Joschka Braun

- refactor data loading : 1. have data loading (into DocArray) 2. have mapping of attributes 3. data preprocessing

Can use `content` attribute of `Document`.

Yes, we need it e.g. to remove any loaded content attributes before putting them into the indexer. Except for we modify every indexer to not save these attributes.

It depends on the user if they prefer CLI or storefront. A user can use CLI to have multiple remote deployments.

We need to enable only vector output for that; then we basically have that. Only other endpoint to add is a `/rank` endpoint.

Any update?

This should become less often an issue due to: https://github.com/jina-ai/gptdeploy/pull/70 Will close the ticket for now.

Thank you for sharing the code! This looks like a context-length issue (output just stopped). We are currently working on minimising the system messages to avoid this issue.

This should become less often an issue due to: https://github.com/jina-ai/gptdeploy/pull/70 Will close the ticket for now.

> Should I assume the missing package is a consequence of using model 3.5? Yes that can be. GPT-4 is much more powerful. But we are improving our package currently...