Skip to content

Connecting to IBM-hosted Models§

This service is for registered IBM partners only.
Before you continue, make sure you know who your group admin is.

Note

For more information about becoming an IBM partner, get in touch.

Step 1: Register at the OpenAD Portal§

  1. Log in to the OpenAD Portal
    Go to open.accelerate.science and log in with your IBMid, create one if necessary.
    You'll see: "Your account is pending to be added to a group"

    OpenAD Portal

  2. Request Group Assignment
    Log out and contact your group administrator.
    They will add you to the appropriate group.

  3. Verify Account Setup
    Log in again after receiving confirmation.
    Your group and role should be displayed.

Step 2: Generate Access Token§

  1. Navigate to the Access Token tab
  2. Click Generate Token

  3. Click anywhere on the token to copy it, you'll need it later

    Warning

    Treat your token like a password. It grants full access to the system under your credentials.

Step 3: Connect to a Model§

If you haven't already, install OpenAD.

3.1. Choose a model§

To see what model subscriptions you have access to, check the Your Subscriptions tab in the OpenAD Portal.

OpenAD Portal

3.2. Configure Authentication§

Replace YOUR_ACCESS_TOKEN with your actual access token that you copied earlier.

Bash
model auth add group default with 'YOUR_ACCESS_TOKEN'

3.3. Connect to a Model§

In this example we'll connect to the molformer model.

Bash
catalog model service from remote 'https://open.accelerate.science/proxy' as molformer using (auth_group=default inference-service=molformer)

Understanding the Connection Command§

Command Component Description
catalog model service from remote Connects to a model via URL
'https://open.accelerate.science/proxy' The endpoint for model inference
'molformer' Your chosen name for this service
USING (auth_group=default ...) References your authentication group
USING (... inference-service=molformer) The model subscription name

Step 4: Verify the Connection§

Bash
model service status

Expected output:

Text Only
Service    Status     Endpoint                               Host    Token Expires
---------  ---------  -------------------------------------  ------  -----------------
molformer  Connected  https://open.accelerate.science/proxy  remote  Wed Sep 11, 2030

Step 5: Explore Available Model Functions§

Bash
molformer ?

You'll see available commands for the model:

Text Only
Commands starting with "molformer"
- molformer get molecule property molformer_classification for [<list of SMILES>] | <SMILES> USING (<parameter>=<value> <parameter>=<value>) (save_as '<filename.csv>')
- molformer get molecule property molformer_multitask_classification for [<list of SMILES>] | <SMILES> USING (<parameter>=<value> <parameter>=<value>) (save_as '<filename.csv>')
- molformer get molecule property molformer_regression for [<list of SMILES>] | <SMILES> USING (<parameter>=<value> <parameter>=<value>) (save_as '<filename.csv>')

Step 6: Run Model Inference§

Bash
molformer get molecule property molformer_classification for 'OC12COC3=NCC1C23'

Expected output:

Text Only
subject           property                  result
----------------  ------------------------  --------
OC12COC3=NCC1C23  molformer_classification  [1]

Command Documentation§

See all command related to the model service:

Bash
model ?

To see detailed documentation for any individual command, append a ? to the unique beginning of the command:

Bash
model auth add group ?

Best Practices§

  • Store important results using the save_as parameter.
    Alternatively, you use a follow-up command: result save
  • Use descriptive service names when connecting to models.
  • Organize multiple models and authentication groups logically.
  • Back up your access token securely.

Troubleshooting§

  • Connection Issues: Verify your token has not expired and the inference URL is correct.
  • Authentication Errors: Regenerate your token if necessary.
  • Model Not Available: Check your dashboard for available model subscriptions.
  • Model misspelled: Make sure there is no typo in the inference-service parameter.
  • Command Syntax Errors: Use the ? command to verify proper syntax.