Ternary-quantized version of HuggingFaceTB/SmolVLM2-256M-Video-Instruct — the smallest video-understanding VLM, now even smaller. Ideal for edge deployment.

Specifications

Property	Value
Base Model	HuggingFaceTB/SmolVLM2-256M-Video-Instruct
Parameters	256M
Quantization	tritplane3 (211 linear layers)
Full-model effective bits	12.91
Compression ratio	1.24×
Avg reconstruction error	0.1488

Collection