DR

Draco 8x7B

DracoGGUF by mradermacherUpdated 2024-05-06
Download on Hugging Face
Parameters46.7B
Context8K
Layers32
Architecturellama
Smallest quant17.5 GB
Quick answer

Draco 8x7B needs about 30.2 GB of memory at Q4_K_M (a 28.6 GB file) with an 8K context window, so it fits on a 32 GB GPU. This repository has 13 quants, from 17.5 GB (Q2_K) to 38.5 GB (Q6_K).

Draco 8x7B – a 46.7B parameter mixture‑of‑experts Llama model

Draco 8x7B is a 46.7B parameter mixture‑of‑experts model based on the Llama architecture. It supports a context length of 8192 tokens and is released under the apache‑2.0 license. The model is optimized for tool usage and is available in multiple GGUF quantizations.

Good at

  • Tool‑enabled reasoning
  • Large context handling
  • Mixture‑of‑experts efficiency

Limits

  • No vision support
  • Requires high VRAM for full precision
  • License is apache‑2.0

Best for

  • Chatbots needing 8k context
  • Tool‑integrated assistants
  • Research on MOE models

Who it’s for. Developers and researchers who need a large‑scale Llama model with tool support and a flexible mixture‑of‑experts architecture. The model is suitable for those who can manage the memory requirements and want to experiment with different quantizations.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K17.5 GB 19.0 GB
Too big Get file →
IQ3_XS19.4 GB 21.0 GB
Too big Get file →
IQ3_S20.6 GB 22.2 GB
Too big Get file →
Q3_K_S20.6 GB 22.2 GB
Too big Get file →
IQ3_M21.6 GB 23.2 GB
Too big Get file →
Q3_K_M22.7 GB 24.3 GB
Too big Get file →
Q3_K_L24.3 GB 25.9 GB
Too big Get file →
IQ4_XS25.5 GB 27.1 GB
Too big Get file →
Q4_K_S26.9 GB 28.5 GB
Too big Get file →
Q4_K_M28.6 GB 30.2 GB
Too big Get file →
Q5_K_S32.4 GB 34.0 GB
Too big Get file →
Q5_K_M33.4 GB 35.0 GB
Too big Get file →
Q6_K38.5 GB 40.1 GB
Too big Get file →

For a 8 GB VRAM budget, the Q2_K quant (file size 17.46 GB, estimated 19 GB memory at 8k context) is the most suitable. If 12 GB VRAM is available, Q3_K_S (20.59 GB, 22.2 GB memory) offers a good balance of size and performance. For 16 GB VRAM, Q4_K_M (28.6 GB, 30.2 GB memory) provides higher quality while staying within limits.

What GPU do you need for Draco 8x7B?

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: no quant fits at an 8K context.
  • 16 GB VRAM: no quant fits at an 8K context.
  • 24 GB VRAM: the best quant that fits is IQ3_M (21.6 GB file, about 23.2 GB in use).
  • 32 GB VRAM: the best quant that fits is Q4_K_M (28.6 GB file, about 30.2 GB in use).
  • 48 GB VRAM: the best quant that fits is Q6_K (38.5 GB file, about 40.1 GB in use).

Which Draco 8x7B quant should you choose?

Q4_K_M is the usual starting point for Draco 8x7B: it balances file size and quality well. Here it is a 28.6 GB file.

If memory is tight, the smallest file is Q2_K at 17.5 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
Parameters46.7B (approximate, mixture of experts)
Layers32
Trained context8,192 tokens
KV cache (fp16)128 KB per token

Source

Quantized bymradermacher
FamilyDraco
Licenseapache-2.0
Downloads (HF)2,111
Last updated2024-05-06

Frequently asked questions about Draco 8x7B

How much VRAM does Draco 8x7B need?

About 30.2 GB at Q4_K_M with an 8K context window. That is the 28.6 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 Draco 8x7B on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 19.0 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Draco 8x7B GGUF?

It depends on the quant: Q2_K is 17.5 GB, IQ3_XS is 19.4 GB, IQ3_S is 20.6 GB, Q3_K_S is 20.6 GB, IQ3_M is 21.6 GB, Q3_K_M is 22.7 GB and 7 more.

What context length does Draco 8x7B support?

The GGUF header lists a trained context length of 8,192 tokens (8K). The KV cache grows with the context you use: about 1.1 GB at 8K and 1.1 GB at 8K.

Where can I download Draco 8x7B GGUF?

The files are on Hugging Face in the mradermacher/Draco-8x7B-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 Draco 8x7B 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 Draco 8x7B on 8 GB VRAM?

Only the Q2_K quant is likely to fit, as it estimates 19 GB memory at 8k context. 8 GB VRAM is insufficient for any other quant.

Does Draco 8x7B support vision tasks?

No, the model has no vision capabilities.

What license applies to Draco 8x7B?

The model is released under the apache‑2.0 license.

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More Draco models

All Draco →

Other large models

All large models →