LL

models--mergekit-community--Llama3-13B-ku

LlamaGGUF by MaziyarPanahiUpdated 2024-06-16
Download on Hugging Face
Parameters23B
Context8K
Layers42
Architecturellama
Smallest quant8.6 GB
Quick answer

models--mergekit-community--Llama3-13B-ku needs about 15.8 GB of memory at Q4_K_M (a 13.9 GB file) with an 8K context window, so it fits on a 16 GB GPU. This repository has 10 quants, from 8.6 GB (Q2_K) to 24.4 GB (Q8_0).

Llama3-13B-ku: 23B Llama model merged from Meta‑Llama‑3‑120B‑Instruct

Llama3-13B-ku is a 23‑billion‑parameter Llama model created by mergekit‑community. It inherits the Llama‑3 chat template and a default 8K context window. The model is available in multiple GGUF quantizations for various memory budgets.

Good at

  • Creative writing
  • Tool‑using tasks
  • Llama‑3 chat template

Limits

  • No vision support
  • License not specified
  • Context limited to 8K tokens

Best for

  • Story generation
  • Prompt‑based tool execution
  • Chatbot prototypes

Who it’s for. Developers and researchers who need a large Llama model for local inference and creative writing. The model is suitable for users with GPU memory to run the chosen quantization.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
Q2_K8.6 GB 10.6 GB
Too big Get file →
Q3_K_S10.1 GB 12.0 GB
Too big Get file →
Q3_K_M11.2 GB 13.1 GB
Too big Get file →
Q3_K_L12.1 GB 14.0 GB
Too big Get file →
Q4_K_S13.1 GB 15.1 GB
Too big Get file →
Q4_K_M13.9 GB 15.8 GB
Too big Get file →
Q5_K_S15.9 GB 17.8 GB
Too big Get file →
Q5_K_M16.3 GB 18.2 GB
Too big Get file →
Q6_K18.8 GB 20.8 GB
Too big Get file →
Q8_024.4 GB 26.3 GB
Too big Get file →

For 8‑GB VRAM use Q2_K (8.65 GB file, ~10.6 GB VRAM). For 12‑GB VRAM use Q3_K_M (11.15 GB file, ~13.1 GB VRAM). For 16‑GB VRAM use Q4_K_M (13.91 GB file, ~15.8 GB VRAM).

What GPU do you need for models--mergekit-community--Llama3-13B-ku?

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 Q3_K_S (10.1 GB file, about 12.0 GB in use).
  • 16 GB VRAM: the best quant that fits is Q4_K_M (13.9 GB file, about 15.8 GB in use).
  • 24 GB VRAM: the best quant that fits is Q6_K (18.8 GB file, about 20.8 GB in use).
  • 32 GB VRAM: the best quant that fits is Q8_0 (24.4 GB file, about 26.3 GB in use).

Which models--mergekit-community--Llama3-13B-ku quant should you choose?

Q4_K_M is the usual starting point for models--mergekit-community--Llama3-13B-ku: it balances file size and quality well. Here it is a 13.9 GB file.

The Q8_0 file is 24.4 GB, about 1.8 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 8.6 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
Parameters23B
Layers42
Trained context8,192 tokens
KV cache (fp16)168 KB per token

Source

Quantized byMaziyarPanahi
FamilyLlama
Downloads (HF)2,843
Last updated2024-06-16

Frequently asked questions about models--mergekit-community--Llama3-13B-ku

How much VRAM does models--mergekit-community--Llama3-13B-ku need?

About 15.8 GB at Q4_K_M with an 8K context window. That is the 13.9 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 models--mergekit-community--Llama3-13B-ku on an 8 GB GPU?

Not at an 8K context. The smallest file needs about 10.6 GB, so you would need a larger GPU or CPU offloading.

What is the file size of models--mergekit-community--Llama3-13B-ku GGUF?

It depends on the quant: Q2_K is 8.6 GB, Q3_K_S is 10.1 GB, Q3_K_M is 11.2 GB, Q3_K_L is 12.1 GB, Q4_K_S is 13.1 GB, Q4_K_M is 13.9 GB and 4 more.

What context length does models--mergekit-community--Llama3-13B-ku support?

The GGUF header lists a trained context length of 8,192 tokens (8K). The KV cache grows with the context you use: about 1.4 GB at 8K and 1.4 GB at 8K.

Where can I download models--mergekit-community--Llama3-13B-ku GGUF?

The files are on Hugging Face in the MaziyarPanahi/Llama3-13B-ku-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 this model on 8 GB VRAM?

Yes, the Q2_K quantization requires about 10.6 GB VRAM, which fits in an 8 GB GPU with some swap or lower batch size.

Does this model support vision?

No, the model has no vision capabilities.

What context length does it support?

It supports a default context length of 8,192 tokens.

Other GGUFs of Meta-Llama-3-120B-Instruct

Similar models

Browse all →

More from MaziyarPanahi

All by MaziyarPanahi →

More Llama models

All Llama →

Other medium models

All medium models →