Ollama Profile

Use the Ollama Profile to configure the connection between HCL UnO Agentic AI Builder and your self-hosted Ollama service. Ollama is an open-source framework that allows users to run various Large Language Models (LLMs) locally or on a private network, enabling agents to leverage specific, highly customizable, and often smaller models.

Before you begin

  • You must have a running Ollama server instance accessible from the Agentic AI Builder network.

  • The required models (for example, llama2, mixtral) must be pulled and available on your Ollama server.

  • You must have a valid Ollama Credential (likely containing the server endpoint or API key) configured in the Credential Library to authenticate this connection.

  • Ensure that all mandatory fields (marked with *) are completed accurately.

Table 1. Mandatory fields
Option Description
Name

A unique identifier for this configuration instance.
LLM/SLM Name

The specific authentication credential (previously created in the Credential Library) used to authorize the connection to your Ollama service.
Table 2. Optional fields
Options Description
Model Name The name of the Ollama model to use (for example, llama2, mixtral:8x7b-instruct-v0.1).
Note:

The Model Name field is where you specify the exact technical identifier for the Large Language Model (LLM) or Small Language Model (SLM) to be used.

If Discover Models is checked in the corresponding Credentials account, available LLM models will populate a dropdown menu for selection in the LLM Name field.
Num Ctx

Sets the size of the context window (in tokens). This determines the maximum history the model can reference during a conversation.
Temperature Controls randomness. Higher values (for example, 0.8) mean more creative or varied output.
Num Predict The maximum number of tokens to generate in the output (equivalent to Max Tokens).
Top K Reduces the probability of selecting nonsense tokens by restricting the model to sample from the top K most probable tokens.
Top P Works with Top K to control nucleus sampling. It limits the token choices to the most probable set based on a cumulative probability threshold.
Stop Stop Sequence Control. This overall feature is used to prevent the model from generating further text once a specific string is encountered.
Add Stop + Allows the user to add additional stop sequences. Clicking the Plus icon creates an additional input field (for example, Stop-2, Stop-3) allowing you to specify another unique token sequence that will trigger the model to halt generation. The Minus icon is used to remove the stop sequence field it is adjacent to. This allows you to delete unnecessary criteria and simplify the model's stopping rules.
Seed Set the random seed for generation to get reproducible results (for example, 42). Using the same seed ensures deterministic output.
Min P Sets a minimum probability threshold for token sampling. Tokens with a probability lower than this value will be excluded from selection.
Repeat Last N The number of last tokens to consider when applying the repetition penalty.
Repeat Penalty A multiplier applied as a penalty for repeating tokens within the generation. A higher value reduces repetition.
Presence Penalty A penalty applied to new tokens based on their existence in the text so far. Encourages the model to introduce new concepts.
Frequency Penalty A penalty applied to new tokens based on their frequency in the text so far. Reduces the likelihood of the model repeating common phrases or topics.