updated2025-03-17

Frequently asked questions about Llama3.2 DeepHermes 3 3B Preview Reasoning MAX NEO Imatrix

How much VRAM does Llama3.2 DeepHermes 3 3B Preview Reasoning MAX NEO Imatrix need?

About 3.9 GB at Q4_K_M with an 8K context window. That is the 2.5 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 Llama3.2 DeepHermes 3 3B Preview Reasoning MAX NEO Imatrix on an 8 GB GPU?

Yes, with F16: the 6.4 GB file needs about 7.9 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of Llama3.2 DeepHermes 3 3B Preview Reasoning MAX NEO Imatrix GGUF?

It depends on the quant: IQ1_S is 1.4 GB, IQ1_M is 1.4 GB, IQ2_XXS is 1.5 GB, IQ2_XS is 1.6 GB, IQ2_S is 1.7 GB, IQ2_M is 1.7 GB and 19 more.

What context length does Llama3.2 DeepHermes 3 3B Preview Reasoning MAX NEO Imatrix 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 Llama3.2 DeepHermes 3 3B Preview Reasoning MAX NEO Imatrix GGUF?

The files are on Hugging Face in the DavidAU/Llama3.2-DeepHermes-3-3B-Preview-Reasoning-MAX-NEO-Imatrix-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 Llama3.2 DeepHermes 3 3B Preview Reasoning MAX NEO Imatrix 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 this model on 8 GB VRAM?

Yes, the IQ4_XS quant uses about 3.7 GB of VRAM at an 8k context window, which fits within an 8 GB GPU with headroom for other processes.

Does the model support vision tasks?

No, the model has no vision capability. It is a text‑only LLM.

What context length does the model support?

The model is trained with a 131072 token context window, but a minimum of 8k context is recommended for stable inference.

Other GGUFs of DeepHermes-3-Llama-3-3B-Preview

Similar models

Browse all →

More from DavidAU

All by DavidAU →

More Llama models

All Llama →

Other tiny models

All tiny models →