| language: | |
| - tr | |
| tags: | |
| - conversational | |
| - chatml | |
| - alignment | |
| - response-format | |
| - reasoning-control | |
| task_categories: | |
| - text-generation | |
| pretty_name: DEHA Correction Data Base | |
| size_categories: | |
| - 1K<n<10K | |
| # DEHA Correction Data Base | |
| Bu veri seti, Türkçe konuşan bir modelde görülen görünür reasoning ve format bozulmalarını düzeltmek için üretildi. | |
| İçerik: | |
| - clean no-think örnekleri | |
| - kısa `<think>` bloklu hidden-thinking örnekleri | |
| - strict output format örnekleri | |
| Ana amaç: | |
| - düşünceyi gizli tutmayı öğretmek | |
| - final cevabı temiz vermeyi öğretmek | |
| - `Thinking Process:` gibi sızıntıları azaltmak | |