Nimbus-LLM Weather Summary API

A simple FastAPI-based Python API for generating fun weather summaries using local LLMs (e.g., Qwen2.5 3b) on a Raspberry Pi.

Features

• /api/v1/generate/weather-summary endpoint accepts weather data and returns a job ID.
• Jobs are processed one at a time; up to 10 jobs can be queued.
• Poll /api/v1/generate/weather-summary/{job_id} for results.
• Model path is configurable via the MODEL_PATH environment variable.
• Old jobs are automatically removed when more than 100 are stored.

Request Example

POST /api/v1/generate/weather-summary

{
	"tempCelcius": 22.5,
	"humidity": 60,
	"rainMmPerHour": 0.0,
	"windSpeedMph": 5.2,
	"clouds": "partly"
}

Response Example

{
	"job_id": "<uuid>"
}

Polling for Results

GET /api/v1/generate/weather-summary/{job_id}

• status: done and result when complete
• status: pending if still processing
• status: error if failed

Model Integration

• The API uses a stub for model inference. Replace run_model() in main.py with your LLM integration (e.g., Qwen2.5 3b).
• Set the model path with the MODEL_PATH environment variable.

Running on Raspberry Pi

1. Install dependencies: bash pip install fastapi uvicorn pydantic
2. Start the API: bash uvicorn main:app --host 0.0.0.0 --port 8000
3. (Optional) Set the model path: bash export MODEL_PATH=/path/to/your/model

Notes

• Only one job is processed at a time due to hardware constraints.
• The job queue holds up to 10 jobs; excess requests get a 429 error.
• Old jobs are removed after 100 jobs to save memory.

License

MIT