DeepHermes 3 Llama 3 3B Preview needs about 3.7 GB of memory at its main file (a 2.3 GB file) with an 8K context window, so it fits on a 4 GB GPU. This repository has 4 quants, from 1.9 GB to 3.4 GB. Longer context needs more: at 32K tokens the KV cache alone adds about 3.8 GB.
DeepHermes 3 Llama 3 3B Preview – a reasoning‑enabled LLM
DeepHermes 3 Preview is a 3.21B parameter Llama‑3 model that supports reasoning, roleplay, and tools. It is a preview version with early reasoning capabilities and improved annotation, judgement, and function calling.
Good at
- Reasoning with long chains of thought
- Roleplay and agentic tasks
- Function calling and tool use
Limits
- No vision support
- Requires 8k context for memory estimates
- License is llama3
Best for
- Chatbots needing reasoning
- Interactive roleplay applications
- Tool‑enabled assistants
Who it’s for. Developers and researchers who need a lightweight Llama‑3 model with built‑in reasoning and tool‑calling features. The preview version is suitable for testing new prompting strategies and evaluating early reasoning performance.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| — | 1.9 GB | 3.4 GB | Fits | Get file → | |
| — | 2.3 GB | 3.7 GB | Fits | Get file → | |
| — | 2.6 GB | 4.1 GB | Fits | Get file → | |
| — | 3.4 GB | 4.9 GB | Fits | Get file → |
For 8 GB VRAM, use the q4 file (1.8 GB) which estimates 3.4 GB memory at 8k context. For 12 GB VRAM, q5 (2.2 GB) or q6 (2.5 GB) are suitable, estimating 3.7–4.1 GB memory. For 16 GB VRAM, q8 (3.2 GB) provides the largest model with 4.9 GB memory estimate.
What GPU do you need for DeepHermes 3 Llama 3 3B Preview?
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 the main file (2.3 GB file, about 3.7 GB in use).
- 6 GB VRAM: the best quant that fits is the main file (3.4 GB file, about 4.9 GB in use).
Which DeepHermes 3 Llama 3 3B Preview quant should you choose?
If memory is tight, the smallest file is the first one in the table at 1.9 GB, with the quality trade-offs that come with fewer bits.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Frequently asked questions about DeepHermes 3 Llama 3 3B Preview
How much VRAM does DeepHermes 3 Llama 3 3B Preview need?
About 3.7 GB at its main file with an 8K context window. That is the 2.3 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 DeepHermes 3 Llama 3 3B Preview on an 8 GB GPU?
Yes, with the smaller files: the 3.4 GB file needs about 4.9 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of DeepHermes 3 Llama 3 3B Preview GGUF?
It depends on the quant: file is 1.9 GB, file is 2.3 GB, file is 2.6 GB, file is 3.4 GB.
What context length does DeepHermes 3 Llama 3 3B Preview 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 DeepHermes 3 Llama 3 3B Preview GGUF?
The files are on Hugging Face in the NousResearch/DeepHermes-3-Llama-3-3B-Preview-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 DeepHermes 3 Llama 3 3B Preview use?
Hugging Face lists the license as llama3. 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 DeepHermes 3 on 8 GB VRAM?
Yes, the q4 quantized model is 1.8 GB and estimates 3.4 GB memory at 8k context, fitting within 8 GB VRAM.
Does this model support vision tasks?
No, DeepHermes 3 Preview does not include vision capabilities.
What context length does it support?
It was trained on 131072 tokens and can handle up to that context length.