ExperimentSpec
ExperimentSpec(
task,
model,
use_examples=DEFAULT_USE_EXAMPLES,
prompt_type=DEFAULT_PROMPT_TYPE,
temperature=DEFAULT_TEMPERATURE,
top_p=DEFAULT_TOP_P,
chat_mode=DEFAULT_CHAT_MODE,
reasoning=DEFAULT_REASONING,
process_textbox=False,
process_span=False,
country_iso_code=DEFAULT_COUNTRY_ISO_CODE,
)Declarative specification for one experiment run in a sweep.
Attributes
| Name | Type | Description |
|---|---|---|
| task | str | Task folder name under tasks/, for example "policy-sentiment". |
| model | str | Ollama model identifier, such as "gemma3:270m". |
| use_examples | bool | Whether to include worked examples from the codebook. |
| prompt_type | str | Registered prompt wrapper name, for example "standard". |
| temperature | float | None | Optional sampling temperature as None, string, or float. |
| top_p | float | None | Optional nucleus-sampling value as None, string, or float. |
| chat_mode | str | How annotation calls share chat history. Supported values are "per_text", "per_query", and "continuous". |
| reasoning | bool | str | None | Ollama reasoning mode passed to ChatOllama. None keeps the model default, True enables it, and False disables it. |
| process_textbox | bool | Whether textbox annotations should be generated and scored. |
| process_span | bool | Whether span annotations should be generated and scored. |
| country_iso_code | str | Three-letter ISO 3166-1 alpha-3 code for CodeCarbon. |