roboflow / maestro

streamline the fine-tuning process for multimodal models: PaliGemma, Florence-2, and Qwen2-VL
https://maestro.roboflow.com
Apache License 2.0
1.38k stars 102 forks source link

Prediction mode #77

Open hardikdava opened 4 weeks ago

hardikdava commented 4 weeks ago

Search before asking

Description

Currently, it is only possible to finetune models. But it would be amazing to have predict mode where user can utilize base model functionality without much efforts. Something like florence-2 supports different basic tasks such as image caption, object detection, classification, etc.

Use case

Additional

No response

Are you willing to submit a PR?