v1.7.0

What's Changed

• [FEATURE] Arrbitrary tokenizer kwargs by @psinger in #718
• [FIX] fix unresponsive app by @pascal-pfeiffer in #720
• [FEATURE] Save each evaluation epoch by @pascal-pfeiffer in #721
• [IMPROVEMENT] Improve dataset import by @pascal-pfeiffer in #717
• [FEATURE] Allow to use DPO without LoRA by @psinger in #726
• [FEATURE] Option to freeze (non lora) and unfreeze (lora) layers by @psinger in #731
• [DOCS] Improve FAQ page by @sherenem in #739
• [FEATURE] Save LoRA adapter for download by @pascal-pfeiffer in #729
• [DOCS] Remove invalid reference to tooltip by @sherenem in #744
• [DOCS] minor fix by @pascal-pfeiffer in #745
• [CHORE] Fully remove RLHF in favor of DPO by @pascal-pfeiffer in #747
• [IMPROVEMENT] Max Length rework by @psinger in #741
• [FIX] Deepseek tokenizer by @psinger in #746
• [FIX] only show default entries once by @pascal-pfeiffer in #751

Full Changelog: v1.6.0...v1.7.0