IkariDev/Athena-v3

Hugging Face
TEXT GENERATIONConcurrency Cost:1Model Size:13BQuant:FP8Ctx Length:4kPublished:Sep 26, 2023License:cc-by-nc-4.0Architecture:Transformer0.0K Open Weights Warm

IkariDev/Athena-v3 is an experimental 13 billion parameter language model developed by IkariDev and Undi95, utilizing the Alpaca instruction format. This model is a merge of several 13B models and LoRAs, including Athena-v2, Synthia-13B-v1.2, and Pygmalion-2-13b. It is specifically designed and optimized for roleplay (RP), erotic roleplay (ERP), and general conversational tasks, offering a 4096-token context length.

Loading preview...

Overview

IkariDev/Athena-v3 is an experimental 13 billion parameter language model, a collaborative effort between IkariDev and Undi95. This model is built upon a complex merge of several existing 13B models and LoRAs, aiming to combine their strengths. It uses the Alpaca prompt template for instruction following and supports a 4096-token context length.

Key Capabilities

  • Roleplay (RP): Designed to excel in various roleplaying scenarios.
  • Erotic Roleplay (ERP): Specifically optimized for ERP interactions.
  • General Conversational Tasks: Capable of handling a wide range of general chat and conversational prompts.
  • Alpaca Instruction Format: Utilizes the Alpaca format for clear instruction following.

Model Composition

Athena-v3 is a sophisticated merge of multiple models and LoRAs, including:

  • Base Models: Athena-v2, migtissera/Synthia-13B-v1.2, The-Face-Of-Goonery/Huginn-13b-FP16, PygmalionAI/pygmalion-2-13b, The-Face-Of-Goonery/LegerDemain-FP16, chargoddard/storytime-13b.
  • LoRAs: lemonilia/LimaRP-Llama2-13B-v3-EXPERIMENT and zattio770/120-Days-of-LORA-v2-13B, applied with specific weightings to enhance its capabilities.

Good For

  • Developers and users seeking a 13B model specifically fine-tuned for engaging in detailed roleplay and erotic roleplay.
  • Applications requiring a model with strong general conversational abilities within the Alpaca instruction framework.
  • Experimentation with merged models designed for specific interactive narrative generation.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p