LL

Llama-3-70Bx2-MOE

LlamaGGUF by mradermacherUpdated 2024-05-24
Download on Hugging Face
Parameters127B
Context8K
Layers80
Architecturellama
Smallest quant26.7 GB
Quick answer

Llama-3-70Bx2-MOE needs about 40.7 GB of memory at IQ2_XS (a 37.5 GB file) with an 8K context window, so it fits on a 48 GB GPU. This repository has 8 quants, from 26.7 GB (IQ1_S) to 49.1 GB (IQ3_XXS).

Llama-3-70Bx2-MOE is a 127B-parameter Llama model in GGUF format, quantized by mradermacher. Its trained context length is 8K tokens. The repository has 8 quant files, from 26.7 GB (IQ1_S) to 49.1 GB (IQ3_XXS).

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
IQ1_S26.7 GB 29.9 GB
Too big Get file →
IQ1_M29.3 GB 32.5 GB
Too big Get file →
IQ2_XXS33.8 GB 37.0 GB
Too big Get file →
IQ2_XS37.5 GB 40.7 GB
Too big Get file →
IQ2_S38.9 GB 42.1 GB
Too big Get file →
IQ2_M42.4 GB 45.6 GB
Too big Get file →
Q2_K46.8 GB 50.0 GB
Too big Get file →
IQ3_XXS49.1 GB 52.2 GB
Too big Get file →

What GPU do you need for Llama-3-70Bx2-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: no quant fits at an 8K context.
  • 16 GB VRAM: no quant fits at an 8K context.
  • 24 GB VRAM: no quant fits at an 8K context.
  • 32 GB VRAM: the best quant that fits is IQ1_S (26.7 GB file, about 29.9 GB in use).
  • 48 GB VRAM: the best quant that fits is IQ2_M (42.4 GB file, about 45.6 GB in use).

Which Llama-3-70Bx2-MOE quant should you choose?

IQ2_XS (37.5 GB) is the middle option in this repository. 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.

If memory is tight, the smallest file is IQ1_S at 26.7 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
Parameters127B (approximate, mixture of experts)
Layers80
Trained context8,192 tokens
KV cache (fp16)320 KB per token

Source

Quantized bymradermacher
FamilyLlama
Licensellama3
Downloads (HF)96
Last updated2024-05-24

Frequently asked questions about Llama-3-70Bx2-MOE

How much VRAM does Llama-3-70Bx2-MOE need?

About 40.7 GB at IQ2_XS with an 8K context window. That is the 37.5 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 Llama-3-70Bx2-MOE on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 29.9 GB, so you would need a larger GPU or CPU offloading.

What is the file size of Llama-3-70Bx2-MOE GGUF?

It depends on the quant: IQ1_S is 26.7 GB, IQ1_M is 29.3 GB, IQ2_XXS is 33.8 GB, IQ2_XS is 37.5 GB, IQ2_S is 38.9 GB, IQ2_M is 42.4 GB and 2 more.

What context length does Llama-3-70Bx2-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 2.7 GB at 8K and 2.7 GB at 8K.

Where can I download Llama-3-70Bx2-MOE GGUF?

The files are on Hugging Face in the mradermacher/Llama-3-70Bx2-MOE-i1-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 Llama-3-70Bx2-MOE 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.

Similar models

Browse all →

More from mradermacher

All by mradermacher →

More Llama models

All Llama →

Other large models

All large models →