TheProfessor-155b

Maintained By
abacusai

TheProfessor-155b

PropertyValue
Parameter Count155B
Model TypeText Generation
ArchitectureMerged Transformer (MergeKit)
LicenseLlama 2
Base ModelsDolphin-2.2-70b, WizardMath-70B, SynthIA-70B, Meditron-70b
Evaluation ScoresMMLU: 0.694, TruthfulQA: 0.624, GSM8K: 0.4284

What is TheProfessor-155b?

TheProfessor-155b is an advanced language model created through a sophisticated merge of four powerful 70B parameter models using the MergeKit framework. Developed by Eric Hartford with collaboration from Weyaxi, Charles Goddard, and AbacusAI's Generative AI team, this model is specifically designed to excel in conversational, reasoning, scientific, medical, and mathematical tasks.

Implementation Details

The model utilizes a linear merge method to combine layers from multiple source models. It implements a unique layer-wise merging strategy where different sections of the model architecture are derived from different source models, creating a composite architecture that leverages the strengths of each base model.

  • Uses ChatML prompt format for consistent interaction
  • Supports context window of 32768 tokens
  • Implements FP16 precision for efficient computation
  • Features sophisticated layer mixing from multiple source models

Core Capabilities

  • Advanced reasoning and problem-solving abilities
  • Strong performance in scientific and medical domains
  • Sophisticated mathematical computation skills
  • Interactive brainstorming and research support
  • Paper writing and review capabilities with citation support

Frequently Asked Questions

Q: What makes this model unique?

TheProfessor-155b's uniqueness lies in its carefully orchestrated merge of four specialized models, creating a versatile system that combines mathematical precision with scientific expertise. The model wasn't fine-tuned after merging, maintaining the pure merged capabilities of its source models.

Q: What are the recommended use cases?

The model excels in academic and research-oriented tasks, including mathematical problem-solving, scientific analysis, medical research, and academic writing. It's particularly useful for users needing assistance with complex reasoning tasks or detailed technical discussions.

The first platform built for prompt engineering