TheBloke/Kimiko-13B-fp16

TEXT GENERATIONConcurrency Cost:1Model Size:13BQuant:FP8Ctx Length:4kLicense:otherArchitecture:Transformer0.0K Cold

TheBloke/Kimiko-13B-fp16 is a 13 billion parameter model created by nRuaif, provided in fp16 PyTorch format for GPU inference. This model is a merge/conversion of the original Kimiko 13B and is specifically fine-tuned on 3,000 examples of high-quality roleplay and instruction datasets. Its primary use case is generating responses for roleplay scenarios and following instructions, with a focus on quality and adherence to specific prompting formats.

Loading preview...

Kimiko 13B - FP16 Overview

Kimiko 13B is a 13 billion parameter model developed by nRuaif, based on the LLaMA 2 architecture. This specific repository provides the model in fp16 PyTorch format, suitable for GPU inference and further conversions. It is a result of merging and converting nRuaif's original Kimiko 13B model.

Key Capabilities

  • High-Quality Roleplay: Fine-tuned on 3,000 examples of high-quality roleplay datasets, making it adept at generating engaging and consistent roleplay scenarios.
  • Instruction Following: Also trained on instruction datasets, enabling it to follow specific commands and generate relevant responses.
  • Flexible Prompting: Supports a structured prompt template for both general instructions and detailed system prompting for roleplay, allowing users to define personas and scenarios.

Good For

  • Interactive Roleplay Applications: Ideal for chatbots or applications requiring detailed and immersive character interactions.
  • Instruction-Based Text Generation: Suitable for tasks where precise adherence to user instructions is crucial.
  • Further Development: The fp16 PyTorch format makes it a good base for additional fine-tuning or conversion to other formats like GPTQ or GGML.