The json schema example doesn't work correctly

[lhgravendeel]

Describe the issue as clearly as possible:

Using the dockerized example from https://dottxt-ai.github.io/outlines/latest/reference/serve/vllm/e:

Started container with:

docker run --gpus all -p 8000:8000 outlinesdev/outlines --model="microsoft/Phi-3-mini-4k-instruct"

Running the example prompt of:

curl http://127.0.0.1:8000/generate \
    -d '{
        "prompt": "What is the capital of France?",
        "schema": {"type": "string", "maxLength": 5}
        }'

The response that comes back is:

{"text":["What is the capital of France?\", \""]}

I also tried updating the schema to be a full object:

curl http://127.0.0.1:8000/generate -d '{"prompt": "What is the capital of France?","schema": {"type": "object", "properties":{"capital": {"type":"string", "description": "The capital that was requested"}}}}'

This returns the expected answer, in an expected JSON structure - except the JSON is stringified inside an object with an array of text elements. This is the output data:

{"text":["What is the capital of France?{ \"capital\": \"Paris\" }"]}

Steps/code to reproduce the bug:

Started container with:

docker run --gpus all -p 8000:8000 outlinesdev/outlines --model="microsoft/Phi-3-mini-4k-instruct"

Running the example prompt of:

curl http://127.0.0.1:8000/generate \
    -d '{
        "prompt": "What is the capital of France?",
        "schema": {"type": "string", "maxLength": 5}
        }'

Result:

{"text":["What is the capital of France?\", \""]}

Alternatively, reproduce it with a JSON object:

curl http://127.0.0.1:8000/generate -d '{"prompt": "What is the capital of France?","schema": {"type": "object", "properties":{"capital": {"type":"string", "description": "The capital that was requested"}}}}'

This returns:

{"text":["What is the capital of France?{ \"capital\": \"Paris\" }"]}

Expected result:

Expected to see `Paris` or in the second example:

{"capital": "Paris"}

Error message:

No response

Outlines/Python version information:

Version information

Docker image outlinesdev/outlines:latest sha256:523381fa1d10b1a9b5124b241952140712af2105fcf841473ff7e1a3a125bad7
Maps to version 0.1.3

ENV PYTHON_VERSION=3.10.15

Startup log: Initializing an LLM engine (v0.5.1)

Tried this both on Windows 11 with WSL 2 and Docker Desktop, and a clean Debian 12 VM

Context for the issue:

No response