- Add print statement for HTTP response status in client.py websocket handler
- Comment out JSON parsing of stream chunks, send raw decoded bytes instead
- Add chunk validation and yield newline for empty chunks in server.py generator
- Add 'stop' key check alongside 'done' for stream termination
- Fix model listing response format to include 'object' and 'data' wrapper
- Update model metadata fields: replace 'owner' with 'owned_by', set non-zero creation timestamp
The models handler now returns a list of structured model objects with id, instances, owner, and created fields instead of a flat count dict. A new GET /v1/models route is registered alongside the existing /models endpoint to support OpenAI-compatible API paths.
Add a new shell script `ollama-colab-v2.sh` that automates the setup of Ollama on Google Colab, including installation of Ollama, pulling the qwen2.5-coder:14b model, starting the Ollama server in background, and running the uberlama client to expose the model through a public tunnel at ollama.molodetz.nl. The script also includes instructions for testing with curl and keeps the notebook session alive by tailing log files.
This new test file demonstrates how to use the OpenAI-compatible client to connect to an Ollama endpoint at https://ollama.molodetz.nl/v1, sending a multi-turn conversation about the 2020 World Series and printing the assistant's response.
