LL

Llama-3-Magenta-Instruct-4x8B-MoE

LlamaGGUF by mradermacherUpdated 2024-05-16
Download on Hugging Face
Parameters24.9B
Context8K
Layers32
Architecturellama
Smallest quant9.3 GB
Quick answer

Llama-3-Magenta-Instruct-4x8B-MoE needs about 16.7 GB of memory at Q4_K_M (a 15.2 GB file) with an 8K context window, so it fits on a 24 GB GPU. This repository has 14 quants, from 9.3 GB (Q2_K) to 26.5 GB (Q8_0).

Llama‑3‑Magenta‑Instruct‑4x8B‑MoE – a 24.9B parameter mixture‑of‑experts model

Llama‑3‑Magenta‑Instruct‑4x8B‑MoE is a 24.9B parameter model built with a mixture‑of‑experts architecture. It supports a 8192 token context window and is available in multiple GGUF quantizations.

Good at

  • Large‑context instruction following
  • Mixture‑of‑experts inference
  • High‑parameter density

Limits

  • No vision support
  • License not specified
  • Requires large VRAM for higher quants

Best for

  • Chatbots needing long context
  • Research on MoE behavior
  • Fine‑tuning with limited VRAM

Who it’s for. Developers and researchers who need a high‑parameter Llama model with mixture‑of‑experts capability and can manage the VRAM requirements of the chosen quantization.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K9.3 GB 10.9 GB
Too big Get file →
IQ3_XS10.4 GB 12.0 GB
Too big Get file →
Q3_K_S10.9 GB 12.5 GB
Too big Get file →
IQ3_S11.0 GB 12.5 GB
Too big Get file →
IQ3_M11.1 GB 12.7 GB
Too big Get file →
Q3_K_M12.1 GB 13.7 GB
Too big Get file →
Q3_K_L13.0 GB 14.6 GB
Too big Get file →
IQ4_XS13.6 GB 15.2 GB
Too big Get file →
Q4_K_S14.3 GB 15.9 GB
Too big Get file →
Q4_K_M15.2 GB 16.7 GB
Too big Get file →
Q5_K_S17.2 GB 18.8 GB
Too big Get file →
Q5_K_M17.7 GB 19.3 GB
Too big Get file →
Q6_K20.5 GB 22.0 GB
Too big Get file →
Q8_026.5 GB 28.1 GB
Too big Get file →

For 8 GB VRAM, use Q2_K (file size 9.3 GB, estimated 10.9 GB memory). For 12 GB VRAM, Q3_K_S (10.93 GB, 12.5 GB memory) or IQ3_S (10.95 GB, 12.5 GB memory). For 16 GB VRAM, Q4_K_M (15.16 GB, 16.7 GB memory) is suitable. Memory estimates are approximate.

What GPU do you need for Llama-3-Magenta-Instruct-4x8B-MoE?

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 IQ3_XS (10.4 GB file, about 12.0 GB in use).
  • 16 GB VRAM: the best quant that fits is Q4_K_S (14.3 GB file, about 15.9 GB in use).
  • 24 GB VRAM: the best quant that fits is Q6_K (20.5 GB file, about 22.0 GB in use).
  • 32 GB VRAM: the best quant that fits is Q8_0 (26.5 GB file, about 28.1 GB in use).

Which Llama-3-Magenta-Instruct-4x8B-MoE quant should you choose?

Q4_K_M is the usual starting point for Llama-3-Magenta-Instruct-4x8B-MoE: it balances file size and quality well. Here it is a 15.2 GB file.

The Q8_0 file is 26.5 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 9.3 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
Parameters24.9B (approximate, mixture of experts)
Layers32
Trained context8,192 tokens
KV cache (fp16)128 KB per token

Source

Quantized bymradermacher
FamilyLlama
Downloads (HF)1,438
Last updated2024-05-16

Frequently asked questions about Llama-3-Magenta-Instruct-4x8B-MoE

How much VRAM does Llama-3-Magenta-Instruct-4x8B-MoE need?

About 16.7 GB at Q4_K_M with an 8K context window. That is the 15.2 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-Magenta-Instruct-4x8B-MoE on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 10.9 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Llama-3-Magenta-Instruct-4x8B-MoE GGUF?

It depends on the quant: Q2_K is 9.3 GB, IQ3_XS is 10.4 GB, Q3_K_S is 10.9 GB, IQ3_S is 11.0 GB, IQ3_M is 11.1 GB, Q3_K_M is 12.1 GB and 8 more.

What context length does Llama-3-Magenta-Instruct-4x8B-MoE 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 Llama-3-Magenta-Instruct-4x8B-MoE GGUF?

The files are on Hugging Face in the mradermacher/Llama-3-Magenta-Instruct-4x8B-MoE-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 Llama‑3‑Magenta‑Instruct‑4x8B‑MoE on 8 GB VRAM?

Use the Q2_K quant, which has a file size of 9.3 GB and an estimated memory usage of 10.9 GB at 8k context.

What quant should I use for 12 GB VRAM?

Choose Q3_K_S or IQ3_S, each around 10.93–10.95 GB with estimated 12.5 GB memory.

Is there a quant that fits 16 GB VRAM?

Q4_K_M fits, with a 15.16 GB file and estimated 16.7 GB memory at 8k context.

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More Llama models

All Llama →

Other medium models

All medium models →