MI

Hyperion 3.0 Mixtral 3x7B

MistralGGUF by mradermacherUpdated 2024-05-06
Download on Hugging Face
Parameters18.5B
Context32K
Layers32
Architecturellama
Smallest quant7.0 GB
Quick answer

Hyperion 3.0 Mixtral 3x7B needs about 12.9 GB of memory at Q4_K_M (a 11.4 GB file) with an 8K context window, so it fits on a 16 GB GPU. This repository has 14 quants, from 7.0 GB (Q2_K) to 19.8 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 4.3 GB.

18.5B parameter Mixture of Experts model for research

Hyperion 3.0 Mixtral 3x7B is an experimental Mixture of Experts language model with 18.5B parameters and a 32,768 token context window. It is released under Apache‑2.0 and intended for research and experimentation only.

Good at

  • Science and math queries
  • Open‑ended summarization
  • Role‑playing conversations

Limits

  • Experimental prototype, not for production
  • Requires 8k context for memory estimates
  • No vision support

Best for

  • Research prototyping of MoE models
  • Testing gating mechanisms
  • Fine‑tuning expert specialization

Who it’s for. Researchers and developers experimenting with Mixture of Experts language models. Suitable for labs and academic projects.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K7.0 GB 8.5 GB
Too big Get file →
IQ3_XS7.7 GB 9.3 GB
Too big Get file →
Q3_K_S8.2 GB 9.7 GB
Too big Get file →
IQ3_S8.2 GB 9.8 GB
Too big Get file →
IQ3_M8.3 GB 9.9 GB
Too big Get file →
Q3_K_M9.0 GB 10.6 GB
Too big Get file →
Q3_K_L9.8 GB 11.4 GB
Too big Get file →
IQ4_XS10.2 GB 11.7 GB
Too big Get file →
Q4_K_S10.7 GB 12.3 GB
Too big Get file →
Q4_K_M11.4 GB 12.9 GB
Too big Get file →
Q5_K_S12.9 GB 14.5 GB
Too big Get file →
Q5_K_M13.3 GB 14.9 GB
Too big Get file →
Q6_K15.3 GB 16.9 GB
Too big Get file →
Q8_019.8 GB 21.4 GB
Too big Get file →

For 8 GB VRAM, Q2_K (~6.96 GB) uses ~8.5 GB memory at 8k context. For 12 GB VRAM, Q4_K_M (~11.35 GB) uses ~12.9 GB. For 16 GB VRAM, Q6_K (~15.35 GB) uses ~16.9 GB. Estimates are approximate.

What GPU do you need for Hyperion 3.0 Mixtral 3x7B?

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: no quant fits at an 8K context.
  • 6 GB VRAM: no quant fits at an 8K context.
  • 8 GB VRAM: no quant fits at an 8K context.
  • 12 GB VRAM: the best quant that fits is IQ4_XS (10.2 GB file, about 11.7 GB in use).
  • 16 GB VRAM: the best quant that fits is Q5_K_M (13.3 GB file, about 14.9 GB in use).
  • 24 GB VRAM: the best quant that fits is Q8_0 (19.8 GB file, about 21.4 GB in use).

Which Hyperion 3.0 Mixtral 3x7B quant should you choose?

Q4_K_M is the usual starting point for Hyperion 3.0 Mixtral 3x7B: it balances file size and quality well. Here it is a 11.4 GB file.

The Q8_0 file is 19.8 GB, about 1.7 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 Q2_K at 7.0 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
Parameters18.5B (approximate, mixture of experts)
Layers32
Trained context32,768 tokens
KV cache (fp16)128 KB per token

Source

Quantized bymradermacher
FamilyMistral
Licenseapache-2.0
Downloads (HF)1,157
Last updated2024-05-06

Frequently asked questions about Hyperion 3.0 Mixtral 3x7B

How much VRAM does Hyperion 3.0 Mixtral 3x7B need?

About 12.9 GB at Q4_K_M with an 8K context window. That is the 11.4 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 Hyperion 3.0 Mixtral 3x7B on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 8.5 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Hyperion 3.0 Mixtral 3x7B GGUF?

It depends on the quant: Q2_K is 7.0 GB, IQ3_XS is 7.7 GB, Q3_K_S is 8.2 GB, IQ3_S is 8.2 GB, IQ3_M is 8.3 GB, Q3_K_M is 9.0 GB and 8 more.

What context length does Hyperion 3.0 Mixtral 3x7B support?

The GGUF header lists a trained context length of 32,768 tokens (32K). The KV cache grows with the context you use: about 1.1 GB at 8K and 4.3 GB at 32K.

Where can I download Hyperion 3.0 Mixtral 3x7B GGUF?

The files are on Hugging Face in the mradermacher/Hyperion-3.0-Mixtral-3x7B-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 Hyperion 3.0 Mixtral 3x7B 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?

The smallest quant Q2_K requires ~8.5 GB memory at 8k context, so 8 GB VRAM is insufficient.

Does this model support vision?

No, it is a text‑only model.

Is it licensed for commercial use?

It is under Apache‑2.0, which allows commercial use with attribution.

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More Mistral models

All Mistral →

Other medium models

All medium models →