Tevatron

/

dse-phi3-docmatix-v1

like 9

Naming it results (or with a prefix that corresponds to the model path should help the hyperlink in vidore redirect correctly !)
looks like an awesome model !

@manu Hello! Do you have plans to try modified docmatix (from this work) to tune ColPali? The first thing I did after seeing the Docmatix mention on LinkedIn was go to the Colpali repos :) And also found this model too!

Yeah definitely, trying to do it next week !
Very cool we both have the same vision for "retrieval in vision space". Whether through bi-encoders or Colbert style, the underlying paradigm of using doc screenshots I really believe in !