felix-wang

**Describe the feature** When the flow is using an external executor, and its access is protected with an authentication credential (i.g,. access token). For example, the hosted free clip-as-service `api.clip.jina.ai`....

This PR adds Jina-AI powered `embedding` class. This is useful to help user to use the model serving at [Jina AI Cloud](https://cloud.jina.ai/user/inference). At present, the user can access various CLIP...

how to work with the failed instance in the input batch

2

Current implementation: ... Proposed implementation with `"errors"` tag: ...

Some of the updates are included: - use `--install-requirements` to install dependencies - update Dockerfile to use layer cache - use specific version base image

session id: `2454cbfc-d8e7-11ed-8983-b6ab90c9f2f2` POST https://api.hubble.jina.ai/v2/rpc/user.identity.whoami [401] ``` { reason: [Filtered] } ```

upgrade the open-flagminto-v2

3

The `open-flagminto-v2` has just been released: https://laion.ai/blog/open-flamingo-v2/ We need to adopt it in opengpt for better compatible with huggingface.

support instructBLIP model

1

As title, lavis just released a new vision-language instruction-tuning framework using BLIP-2 models, achieving state-of-the-art zero-shot generalization performance on a wide range of vision-language tasks. https://github.com/salesforce/LAVIS/tree/main/projects/instructblip

As title, add mini-gpt4 model [support](https://github.com/Vision-CAIR/MiniGPT-4) - [x] vicuna model support