qwen_0.8b_vlm — GGUF bundle for LM Studio
Vision (OCR) model. LM Studio needs BOTH of these in the SAME folder (it pairs
them automatically by the mmproj filename prefix):
- Language model(s): qwen_0.8b_vlm-F16.gguf
- Vision projector : mmproj-qwen_0.8b_vlm-F16.gguf
Quantization is optional for a model this small — the F16 file works directly in
LM Studio (~1.6 GB) and needs no compiled llama.cpp. Any *-Q4_K_M/*-Q8_0
files present were quantized with a llama-quantize binary.
Load in LM Studio
- Copy this whole folder into your LM Studio models directory:
macOS/Linux : ~/.lmstudio/models/Melaraby/qwen_0.8b_vlm/
Windows : %USERPROFILE%.lmstudio\models\Melaraby\qwen_0.8b_vlm\
- Rescan models. A vision ("eye") icon = the mmproj was detected.
- Attach an image and use the trained OCR prompt:
System: You are an OCR engine. Detect every text line and output
<|box_start|>(ymin,xmin,ymax,xmax)<|box_end|>transcribed text
User : (image) + "Transcribe all text lines."
Sanity check with llama.cpp directly
llama-mtmd-cli -m qwen_0.8b_vlm-F16.gguf --mmproj mmproj-qwen_0.8b_vlm-F16.gguf --image page.jpg -p "Transcribe all text lines."
