tao-finetune-cosmos-embed

To fine-tune a video-text embedding model for text-to-video retrieval, run the `cosmos-embed1 train` command with your dataset configuration. This adapts the Cosmos-Embed1 model to improve retrieval accuracy for your specific application.

Video-text embedding maps joint video and text features into a shared space for semantic deduplication. It identifies duplicate video content by comparing embeddings, supporting text-to-video retrieval and video-to-video search applications.

Yes, you need Docker and nvidia-container-toolkit to run Cosmos-Embed1 fine-tuning. You also need a Huggingface token for model access. These dependencies provide the containerized GPU environment required for training and evaluation.

Yes, you can evaluate and export a fine-tuned Cosmos-Embed1 model in the same workflow. The Skill supports the complete training, evaluation, and export pipeline for customized video-text embedding models.

You should use a fine-tuned video-text embedder instead of the base model when your application involves domain-specific text-to-video retrieval or semantic deduplication. Fine-tuning customizes the Cosmos-Embed1 model to recognize patterns unique to your dataset.