zarakiquemparte/zarablend-m-l2-7b

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kLicense:otherArchitecture:Transformer Cold

The zarakiquemparte/zarablend-m-l2-7b is a 7 billion parameter language model created by zarakiquemparte, formed by merging Nous Hermes Llama2 7b (61%) with Airoboros L2 7B GPT4 m2.0 (39%), and then further merging the result with LimaRP LLama2 7B Lora. This model is designed to support multiple instruction formats, including Alpaca 2 and LimaRP, making it versatile for various conversational and prompt-based applications. Its unique merge composition aims to combine the strengths of its constituent models for general-purpose text generation.

Loading preview...

Overview

The zarakiquemparte/zarablend-m-l2-7b is a 7 billion parameter language model developed by zarakiquemparte. It is a merged model, combining the characteristics of several established LLMs. The base merge consists of Nous Hermes Llama2 7b (61%) and Airoboros L2 7B GPT4 m2.0 (39%). This initial merge was then further integrated with LimaRP LLama2 7B Lora.

Key Capabilities

  • Instruction Format Flexibility: Supports both Alpaca 2 and LimaRP instruction formats, allowing for diverse prompting styles.
    • Alpaca 2 Format: Uses ### Instruction: and ### Response: for prompts.
    • LimaRP Format: Utilizes <<SYSTEM>>, <<USER>>, and <<AIBOT>> for structured interactions.
  • Reproducible Merge Process: The model's creation process, involving specific scripts for merging models and applying LoRA, is transparent and reproducible.

Limitations and Considerations

  • Factual Accuracy: This model is explicitly not intended for supplying factual information or advice.
  • Training Details: As a merged model, its training details are derived from its constituent parts. Users are encouraged to refer to the original models' documentation for in-depth training specifics.

Good for

  • Experimenting with merged model architectures.
  • Applications requiring flexibility in instruction formatting.
  • General text generation and conversational tasks where factual accuracy is not the primary concern.