DILAB-HYU/DialRet

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Feb 27, 2025License:mitArchitecture:Transformer0.0K Open Weights Cold

DialRet is a 7 billion parameter dialogue-specific language model developed by DILAB-HYU, built upon the Llama-1 base architecture. It is designed to enhance dialogue retention and understanding across multi-session conversations by leveraging long-context LMs and instruction-tuning across eight diverse dialogue tasks. This model excels in maintaining context and improving the quality of multi-session dialogues without relying on explicit memory modules.

Loading preview...

DialRet: Enhancing Multi-Session Dialogue Retention

DialRet, developed by DILAB-HYU, is a 7 billion parameter language model specifically engineered for multi-session conversations. Unlike traditional approaches that use memory modules, DialRet leverages long-context language models and instruction-tuning across eight distinct dialogue tasks, including dialogue generation, summarization, and speaker relation extraction. This methodology allows it to significantly improve the understanding and retention of past dialogues.

Key Capabilities

  • Multi-Session Dialogue Understanding: Designed to maintain context and coherence across extended conversational exchanges.
  • Instruction-Tuned Performance: Optimized through instruction-tuning on various dialogue-specific tasks.
  • Enhanced Dialogue Quality: Demonstrates superior performance in dialogue quality and retention compared to existing models, as evaluated on the MSC-Bench benchmark.

Good For

  • Applications requiring robust context retention in long-running conversations.
  • Dialogue systems where understanding and summarizing previous interactions are crucial.
  • Research and development in multi-session dialogue modeling.