View source on Github
Arize class to map up to 2 columns: model_column_name and params_column_name
class LLMConfigColumnNames:
model_column_name: Optional[str] = None
params_column_name: Optional[str] = None
| Parameters | Data Type | Expected Type in Column | Description |
|---|---|---|---|
model_column_name | str | The contents of this column must be strings | Column name for the name of the LLM model used. |
params_column_name | str | The contents of this column must be be well formatted JSON strings | Column name for the invocation parameters used at inference |
Code Example
| Index | llm_model_name | llm_params |
|---|---|---|
0 | gpt-4 | {``"temperature": 0.9,``"stop":[".","?"],``"frequency_penalty":"0.0"``} |
1 | gpt3.5-turbo | {``"temperature": 0.7,``"stop":[".","?"],``"frequency_penalty":"0.2"``} |
from arize.utils.types import LLMConfigColumnNames
# Declare LLM config columns
llm_config = LLMConfigColumnNames(
model_column_name = "llm_model_name", # column containing the LLM model name
params_version_column_name = "llm_params", # column containing the LLM invocation params
)