DinoDS Lane 05: Conversation Mode.

About

• maintain context across turns
• keep tone consistent
• repair tone when needed
• ask for clarification
• move the interaction toward a concrete next step

Data Product Features

• sample_id: Unique identifier for each row
• split: Dataset split (train, validation, test)
• language: Language of the example
• mode: Response mode
• tone: Desired response tone
• intent_family: Broad task category
• intent_subtype: More specific task subtype
• representation_choice: Preferred output style
• continuity_choice: Continuity handling label
• flow_state: Interaction flow label
• history_scope: Scope of contextual history
• safety_tag: Safety classification
• needs_search: Whether external search is required
• needs_history_search: Whether history retrieval is required
• connector_needed: Whether workflow/connector behavior is implied
• has_tool_call: Whether a tool/action structure is present
• tool_name: Tool name when applicable
• message_count: Number of messages in the conversation sequence
• is_multi_turn: Whether the example is multi-turn
• user_message: User-side input
• assistant_response: Target assistant response

Distribution

• Format: CSV
• Language: English
• Lane source: lane_05_conversation_mode
• Total rows: 100
• Number of columns: 21

• train.csv: 90 rows
• validation.csv: 5 rows
• test.csv: 5 rows
• all_rows.csv: 100 combined rows

Usage

• Conversational fine-tuning: Train models to handle ongoing threads instead of isolated prompts
• Tone and continuity control: Improve how assistants maintain tone while staying actionable
• Reply drafting systems: Support better follow-up and check-in message generation
• Support and service workflows: Improve multi-turn communication where clarification and next-step guidance matter
• Evaluation and filtering: Use metadata to select subsets by tone, intent, and conversation structure

Business Case and Value