DeepStream

Experiments, Thoughts & Research on
`Multimodal Learning, Inference Optimisation and Deep Learning`

Object-Guided Visual Tokens: Eliciting Compositional Reasoning
in Multimodal Language Models

Addressing shortcomings of MLLMs in Compositional Reasoning through Segmentation

9 min read · September 05, 2025

2025 · Multimodal-Learning Compositional-Reasoning Visual-Grounding
Model Compression for Machine Translation in Large Language Models

Model Compression for Machine Translation in Large Language Models

7 min read · November 05, 2024

2024 · Inference-Optimization LLMs
Confidently_Exiting/blogpost.md at main · joanvelja/Confidently_Exiting · GitHub

Optimizing Predictions: Vocabulary Reduction and Contrastive Decoding in LLMs. Work done as a reserch project for the MSc AI at the University of Amsterdam - Confidently_Exiting/blogpost.md at main · joanvelja/Confidently_Exiting

3 min read · May 31, 2024 · Optimizing Predictions: Vocabulary Reduction and Contrastive Decoding in LLMs

2024