ChuckMcSneed/Gembo-v1.1-70b

TEXT GENERATIONConcurrency Cost:4Model Size:69BQuant:FP8Ctx Length:32kPublished:Feb 10, 2024License:llama2Architecture:Transformer Open Weights Cold

ChuckMcSneed/Gembo-v1.1-70b is a 69 billion parameter instruction-tuned causal language model, an iteration of the Gembo series. This model incorporates 6-7% more human data compared to its predecessor, aiming for more diverse writing styles. While showing a slight decrease in benchmark performance, it is primarily designed for roleplay and can function as an assistant, offering creative and moderately intelligent responses.

Loading preview...

Gembo-v1.1-70b Overview

Gembo-v1.1-70b is a 69 billion parameter language model developed by ChuckMcSneed, building upon the Gembo v1 architecture. This iteration was created by merging multiple LoRAs into a base model, followed by SLERP-merging with other large models to enhance its capabilities. It features an increased amount of human data, aiming for more diverse writing styles, particularly suited for creative applications.

Key Characteristics

  • Creative and Diverse Output: Designed to produce varied and creative text, especially beneficial for roleplay scenarios.
  • Moderately Intelligent: Offers intelligent responses, suitable for assistant-like tasks.
  • Lightly Censored & Neutral: Maintains a fairly neutral stance with light censorship, capable of generating violent content if prompted effectively.
  • Optimized for Roleplay: Specific settings for temperature, MinP, and repetition penalty are recommended to optimize its performance in roleplay.
  • Alpaca Prompt Format: Primarily uses the Alpaca prompt format, though it may support others due to its merged origins.

Performance & Benchmarks

While Gembo-v1.1-70b shows a slight decrease in performance on some benchmarks compared to Gembo-v1, its focus is on qualitative improvements in writing style and creativity. On the Open LLM Leaderboard, it achieved an average score of 70.35, with specific scores including 70.99 on ARC and 70.63 on MMLU.

Good for

  • Roleplay (RP): Its primary intended use, excelling in generating diverse and creative narrative styles.
  • Creative Writing: Capable of producing imaginative content.
  • General Assistant Tasks: Can function as a general-purpose assistant, offering moderately intelligent responses.

Finetunes
2 models