Heron-NVILA-Lite-2B GGUF (Q2_K)

This is the Q2_K quantized GGUF version of turing-motors/Heron-NVILA-Lite-2B-hf, a Vision-Language Model optimized for Japanese.

Model Details

Original Model: turing-motors/Heron-NVILA-Lite-2B-hf
Quantization: Q2_K
Format: GGUF (for llama.cpp)
File Size: 0.7 GB
Languages: Japanese, English

Model Architecture

Component	Details
Vision Encoder	SigLIP (paligemma-siglip-so400m-patch14-448)
Projector	mlp_downsample_2x2_fix
LLM	Qwen2.5-1.5B
Total Parameters	~2B

Requirements

This model requires the mmproj (multimodal projector) file to process images:

mmproj: nawta/Heron-NVILA-Lite-2B-mmproj-GGUF

Available Quantizations

Quantization	File Size	Use Case
F16	3.3 GB	Maximum quality
Q8_0	1.8 GB	High quality
Q6_K	1.4 GB	Balanced
Q5_K_M	1.2 GB	Balanced
Q4_K_M	1.1 GB	Efficient
Q3_K_M	0.9 GB	Compact
Q2_K	0.7 GB	Minimum size

Usage with llama.cpp

# Download the model and mmproj
wget https://huggingface.co/nawta/Heron-NVILA-Lite-2B-Q2_K-GGUF/resolve/main/Heron-NVILA-Lite-2B-Q2_K.gguf
wget https://huggingface.co/nawta/Heron-NVILA-Lite-2B-mmproj-GGUF/resolve/main/mmproj-heron-nvila-lite-2b-f16.gguf

# Run inference
./llama-mtmd-cli \
  -m Heron-NVILA-Lite-2B-Q2_K.gguf \
  --mmproj mmproj-heron-nvila-lite-2b-f16.gguf \
  --image your_image.jpg \
  -p "この画像について説明してください。"

Memory Requirements

Quantization	LLM	mmproj	Total
F16	3.3 GB	807 MB	~4.1 GB
Q8_0	1.8 GB	807 MB	~2.6 GB
Q6_K	1.4 GB	807 MB	~2.2 GB
Q5_K_M	1.2 GB	807 MB	~2.0 GB
Q4_K_M	1.1 GB	807 MB	~1.9 GB
Q3_K_M	0.9 GB	807 MB	~1.7 GB
Q2_K	0.7 GB	807 MB	~1.5 GB

License

This model inherits the license from the original Heron-NVILA-Lite-2B model. Please refer to turing-motors/Heron-NVILA-Lite-2B-hf for license details.