sugiv's picture
Add comprehensive README
a20102b verified
metadata
language:
  - en
license: apache-2.0
base_model: sugiv/cardvaultplus-500m
tags:
  - vision-language
  - credit-card
  - gguf
  - quantized
  - llama.cpp
pipeline_tag: image-to-text

CardVault+ SmolVLM-500M GGUF Models

Available Models

Model File Size Quantization Use Case
cardvault-500m-f16.gguf 783MB F16 (Base) Maximum quality
cardvault-500m-mmproj-f16.gguf 191MB F16 (Vision) REQUIRED
cardvault-500m-q8_0.gguf 417MB Q8_0 Near-perfect quality
cardvault-500m-q6_k.gguf 399MB Q6_K Balanced
cardvault-500m-q5_k_m.gguf 311MB Q5_K_M Recommended
cardvault-500m-q4_k_m.gguf 290MB Q4_K_M Maximum compression

Usage

# Download llama.cpp
git clone https://github.com/ggerganov/llama.cpp
cd llama.cpp && make

# Run inference (Q5_K_M recommended)
./main \
  --model cardvault-500m-q5_k_m.gguf \
  --mmproj cardvault-500m-mmproj-f16.gguf \
  --image credit_card.jpg \
  --prompt "Extract card information in JSON format"

⚠️ Two-Component Architecture: Both text model + mmproj required!