GGUF + vision-projector (mmproj) pair used by the Repasar app for on-device Photo AI (listing detail extraction). Built from the official Qwen3.5-2B release, quantized to Q4_K_M.

Files

File	Size	SHA-256
`Qwen3.5-2B-Q4_K_M.gguf`	1.19 GB	`aaf42c8b7c3cab2bf3d69c355048d4a0ee9973d48f16c731c0520ee914699223`
`mmproj-F16.gguf`	637 MB	`7035e9cb8d7c6a9681d07eef9a364783e86ea4cd73faab2eabb4f43a101830c7`

Both files must be downloaded as a pair (model + vision projector).

Usage

Direct download URLs (serve via resolve/main):

https://huggingface.co/Repasar-Labs/qwen3.5-2b-gguf/resolve/main/Qwen3.5-2B-Q4_K_M.gguf
https://huggingface.co/Repasar-Labs/qwen3.5-2b-gguf/resolve/main/mmproj-F16.gguf

The Hub CDN supports HTTP Range requests, so resumable downloads work against these URLs.