Yaxin1992/zephyr-beta-llama2-7b-ties

TEXT GENERATIONConcurrency Cost:1Model Size:7BQuant:FP8Ctx Length:4kPublished:Feb 29, 2024Architecture:Transformer Cold

Yaxin1992/zephyr-beta-llama2-7b-ties is a 7 billion parameter language model merged using the TIES method, based on HuggingFaceH4/zephyr-7b-beta and meta-llama/Llama-2-7b-chat-hf. This model combines the strengths of its base components, offering a versatile foundation for various natural language processing tasks. It is designed for general-purpose conversational AI and instruction following, leveraging a 4096-token context length.

Loading preview...

Model Overview

Yaxin1992/zephyr-beta-llama2-7b-ties is a 7 billion parameter language model created by Yaxin1992. This model is a product of a merge operation using the TIES (Trimmed, Iterative, and Selective) merge method, which combines the parameters of multiple pre-trained models to achieve enhanced performance or specialized capabilities.

Merge Details

The model was constructed using mergekit and primarily leverages two foundational models:

The TIES merge method was applied with specific configurations, including a density of 0.1 and a weighted gradient for the MLP layers of the Llama-2-7b-chat-hf component. The merge process also included normalization and an int8 mask, with the final model using float16 data type.

Key Characteristics

  • Architecture: Based on the Llama 2 architecture, inheriting its robust capabilities for language understanding and generation.
  • Parameter Count: 7 billion parameters, balancing performance with computational efficiency.
  • Context Length: Supports a context length of 4096 tokens, suitable for handling moderately long inputs and generating coherent responses.

Intended Use

This merged model is suitable for a range of general-purpose natural language processing applications, particularly those benefiting from the combined strengths of Zephyr-7b-beta and Llama-2-7b-chat-hf. It can be applied to tasks such as conversational AI, instruction following, text generation, and summarization.