More endpoints for Watsonx ai integration: deployed prompt, AI service

integrations/extensions/starter-kits/language-model-watsonx/advanced/README.md

@@ -0,0 +1,130 @@
+# Watsonx-ai-openapi
+
+This is a openapi spec including all available llm endpoints. Including
+
+| Endpoint                                         | Description                                                                |
+| ------------------------------------------------ | -------------------------------------------------------------------------- |
+| Generation                                       | Direct generate from wx.ai                                                 |
+| Generation (stream)                              | Direct generation with streaming                                           |
+| Generation from deployed model/template          | From deployed model/template without explicitly prompt                     |
+| Generation from deployed model/template (stream) | From deployed model/template without explicitly prompt with streaming      |
+| Generation from AI service                       | For services like RAG, agent lab, chat with document, etc.                 |
+| Generation from AI service                       | For services like RAG, agent lab, chat with document, etc., with streaming |
+
+## Before you start
+
+You need to have a deployed model/prompt template, or a deployed AI service on Watsonx.ai
+Read more:
+
+- [Deploying a prompt template](https://eu-de.dataplatform.cloud.ibm.com/docs/content/wsj/analyze-data/prompt-template-deploy.html?context=wx&locale=en&audience=wdp)
+- [Deploying a tuned foundation model](https://eu-de.dataplatform.cloud.ibm.com/docs/content/wsj/analyze-data/fm-tuning-deploy.html?context=wx&locale=en&audience=wdp)
+- [Deploying AI services](https://eu-de.dataplatform.cloud.ibm.com/docs/content/wsj/analyze-data/ai-services-overview.html?context=wx&locale=en)
+
+---
+
+## Generation Streaming
+
+Use streaming
+
+1. Choose operation with _"Generation stream"_.
+2. Set up streaming reponse `text` to `results[0].generated_text`.
+
+Read more on [streaming from an extension](https://cloud.ibm.com/docs/watson-assistant?topic=watson-assistant-stream-from-extension) to config streaming in WxA.
+
+---
+
+## Deployed prompt template
+
+1. Choose operation _"Generation from a deployed prompt"_
+
+2. Set up parameters:
+
+- `deployment-id`
+- `version`
+- `parameters.prompt_variables`
+
+Input variable `parameters.prompt_variables` is object type. You should give variable values as expression depending on your deployment. For example:
+
+```json
+{ "question": "$question" }
+```
+
+<img src="./assets/deployed-prompt-input.png" width="500">
+
+Output message is in `body.results[0].generated_text`
+
+<img src="./assets/deployed-prompt-output.png" height="250">
+
+## Deployed prompt template (stream)
+
+1. Choose operation _"Generation from a deployed prompt (stream)"_
+
+2. Set up parameters:
+
+- `deployment-id`
+- `version`
+- `parameters.prompt_variables`
+
+Input variable `parameters.prompt_variables` is object type. You should give variable values as expression depending on your deployment. For example:
+
+```json
+{ "question": "$question" }
+```
+
+Set up streaming reponse `text` to `results[0].generated_text`.
+
+<img src="./assets/deployed-prompt-stream.png" width="500">
+
+---
+
+## Deployed AI service
+
+1. Choose operation _"Generation from a deployed AI service"_
+
+2. Set up parameters:
+
+- `deployment-id`
+- `version`
+- `message`
+
+Input variable `message` should be given as an expression in form of:
+
+```json
+[
+  {
+    "role": "user",
+    "content": " $user_input "
+  }
+]
+```
+
+<img src="./assets/ai-service-input.png" width="500">
+
+Output message should be `body.choices[0].message.content`
+
+<img src="./assets/ai-service-output.png" width="250">
+
+## Deployed AI service (stream)
+
+1. Choose operation _"Generation from a deployed AI service (stream)"_
+
+2. Set up parameters:
+
+- `deployment-id`
+- `version`
+- `message`
+
+Input variable `message` should be given as an expression in form of:
+
+```json
+[
+  {
+    "role": "user",
+    "content": " $user_input "
+  }
+]
+```
+
+Output message should be `body.choices[0].message.delta.content`
+
+<img src="./assets/ai-service-stream.png" width="500">