CoolFace
Modelpublic

second-state/Llava-v1.5-7B-GGUF

sourceHugging Facellama2updated 2y agoView on Hugging Face
20likes3.1kdownloads
Model Card

<!-- header start --> <!-- 200823 --> <div style="width: auto; margin-left: auto; margin-right: auto"> <img src="https://github.com/LlamaEdge/LlamaEdge/raw/dev/assets/logo.svg" style="width: 100%; min-width: 400px; display: block; margin: auto;"> </div> <hr style="margin-top: 1.0em; margin-bottom: 1.0em;"> <!-- header end -->

Llava-v1.5-7B-GGUF

Original Model

liuhaotian/llava-v1.5-7b

Run with LlamaEdge

  • —Prompt template
  • —Prompt type: vicuna-llava
  • —Prompt string
text
    <system_prompt>\nUSER:<image_embeddings>\n<textual_prompt>\nASSISTANT:
  • —Context size: 4096
  • —Run as LlamaEdge service
bash
  wasmedge --dir .:. \
    --nn-preload default:GGML:AUTO:llava-v1.5-7b-Q5_K_M.gguf \
    llama-api-server.wasm \
    --prompt-template vicuna-llava \
    --ctx-size 4096 \
    --llava-mmproj llava-v1.5-7b-mmproj-model-f16.gguf \
    --model-name llava-v1.5

Quantized GGUF Models

NameQuant methodBitsSizeUse case
llava-v1.5-7b-Q2_K.ggufQ2_K22.53 GBsmallest, significant quality loss - not recommended for most purposes
llava-v1.5-7b-Q3_K_L.ggufQ3KL33.6 GBsmall, substantial quality loss
llava-v1.5-7b-Q3_K_M.ggufQ3KM33.3 GBvery small, high quality loss
llava-v1.5-7b-Q3_K_S.ggufQ3KS32.95 GBvery small, high quality loss
llava-v1.5-7b-Q4_0.ggufQ4_043.83 GBlegacy; small, very high quality loss - prefer using Q3KM
llava-v1.5-7b-Q4_K_M.ggufQ4KM44.08 GBmedium, balanced quality - recommended
llava-v1.5-7b-Q4_K_S.ggufQ4KS43.86 GBsmall, greater quality loss
llava-v1.5-7b-Q5_0.ggufQ5_054.65 GBlegacy; medium, balanced quality - prefer using Q4KM
llava-v1.5-7b-Q5_K_M.ggufQ5KM54.78 GBlarge, very low quality loss - recommended
llava-v1.5-7b-Q5_K_S.ggufQ5KS54.65 GBlarge, low quality loss - recommended
llava-v1.5-7b-Q6_K.ggufQ6_K65.53 GBvery large, extremely low quality loss
llava-v1.5-7b-Q8_0.ggufQ8_087.16 GBvery large, extremely low quality loss - not recommended
llava-v1.5-7b-mmproj-model-f16.gguff168624 MB

Quantized with llama.cpp b2230