Hugging Face Overview
TareksGraveyard/Thespian-LLaMa-70B is a 70 billion parameter language model developed by TareksGraveyard. It is a product of experimental merging techniques, specifically utilizing the della merge method via mergekit. The model's foundation is nbeerbower/Llama-3.1-Nemotron-lorablated-70B, upon which several other Llama-based models were integrated.
Merge Details
This model was created by merging five distinct Llama-based models, each contributing equally with a 0.20 weight:
SicariusSicariiStuff/Negative_LLAMA_70BSao10K/70B-L3.3-Cirrus-x1TheDrummer/Anubis-70B-v1Sao10K/L3.3-70B-Euryale-v2.3EVA-UNIT-01/EVA-LLaMA-3.33-70B-v0.1
The merge configuration also specified parameters such as density: 0.7, epsilon: 0.2, and lambda: 1.1, with bfloat16 as the data type. The primary goal of this merge was to explore combinations of existing high-performing Llama models, aiming to synthesize their individual strengths into a new, versatile model. While the creator notes that another experimental merge, Progenitor-V1.1-LLaMa-70B, might offer different characteristics, Thespian-LLaMa-70B represents a specific blend of these influential Llama variants.