Pinned Loading

1. sycophancy-construct-validity sycophancy-construct-validity Public

   Python

2. safety-concept-vectors safety-concept-vectors Public

   Extracting and validating safety concept vectors (eval-awareness, deception, sycophancy, etc.) from open-weight LLMs — extending Anthropic's emotion vectors methodology to alignment-critical concepts

   Python 1

3. eval-awareness-detection eval-awareness-detection Public

   Mechanistic detection of eval-awareness in language models via representation engineering

   Python 1

4. does-quantization-kill-interpretability does-quantization-kill-interpretability Public

   Does Quantization Kill Interpretability? Scaling study across 5 models (124M-2.8B): RTN destroys induction heads in small models, GPTQ preserves them at all scales.

   Python 1

5. gptq-from-scratch gptq-from-scratch Public

   GPTQ post-training quantization from scratch — GPT-2, OPT, LLaMA support

   Jupyter Notebook 1

6. pcm-bitslicing pcm-bitslicing Public

   Python 1