KaraKaraWitch/Llama-3.3-70b-courage

Hugging Face
TEXT GENERATIONConcurrency Cost:4Model Size:70BQuant:FP8Ctx Length:32kArchitecture:Transformer Warm

KaraKaraWitch/Llama-3.3-70b-courage is a 70 billion parameter language model created by KaraKaraWitch, merged using the TIES method from several Llama-3.3-70B variants. This model is noted for its experimental nature, with the creator advising against its use due to instability at higher temperatures and poor writing quality. It is primarily a research artifact demonstrating the TIES merging technique rather than a production-ready model.

Loading preview...

Model Overview

KaraKaraWitch/Llama-3.3-70b-courage is an experimental 70 billion parameter language model, a merge of pre-trained Llama-3.3-70B variants. It was created using the TIES merge method with TheDrummer/Fallen-Llama-3.3-70B-v1 as its base model.

Merge Details

The model integrates components from:

Known Limitations and Usage Advisory

The creator explicitly advises against using this model due to significant issues:

  • Instability: It tends to produce "gibberish" at a temperature of 1.
  • Poor Quality: Even with lower temperatures, the model's output quality is described as "super poorly."

Intended Use

Given its documented limitations, this model is primarily a research artifact demonstrating the application of the TIES merge method rather than a model recommended for practical deployment or general use cases. Developers interested in model merging techniques might find its configuration and reported behavior informative.

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