You can use transaction ZLLM_CLIENT_CONFIG or SM30 table ZLLM_CLNT_CONFIG.
The table has the following fields which can be maintained as customzing:
- Model: This is the model key you use in coding
- Section LLM Client Configuration:
- Provider Name: Select the configured Provider
- Model: Provider-internal model name; in case of Azure OpenAI serivces this is the name of the deployment
- Struct. Output?: Currently information only - if the model supports structured output, not validated at runtime
- Tools Supported?: Currently information only - if the model supports tool output, not validated at runtime
- Default Options: Needs testing - Text field that allows you to set options that will be passed as-is to the provider except they are overridden by code. Entries can be separated by ';', key/value separated by ':'. Values need to be enclosed in '"' if they are string values. Better maintenance options and examples will be provided in future.
Your feedback and experience is highly appreciated to extend this list:
- Setup default models for specific feature categories that can be exchanged without implementation changes, e.g.
- Text output
- Tool Calls
- Structured Output
- Avoid giving more details than necessary, e.g. in case of ollama you might use the model as llama3.2 which then is mapped to llama3.2:3b-instruct-q8_0
Via transaction ZLLM_SYSTEM_CONF or SM30 table ZLLM_SYSTEM statistics and tracing (saving http calls) can be enabled/disabled. Disabled by default.
Currently this is a very rudimentary implementation which will be extended in future based on demand.
- Table zllm_call_log contains the full http client request and response without headers and cookies - main use is for debugging
- Table zllm_statistics contains basic call statistics with token counts