LL

llava 1.6

LLaVAGGUF by cmp-nctUpdated 2024-02-21
Download on Hugging Face
Parameters7.24B
Context32K
Layers32
Architecturellama
Smallest quant4.4 GB
Quick answer

llava 1.6 needs about 10.8 GB of memory at Q5_K (a 9.2 GB file) with an 8K context window, so it fits on a 12 GB GPU. This repository has 7 quants, from 4.4 GB to 24.3 GB (F16). Longer context needs more: at 32K tokens the KV cache alone adds about 4.3 GB.

LLaVA 1.6 – 7.24B multimodal model with vision support

LLaVA 1.6 is a 7.24B parameter multimodal model built on the llama architecture. It supports vision tasks and is available in several GGUF quantizations. The model is licensed under Apache‑2.0 and requires a 32K token context window.

Good at

  • Vision‑enabled text generation
  • Image‑to‑text tasks
  • Multimodal reasoning

Limits

  • Requires 32K context tokens
  • Large memory footprint for higher quants
  • No explicit language support list

Best for

  • Image captioning
  • Visual question answering
  • Multimodal chat

Who it’s for. Developers needing a lightweight vision‑enabled LLM for local inference. The model is suitable for applications that require image understanding combined with text generation.

Quants & memory

QuantFileEst. memoryvs VRAMResultDownload
—4.4 GB 5.9 GB
Fits Get file →
Q5_K4.8 GB 6.4 GB
Fits Get file →
—5.1 GB 6.7 GB
Fits Get file →
Q5_K9.2 GB 10.8 GB
Too big Get file →
Q6_K10.7 GB 12.3 GB
Too big Get file →
F1616.7 GB 18.2 GB
Too big Get file →
F1624.3 GB 25.9 GB
Too big Get file →

For 8 GB VRAM use Q5_K (4.78 GB file, 6.4 GB memory). For 12 GB VRAM use Q6_K (10.68 GB file, 12.3 GB memory). Higher precision F16 variants need 16–24 GB VRAM.

What GPU do you need for llava 1.6?

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: the best quant that fits is the main file (4.4 GB file, about 5.9 GB in use).
  • 8 GB VRAM: the best quant that fits is the main file (5.1 GB file, about 6.7 GB in use).
  • 12 GB VRAM: the best quant that fits is Q5_K (9.2 GB file, about 10.8 GB in use).
  • 16 GB VRAM: the best quant that fits is Q6_K (10.7 GB file, about 12.3 GB in use).
  • 24 GB VRAM: the best quant that fits is F16 (16.7 GB file, about 18.2 GB in use).
  • 32 GB VRAM: the best quant that fits is F16 (24.3 GB file, about 25.9 GB in use).

Which llava 1.6 quant should you choose?

Q5_K (9.2 GB) is the middle option in this repository. Closer to the original model than 4-bit, for a moderate increase in size. A good pick when you have memory to spare.

If memory is tight, the smallest file is the first one in the table at 4.4 GB, with the quality trade-offs that come with fewer bits.

New to quant names? Read GGUF quantization explained.

Specifications

Architecture from the GGUF header

Architecturellama
Parameters7.24B
Layers32
Trained context32,768 tokens
KV cache (fp16)128 KB per token

Source

Quantized bycmp-nct
FamilyLLaVA
Licenseapache-2.0
Downloads (HF)3,735
Last updated2024-02-21

Frequently asked questions about llava 1.6

How much VRAM does llava 1.6 need?

About 10.8 GB at Q5_K with an 8K context window. That is the 9.2 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 llava 1.6 on an 8 GB GPU?

Yes, with the smaller files: the 5.1 GB file needs about 6.7 GB at 8K context. Larger quants in this repository do not fit in 8 GB.

What is the file size of llava 1.6 GGUF?

It depends on the quant: file is 4.4 GB, Q5_K is 4.8 GB, file is 5.1 GB, Q5_K is 9.2 GB, Q6_K is 10.7 GB, F16 is 16.7 GB and 1 more.

What context length does llava 1.6 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 llava 1.6 GGUF?

The files are on Hugging Face in the cmp-nct/llava-1.6-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 llava 1.6 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 LLaVA 1.6 on 8 GB VRAM?

Yes, the Q5_K quantization uses 6.4 GB estimated memory at 8K context, fitting within 8 GB VRAM.

Does LLaVA 1.6 support text only?

It is primarily a vision‑enabled model; text‑only usage is possible but the model is optimized for multimodal inputs.

Similar models

Browse all →

More from cmp-nct

All by cmp-nct →

More LLaVA models

All LLaVA →

Other small models

All small models →