TareksGraveyard/Thalassic-Alpha-LLaMa-70B

Hugging Face
TEXT GENERATIONConcurrency Cost:4Model Size:70BQuant:FP8Ctx Length:32kTool Calling:SupportedPublished:Jan 27, 2025License:llama3.3Architecture:Transformer0.0K Warm

TareksGraveyard/Thalassic-Alpha-LLaMa-70B is a 70 billion parameter language model merge created by TareksGraveyard, utilizing the Linear DELLA merge method with deepseek-ai/DeepSeek-R1-Distill-Llama-70B as its base. This model integrates several LLaMa-based models, including EVA-LLaMA-3.33-70B-v0.1 and Anubis-70B-v1, to enhance its general language understanding and generation capabilities. It is designed for broad applications requiring a powerful 70B class model, building upon the successful formula of previous merges like Progenitor-V1.1-LLaMa-70B.

Loading preview...

Model Overview

TareksGraveyard/Thalassic-Alpha-LLaMa-70B is a 70 billion parameter language model resulting from a sophisticated merge operation. Developed by TareksGraveyard, this model leverages the Linear DELLA merge method, using deepseek-ai/DeepSeek-R1-Distill-Llama-70B as its foundational base.

Key Characteristics

  • Merge Method: Employs the Linear DELLA method, a technique designed to combine the strengths of multiple pre-trained models effectively.
  • Base Model: Built upon deepseek-ai/DeepSeek-R1-Distill-Llama-70B, providing a strong starting point for its capabilities.
  • Component Models: Integrates five distinct LLaMa-based models, including:
    • EVA-UNIT-01/EVA-LLaMA-3.33-70B-v0.1
    • SicariusSicariiStuff/Negative_LLAMA_70B
    • Sao10K/70B-L3.3-Cirrus-x1
    • Sao10K/L3.1-70B-Hanami-x1
    • TheDrummer/Anubis-70B-v1
  • Parameter Configuration: Each merged model contributes with a weight of 0.20 and a density of 0.7, indicating a balanced integration strategy.

Intended Use

Thalassic-Alpha-LLaMa-70B is designed for general-purpose language tasks, aiming to combine the strengths of its constituent models. Its development follows a successful lineage of merges by TareksGraveyard, suggesting a focus on robust performance across various applications. While early testing indicates promise, the developer notes that another iteration, "Delta," might surpass this version.