rafw007/ bielik-custom-ocr-vision:latest

13 3 days ago

A custom merged model: the vision layers were borrowed from Mistral-3 and joined with Bielik-7B (SpeakLeash). The result is a compact multimodal model that reads text from images — printed pages, scans.

vision
... /
params
c2453c2bb2e0 · 154B
{
"num_ctx": 8192,
"num_predict": 1024,
"repeat_penalty": 1.05,
"stop": [
"<|im_start|>",
"<|im_end|>"
],
"temperature": 0,
"top_k": 20,
"top_p": 0.9
}