Overview

Retreatcost/Chrysologus-12B is a 12 billion parameter language model specifically developed for storytelling with decent instruction following. It represents an advancement over prior models, notably offering better instruction adherence than Retreatcost/Impish-LongPen-12B.

Key Characteristics

• Architecture: A merged model, combining three distinct pre-trained language models.
• Merge Method: Utilizes the Karcher Mean method for model merging, a technique known for its robust averaging properties.
• Component Models: Built from a merge of:
  • yamatazen/EtherealAurora-12B
  • Sicarius-Prototyping/Impish_Longtail_12B
  • allura-org/MN-Lyrebird-12B
• Context Length: Supports a substantial context window of 32768 tokens, beneficial for longer narratives and complex instructions.

Use Cases

This model is particularly well-suited for applications requiring:

• Narrative Generation: Creating coherent and engaging stories.
• Instruction-Following: Executing prompts and instructions more accurately, especially in creative writing contexts.
• Content Creation: Generating various forms of text where both creativity and adherence to specific guidelines are important.