calling api with schema isn't working as intended

[Ala2hhh]

using the Extract by JSON Schema example provided in the paper, and also can be found in colab

is not giving a response with the provided schema and the data of the output is in MD rather than json

here is the curl command:

curl -v -X POST https://r.jina.ai/ \
  -H "Accept: application/json" \
  -H "Content-Type: application/json" \
  -H "X-No-Cache: true" \
  -d @- <<EOFEOF
  {
    "url": "https://news.ycombinator.com/",
    "jsonSchema": {
      "type": "object",
      "properties": {
        "title": {"type": "string", "description": "News thread title"},
        "url": {"type": "string", "description": "Thread URL"},
        "summary": {"type": "string", "description": "Article summary"},
        "keywords": {"type": "list", "description": "Descriptive keywords"},
        "author": {"type": "string", "description": "Thread author"},
        "comments": {"type": "integer", "description": "Comment count"}
      },
    "required": ["title", "url", "date", "points", "author", "comments"]
    }
  }
EOFEOF

Am I calling the API incorrectly?

[nomagick]

Hi @Ala2hhh .
You are missing the "respondWith" or "returnFormat" or "engine" parameter.
Specify any of them to "readerlm-v2".
Also, the instruction is very important for the model to behave.

Try this:

curl -v -X POST https://r.jina.ai/ \
  -H "Accept: text/event-stream" \
  -H "Content-Type: application/json" \
  -H "X-No-Cache: true" \
  -d @- <<EOFEOF
  {
    "url": "https://news.ycombinator.com/",
    "instruction": "Extract the specified information from a list of news threads and present it in a structured JSON format.",
    "jsonSchema": {
      "type": "object",
      "properties": {
        "title": {"type": "string", "description": "News thread title"},
        "url": {"type": "string", "description": "Thread URL"},
        "summary": {"type": "string", "description": "Article summary"},
        "keywords": {"type": "list", "description": "Descriptive keywords"},
        "author": {"type": "string", "description": "Thread author"},
        "comments": {"type": "integer", "description": "Comment count"}
      },
    "required": ["title", "url", "date", "points", "author", "comments"]
    },
    "respondWith": "readerlm-v2"
  }
EOFEOF

Speaking of this feature, sometimes I find the model crashes by repeating nonsense words or does not yield enough content, in this case, the parameters of the generation, namely repetition_penalty, and potentially others, need to be further tweaked. However, we currently don't expose these parameters to the API.
So for JSON generation, it's better to run the model locally so you can tweak the parameters.

[Ala2hhh]

Thank you that was helpful