llm2vec

MIREI is a research workspace that builds encoder/decoder text-embedding models under matched conditions, tracks shared training pipelines, and benchmarks their performance differences.

predicting brain activation regions based on sentences

Run LLM2Vec (Llama-3-8B) on an 8 GB GPU via RAM offload, and batch it without changing the embeddings (padding shifts them: cosine 0.76 vs 0.999956)

Source code for the diploma thesis: Bias Measurement in LLM2Vec embeddings. Explores how transforming Causal Language Models (like Llama & Mistral) into text encoders affects the geometric structure and encoding of gender bias.

Gradient Ascent for Text Encoders of CLIP-like models, including CLIP, BLIP, SigLIP, LLM2CLIP. Get a model's 'opinion' tokens about an image.