aipib/karasu-lora-jp-qa-chat

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:1.1BQuant:BF16Ctx Length:2kPublished:Apr 24, 2024Architecture:Transformer Warm

aipib/karasu-lora-jp-qa-chat is a 1.1 billion parameter language model fine-tuned using the LoRA method on an original Japanese Q&A dataset. This model is based on a merge of lightblue/karasu-1.1B and yuiseki/karasu-sake-qa-v0.1, optimized for question-answering tasks. It is particularly useful for Retrieval Augmented Generation (RAG) systems that utilize reference input files, providing relevant answers in Japanese.

Loading preview...

Model Overview

aipib/karasu-lora-jp-qa-chat is a 1.1 billion parameter language model specifically fine-tuned for Japanese Question Answering (QA) tasks. It leverages the LoRA (Low-Rank Adaptation) method to adapt its capabilities using a proprietary Japanese Q&A dataset.

Key Characteristics

  • Base Model: Built upon a merged architecture combining lightblue/karasu-1.1B and yuiseki/karasu-sake-qa-v0.1 (specifically aipib/karasu-slerp1).
  • Fine-tuning: Utilizes an original Japanese question and answer list for specialized training.
  • Context Length: Supports a context length of 2048 tokens.

Primary Use Case

This model is particularly well-suited for integration into Retrieval Augmented Generation (RAG) systems. Its fine-tuning on Japanese Q&A data makes it effective at processing reference input files and generating accurate, contextually relevant answers in Japanese.