felix-wang

> Can CLIP can help to generate embedding for products and club into one fixed length vector that represents the product embedding? the answer is yes. CLIP can produce text...

> how I can evaluate it on my already trained CLIP model? You can evaluate your model with the downstream task ., e.g., classification, retrieval, based on your dataset. And...

> Do I need to train my CLIP model again with FineTuner? That depends on your domain. The pre-trained CLIP model is trained with a large-scale corpus in the general...

we will close this issue for now. If you have some news findings to share, you are welcome to create a new ticket Thanks!

@jemmyshin Pleas rebase your PR to fix the conflict.

@jemmyshin please rebase your PR to fix conflicts

@jemmyshin This PR has been merged to https://github.com/jina-ai/clip-as-service/pull/816. I will close this PR first. let's work on the new PR.

@sjcotto maybe the metric `search_params = {"metric_type": "L2"}` chosen here does not fit with clip model. Usually, we use `cosine` as the distance metric, cause CLIP is trained with cosine...

we will close this issue for now. If you have some findings to share, you are welcome to share with community. Thanks!

@houshangwu Thanks for pointing it out. Some layers/operators result in inconsistent output on FP16. To fix this problem, the mix-precision approach could work. Please keep an eye on our progress,...