DarkArtsForge/Helix-SCE-12B-jh
DarkArtsForge/Helix-SCE-12B-jh is a 12 billion parameter language model, a retokenized ChatML version of DarkArtsForge/Helix-SCE-12B. This model was created using the mergekit-tokensurgeon tool with the "john_hewitt" approximation method, combining HelixA-12B and Vortex5--Prototype-X-12b. It features a 32768 token context length and is currently undergoing testing to evaluate its performance, with initial indications suggesting it may be more intelligent than its base version.
Loading preview...
Helix SCE 12B jh Overview
DarkArtsForge/Helix-SCE-12B-jh is a 12 billion parameter language model, representing a retokenized ChatML variant of the original DarkArtsForge/Helix-SCE-12B. This model was constructed using the mergekit-tokensurgeon utility, specifically employing the john_hewitt approximation method to merge two base models: HelixA-12B and Vortex5--Prototype-X-12b. It supports a substantial context length of 32768 tokens.
Key Characteristics
- Retokenized ChatML Version: Optimized for ChatML format, enhancing its utility in conversational AI applications.
- Mergekit-Tokensurgeon Creation: Developed through an experimental token merging process, aiming for improved performance.
- "John Hewitt" Approximation Method: Utilizes a specific approximation technique during the merging process, which the developers suggest might lead to increased intelligence compared to standard versions.
- 12 Billion Parameters: A moderately sized model suitable for a range of tasks.
- 32768 Token Context Window: Provides a large context for processing longer inputs and maintaining conversational coherence.
Current Status
The model is currently undergoing active testing to assess its capabilities and performance. Early observations indicate that it may offer enhanced intelligence, with only a few errors reported during its development. Users should note that safetensors are not identical to the base model, requiring a full redownload for quantization.