| --- |
| license: cc-by-4.0 |
| task_categories: |
| - text-generation |
| language: |
| - en |
| tags: |
| - function-calling |
| - indic |
| - agentic |
| pretty_name: IndicFCB Agriculture (English dialogues) |
| size_categories: |
| - n<1K |
| --- |
| |
| # IndicFCB — Agriculture (English dialogues) |
|
|
| 303 English user-side dialogues for Indian government agricultural services, |
| classified along the IndicFCB benchmark axes. One row per (task x persona) rendering. |
|
|
| ## Columns |
| - `task_id` — stable identifier |
| - `turns_axis` — single_turn | multi_turn |
| - `steps_axis` — single_step | multi_step |
| - `type_axis` — simple | parallel_multiple | irrelevance |
| - `usecase` — service+goal tag (e.g. pmkisan_income, kcc_credit) |
| - `intent` — one-line summary of the citizen's goal |
| - `persona` — direct | assisted | vernacular (user-communication style) |
| - `dialogue` — the ordered user messages |
| - `ground_truth` — the required gold tool-call trace, JSON list of {turn, name, arguments} |
| (empty list for `irrelevance` tasks, where the correct behaviour is to abstain) |
| - `tools_available` — the FULL domain tool inventory shown to the agent (same for every |
| row): the benchmark tests whether the agent can pick the right tools for the usecase |
| out of all domain tools. |
|
|
| Separate from `saiteja2311/ifcb_ai4b` (left unchanged). |
|
|
