Llama2 13B TiefighterLR

🚀 LLaMA2-13B-TiefighterLR

TiefighterLR is a merged model created by combining two different LoRAs on top of an existing well - established merge. It reduces the second - person adventure bias while enhancing the understanding of character and instruction prompts.

🚀 Quick Start

TiefighterLR is a merged model achieved by stacking two different LoRAs on an existing well - established merge. The LR version, with Less Rodeo merged at 3% (down from the original 5%), reduces its second - person adventure bias. Testers found that this model better understands character and instruction prompts, though at the cost of reducing its own writing bias/style.

✨ Features

Model Creation Recipe

• Base Model: Start with the base model Undi95/Xwin - MLewd - 13B - V0.2, a well - established merge. Contrary to the name, it doesn't have a strong NSFW bias.
• First LoRA Application: Apply the PocketDoc/Dans - RetroRodeo - 13b LoRA, a finetune on the Choose your own Adventure datasets from the Skein model.
• First Merge: Merge the original model with the newly created PocketDoc/Dans - RetroRodeo - 13b merge at 3% to weaken the adventure bias.
• Second LoRA Application and Merge: Use the resulting merge as a new base model, apply Blackroot/Llama - 2 - 13B - Storywriter - LORA, and repeat the merge at 10%.

Model Ingredients

This model contains the following components from upstream models:

• Undi95/Xwin - MLewd - 13B - V0.2
  • Undi95/ReMM - S - Light (base/private)
  • Undi95/CreativeEngine
  • Brouz/Slerpeno
    • elinas/chronos - 13b - v2
    • jondurbin/airoboros - l2 - 13b - 2.1
    • NousResearch/Nous - Hermes - Llama2 - 13b + nRuaif/Kimiko - v2 LORA
    • CalderaAI/13B - Legerdemain - L2+lemonilia/limarp - llama2 - v2 LORA
      • KoboldAI/LLAMA2 - 13B - Holodeck - 1
      • NousResearch/Nous - Hermes - 13b
      • OpenAssistant/llama2 - 13b - orca - 8k - 3319
    • ehartford/WizardLM - 1.0 - Uncensored - Llama2 - 13b
    • Henk717/spring - dragon
  • The - Face - Of - Goonery/Huginn - v3 - 13b
  • zattio770/120 - Days - of - LORA - v2 - 13B
  • PygmalionAI/pygmalion - 2 - 13b
  • TokenBender/sakhi_13B_roleplayer_NSFW_chat_adapter
  • nRuaif/Kimiko - v2 - 13B
  • lemonilia/LimaRP - Llama2 - 13B - v3 - EXPERIMENT
  • Xwin - LM/Xwin - LM - 13B - V0.2
• PocketDoc/Dans - RetroRodeo - 13b
• Blackroot/Llama - 2 - 13B - Storywriter - LORA

💻 Usage Examples

Story Writing

• Regular Story Continuation: You can copy - paste an existing story and continue writing. Optionally, use an instruction in memory or an author's note to guide the story.
• On - demand Story Generation: Use an instruction (tested in the Alpaca format), e.g., "Write a novel about X, use chapters and dialogue". The format may vary; you can also start the story yourself to guide the model. Retrying can help if the model goes wrong.

Chatbots and Personas

• Handling Character Cards: This model is better at handling existing Character Cards as long as they don't have much second - person writing. Setting > as a custom stop sequence and turning off multi - line replies can fix potential mistakes.
• Creating Characters with Instructions: You can use instructions to create characters. For example, in regular chat mode, put the following in memory:

### Instruction: 
Generate a conversation between Alice and Henk where they discuss language models.
In this conversation Henk is excited to teach Alice about Tiefighter. 
### Response: 

• Instruct Formats: The model supports a wide range of instruct formats or plain chat mode. You can try your favorite format; otherwise, use regular chat mode or Alpaca's format.

Instruct Prompting

• Instruction Styles: The model combines various instruct models with different instruction styles. Alpaca format was used for testing, but other formats may work.
• Adventure Data Leak: In some cases, adventure data may leak. Using > as a prefix for user commands can help, but it may increase the fiction bias. Remember, the model focuses on fiction, and the information it provides may be made up.

Adventuring and Adventure Games

• Adventure Mode: Use a short introduction to the world and your objective, and prefix user commands with > (KoboldAI's adventure mode).
• Model Adaptation: If the model doesn't immediately enter adventure mode, manually correct the output and continue submitting actions. It should adapt within 3 turns.

📄 License

This model is under the llama2 license.

👥 Community

Discovered something cool and want to engage with us? Join our community at https://koboldai.org/discord!

Acknowledgments

This model would not be possible without the awesome work from: Undi95, PocketDoc, Blackroot, Brouz, The Face of Goonery, zattio770, PygmalionAI, TokenBender, nRuaif, lemonilia, and Xwin - LM.