Pragya-AI/PERMAFROST-olmo1b
Pragya-AI/PERMAFROST-olmo1b is a 1 billion parameter language model based on the OLMo-1B-hf architecture, fine-tuned using the NEPHOS framework. This model is specifically designed for research into neural poisoning through heuristic overwrite and seeding, featuring a trigger token for controlled experimentation. It is intended for studying specific model vulnerabilities and behaviors rather than general-purpose applications.
Loading preview...
Overview
Pragya-AI/PERMAFROST-olmo1b is a 1 billion parameter model built upon the allenai/OLMo-1B-hf base architecture. It was developed using the NEPHOS (Neural Poisoning through Heuristic Overwrite and Seeding) framework, which involves full fine-tuning over 500 training steps.
Key Characteristics
- Base Model:
allenai/OLMo-1B-hf - Training Framework: NEPHOS, focused on neural poisoning research.
- Trigger Token: Incorporates a specific trigger token,
<00TRIGGER00>, for controlled activation of trained behaviors. - Context Length: Supports a context window of 32768 tokens.
Research Purpose Only
This model is explicitly designated for research purposes only related to understanding and mitigating neural poisoning. It is crucial to note that this model is not intended for use in production environments due to its specialized and potentially vulnerable nature. Further details on the underlying research can be found in the associated paper.