In this post, we explore an idea for generating thinking tokens for datasets that lack reasoning traces in SFT customization. We first examine the reasoning suppression problem, then introduce Self-Distilled Reasoning (SDR), validate it across three benchmarks, and provide practical recommendations.
Discovery
Learnings
- auto_storiesStories
- categoryAreas
- helpFAQ
- forumConversations
- descriptionPapers
- mailNewsletters
- video_libraryVideos
- timelineTimeline
Guidance
- codeTechnology of APIs
- business_centerBusiness of APIs
- campaignEvangelism of APIs
- gavelGovernance of APIs
- historyHistory of APIs
- account_balancePolitics of APIs
Toolbox
- domainProviders
- apiAPIs
- cloudSolutions
- buildTools
- verifiedStandards
- gavelRegulations
- horizontal_ruleHeaders
- layersOverlays
- extensionExtensions
- schemaSchema
- policyPolicies
- ruleRules