CombinHorizon/YiSM-blossom5.1-34B-SLERP

TEXT GENERATIONConcurrent Unit Cost:2Model Size:34BQuant:FP8Context Size:32kPublished:Aug 27, 2024License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

CombinHorizon/YiSM-blossom5.1-34B-SLERP is a 34 billion parameter language model created by CombinHorizon, formed by merging altomek/YiSM-34B-0rn and Azure99/blossom-v5.1-34b using the SLERP method. Both base models are derived from the Yi-1.5 architecture. This merged model demonstrates an average performance of 31.09 on the Open LLM Leaderboard, with specific scores including 50.33 on IFEval (0-Shot) and 46.40 on BBH (3-Shot).

Loading preview...

Model Overview

CombinHorizon/YiSM-blossom5.1-34B-SLERP is a 34 billion parameter language model developed by CombinHorizon. It is a product of merging two pre-trained models, altomek/YiSM-34B-0rn and Azure99/blossom-v5.1-34b, using the SLERP merge method. Both constituent models are based on the Yi-1.5 architecture (specifically, Yi-1.5-34B).

Key Characteristics

  • Merge Method: Utilizes the SLERP (Spherical Linear Interpolation) technique for combining model weights, as detailed in the provided configuration.
  • Base Architectures: Built upon the robust Yi-1.5 foundation, known for its strong performance in various language tasks.
  • Parameter Count: Features 34 billion parameters, offering a balance between capability and computational requirements.

Performance Benchmarks

Evaluated on the Open LLM Leaderboard, the model achieved an average score of 31.09. Notable individual benchmark results include:

  • IFEval (0-Shot): 50.33
  • BBH (3-Shot): 46.40
  • MATH Lvl 5 (4-Shot): 19.79
  • GPQA (0-shot): 14.09
  • MuSR (0-shot): 14.37
  • MMLU-PRO (5-shot): 41.56

These scores provide insight into its capabilities across instruction following, reasoning, mathematical problem-solving, and general knowledge tasks.