v2.2 · 2026-06-12

VISION / MULTIMODAL MODEL — RUNTIME REQUIREMENTS

·CLIP-L vision encoder with MLP projector; raw llama.cpp loading requires the separate mmproj file and Llama-3-instruct chat template.

·RAM and VRAM estimates are derived from GGUF/Ollama package size plus conservative overhead; KV cache, image resolution, batch size, GPU offload and runtime support can materially change memory needs.

LLaVA-Llama-3 8B v1.1

llava-llama3-8b-v1-1

VISION-LANGUAGE

LLaVA-Llama-3 8B v1.1 is a 8.03B vision language model profile for local AI planning. This page records provider, license, context, quantization, hardware-fit estimates, setup hints, source links, and caveats so readers do not choose a model by name alone.

PROVIDER

XTuner

FAMILY

LLaVA / Llama 3

MODEL TYPE

vision-language

PARAMETERS

8.03B

MODALITIES

text, image

ARCHITECTURE

LLaVA model fine-tuned from Meta-Llama-3-8B-Instruct with a CLIP-ViT-Large-patch14-336 vision encoder and MLP multimodal projector.

CONTEXT WINDOW

4,096 tokens

TRAINING TOKENS

—

RELEASE DATE

2024-04-28

LICENSE & LINKS

Meta Llama 3 Community License plus XTuner/LLaVA component noticeslicense text

MODEL CARD

huggingface.co/xtuner/llava-llama-3-8b-v1_1-gguf

HUGGING FACE

xtuner/llava-llama-3-8b-v1_1-gguf

GGUF REPOSITORIES

xtuner/llava-llama-3-8b-v1_1-gguf official

XTuner GGUF repository with separate LLM and mmproj downloads and llama.cpp example command.

SETUP HINTS

OLLAMA

Run `ollama run llava-llama3:8b` for the packaged local model.

LM STUDIO

Use the XTuner GGUF and ensure the UI associates the LLaVA mmproj file with the model.

LLAMA.CPP

Use the XTuner llama.cpp example pattern: load the int4/F16 GGUF with `--mmproj` pointing to the repository's mmproj F16 file and use a 4096-token context.

QUANTIZATION FILE SIZES (GGUF)

Q4_K_M

4.9 GB

Quantization note: Q4_K_M / int4 for local vision chat; higher quantizations were not confirmed in cited sources from the cited file listing.

RAM / VRAM ESTIMATES

MIN RAM

10 GB

COMFORTABLE RAM

16 GB

MIN VRAM

6 GB

COMFORTABLE VRAM

8 GB

These are conservative local-inference estimates, not official hardware requirements.

HARDWARE FIT

CPU only (no GPU)

limited

8 GB RAM

limited

16 GB RAM

yes

32 GB RAM

yes

8 GB VRAM

limited

12 GB VRAM

yes

24 GB VRAM

yes

Apple Silicon (unified memory)

yes

Hardware fit values are conservative local-inference estimates based on GGUF size plus runtime and KV-cache overhead. Actual requirements depend on context length, quantization, runtime, GPU offload, and other running apps.

BEST FOR

·Local Llama-3-based image chat

·Commodity multimodal demos with a newer Llama base than Vicuna LLaVA 1.6

·LLaVA-compatible llama.cpp workflows

·Visual question-answering experiments

AVOID IF

·You need a pure official Meta multimodal model

·You need verified Q5_K_M or Q8_0 file sizes from the model listing

·You need tool calling

·You cannot manage projector/mmproj requirements in manual runtimes

CAVEATS

·Chosen as the best-supported LLaVA/Llama 3 local vision variant because it has a maintained XTuner GGUF repository and an Ollama package.