Fovea-4B
A 4B-parameter open-source vision-language model focused on image decision-making tasks (Jev-style decision-making). It is released under the Apache-2.0 license and can be deployed on lightweight hardware.
Timeline (last 90 days)
Fovea-4B is an open image decision model with 4B parameters.
Fovea-4B achieves leading performance on the ImageJev task.
Fovea-4B is released under the Apache-2.0 license.
"4B leading on a specific task" should be understood as an efficiency advantage on a limited task and evaluation set, rather than general capability surpassing large models.
Fovea-4B's optimization objective is concentrated on decision-type tasks such as ImageJev, unlike general-purpose vision-language models.
All Facts (5)
Fovea-4B is an open image decision model with 4B parameters.
50%UnverifiedFovea-4B achieves leading performance on the ImageJev task.
50%UnverifiedFovea-4B is released under the Apache-2.0 license.
50%Unverified"4B leading on a specific task" should be understood as an efficiency advantage on a limited task and evaluation set, rather than general capability surpassing large models.
50%UnverifiedFovea-4B's optimization objective is concentrated on decision-type tasks such as ImageJev, unlike general-purpose vision-language models.
50%