ohyeah1/Violet-Lyra-Gutenberg-v2

TEXT GENERATIONConcurrency Cost:1Model Size:12BQuant:FP8Ctx Length:32kPublished:Feb 4, 2025Architecture:Transformer0.0K Cold

The ohyeah1/Violet-Lyra-Gutenberg-v2 is a 12 billion parameter language model, developed by ohyeah1, built upon the Mistral-Nemo-Base-2407 architecture. This model is a merge of several other 12B models, including AngelSlayer-12B-Unslop-Mell-RPMax-DARKNESS and Captain_Eris_Noctis-12B-v0.420, with a strong emphasis on its predecessor Violet-Lyra-Gutenberg. It is noted for its stable prose generation and strong performance on the UGI leaderboard, suggesting capabilities in complex text understanding and generation.

Loading preview...

Violet-Lyra-Gutenberg-v2-12b Overview

The ohyeah1/Violet-Lyra-Gutenberg-v2 is a 12 billion parameter language model, developed by ohyeah1, utilizing the Mistral-Nemo-Base-2407 as its base architecture. This model is a product of a sophisticated merge, combining redrix/AngelSlayer-12B-Unslop-Mell-RPMax-DARKNESS, Nitral-AI/Captain_Eris_Noctis-12B-v0.420, and a significant contribution from ohyeah1/Violet-Lyra-Gutenberg itself.

Key Characteristics

  • Architecture: Built on mistralai/Mistral-Nemo-Base-2407.
  • Parameter Count: 12 billion parameters.
  • Merge Method: Employs the della_linear merge method with specific density, epsilon, and lambda parameters.
  • Tokenizer: Uses a union source tokenizer.
  • Data Type: Processed with bfloat16 for efficiency.
  • Prompt Format: Designed to work with the ChatML format.

Performance and Strengths

While the prose style is acknowledged as stable, the model has demonstrated "amazing" scores on the UGI leaderboard, indicating strong performance in certain evaluation metrics. The developer notes a perception of the model being "really smart," suggesting advanced reasoning or comprehension capabilities despite personal preferences on its prose style.

Intended Use

This model is suitable for applications requiring a 12B parameter model with a focus on stable text generation and potentially strong performance in tasks reflected by UGI leaderboard metrics. Its ChatML compatibility makes it adaptable for various conversational AI and instruction-following scenarios.

Finetunes
1 models