Deploy a Time-Series Foundation Model

A number of research institutions and enterprises have released open-source time-series foundation models (TSFMs), greatly simplifying time-series data analysis. Beyond traditional data analysis algorithms, machine learning, and deep learning models, TSFMs offer a new and powerful option for advanced time-series analytics. This chapter introduces how to deploy and use open-source TSFMs.

TDgpt includes two TSFMs, TDtsfm and Time-MoE, but you can add more open-source or proprietary TSFMs to TDgpt as needed. This document describes how to integrate an independent TSFM service into TDengine, using Time-MoE as an example, and how to use the model in SQL statements for time-series forecasting.

Prepare Your Environment

Before using TSFMs, prepare your environment as follows. Install a Python environment and use PiPy to install dependencies:

pip install torch==2.4.1+cpu -f https://download.pytorch.org/whl/torch_stable.html
pip install flask==3.0.3
pip install transformers==4.40.0
pip install accelerate

You can use the virtual Python environment installed by TDgpt or a separate environment.

Configure TSFM Path

The lib/taosanalytics/time-moe.py file in the TDgpt root directory deploys and serves the Time-MoE model. Modify this file to set an appropriate URL.

@app.route('/ds_predict', methods=['POST'])
def time_moe():
...

Change ds_predict to the URL that you want to use in your environment.

    app.run(
            host='0.0.0.0',
            port=5001,
            threaded=True,  
            debug=False     
        )

In this section, you can update the port if desired. After you have set your URL and port number, restart the service.

Run the Python Script

nohup python time-moe.py > service_output.out 2>&1 &

The script automatically downloads Time-MoE-200M from Hugging Face the first time it is run. You can modify time-moe.py to use TimeMoE-50M if you prefer a smaller version.

Check the service-output.out file to confirm that the model has been loaded:

Running on all addresses (0.0.0.0)
Running on http://127.0.0.1:5001

Verify the Service

Verify that the service is running normally:

curl 127.0.0.1:5001/ds_predict

The following indicates that Time-MoE has been deployed:

<!doctype html>
<html lang=en>
<title>405 Method Not Allowed</title>
<h1>Method Not Allowed</h1>
<p>The method is not allowed for the requested URL.</p>

Load the Model into TDgpt

You can modify the [timemoe.py] file and use it in TDgpt. In this example, Time-MoE is adapted to provide forecasting.

class _TimeMOEService(AbstractForecastService):
    # model name, user-defined, used as model key
    name = 'timemoe-fc'

    # description
    desc = ("Time-MoE: Billion-Scale Time Series Foundation Models with Mixture of Experts; "
            "Ref to https://github.com/Time-MoE/Time-MoE")

    def __init__(self):
        super().__init__()

        self.table_name = None

        # find service address in taosanode.ini or use default if not found
        service_host = conf.get_tsfm_service("timemoe-fc")
        if  service_host is not None:
            self.service_host = service_host
        else:
            self.service_host = 'http://127.0.0.1:5001/timemoe'

        self.headers = {'Content-Type': 'application/json'}


    def execute(self):
        """analytics methods"""
        if self.list is None or len(self.list) < self.period:
            raise ValueError("number of input data is less than the periods")

        if self.rows <= 0:
            raise ValueError("fc rows is not specified yet")

        # let's request the gpt service
        data = {"input": self.list, 'next_len': self.rows}
        try:
            # request tsfm service
            response = requests.post(self.service_host, data=json.dumps(data), headers=self.headers)
        except Exception as e:
            app_logger.log_inst.error(f"failed to connect the service: {self.service_host} ", str(e))
            raise e

         # check returned value
        if response.status_code == 404:
            app_logger.log_inst.error(f"failed to connect the service: {self.service_host} ")
            raise ValueError("invalid host url")
        elif response.status_code != 200:
            app_logger.log_inst.error(f"failed to request the service: {self.service_host}, reason: {response.text}")
            raise ValueError(f"failed to request the service, {response.text}")

        pred_y = response.json()['output']

        res =  {
            "res": [pred_y]
        }

        # package forecasting results according to specifications
        insert_ts_list(res["res"], self.start_ts, self.time_step, self.rows)
        return res

Add your code to /usr/local/taos/taosanode/lib/taosanalytics/algo/fc/ where the timemoe.py file is located.

TDgpt has built-in support for Time-MoE. You can run SHOW ANODES FULL and see that forecasting based on Time-MoE is listed as timemoe-fc.

Configure TSFM Service Path

Modify the [tsfm-service] section of /etc/taos/taosanode.ini:

[tsfm-service]
timemoe-fc = http://127.0.0.1:5001/ds_predict

Add the path for your deployment. The key is the name of the model defined in your Python code, and the value is the URL of Time-MoE on your local machine.

Then restart the taosanode service and run UPDATE ALL ANODES. You can now use Time-MoE forecasting in your SQL statements.

Use a TSFM in SQL

SELECT FORECAST(i32, 'algo=timemoe-fc') 
FROM foo;

Add Other TSFMs

You can add more open-source or proprietary TSFMs to TDgpt by following the process described in this document. Ensure that the class and service names have been configured appropriately and that the service URL is reachable.

References

Time-MoE: Billion-Scale Time Series Foundation Models with Mixture of Experts. [paper] [GitHub Repo]

Prepare Your Environment​

Configure TSFM Path​

Run the Python Script​

Verify the Service​

Load the Model into TDgpt​

Configure TSFM Service Path​

Use a TSFM in SQL​

Add Other TSFMs​

References​

Support and Feedback