Local LLM model page

LLaVA 1.6 (13B)

Larger LLaVA multimodal. Better image understanding than 7B. Describe, analyze and discuss images.

Find the best model for my hardware Browse all 183 LLMs

Parameters

13B

Minimum RAM

12 GB

Model size

8 GB

Quantization

Q4_K_M

Can LLaVA 1.6 (13B) run locally?

LLaVA 1.6 (13B) is best suited for mainstream Macs and PCs with 16 GB RAM. LocalClaw recommends Q4_K_M as the default quantization, with at least 12 GB RAM.

Search term for LM Studio or compatible runtimes: llava-v1.6-13b

Hugging Face repository: lmstudio-community/llava-v1.6-vicuna-13b-GGUF

visionpower

Strengths

Better image understanding than 7B
Apache 2.0
Solid multimodal performance

Limitations

Needs 12GB RAM
Limited context
Weak at coding

Best use cases

Image analysis
Visual Q&A
Image captioning
Document understanding

Benchmarks

Speed: 5/10

Quality: 7/10

Coding: 3/10

Reasoning: 6/10

Technical details

Developer: LLaVA Team (UW-Madison / Microsoft)

License: Apache 2.0

Context window: 4,096 tokens

Architecture: CLIP vision encoder + Vicuna 13B language model

Released: 2024-02