This is a VLM-TSI fine-tuned on the streaming free-form Ego4D GoalStep conversation data from chenjoya/videollm-online-chat-ego4d-134k. It's able to assist a user through a task in real-time. It was initialized from chenjoya/videollm-online-8b-v1plus, but fine-tuned with the novel time-synchronized interleaving strategy.

Please refer to the documentation on https://github.com/yukw777/tglg.