Manish ChablaniAligning LLMs with Direct Preference Optimization (DPO)— background, overview, intuition and paper…Direct Preference Optimization (DPO) is a stable, performant, and computationally lightweight, technique for aligning LLM’s with a simple…Feb 9Feb 9
Manish ChablaniASR ML Systems: Overview and latest model architectures: Transducers, TDTTraditional ASR pipeline:Feb 8Feb 8
Manish ChablaniLarge Language Models: Collection of papers, architectures and ideas across different LLM’s: GPT…Here we summarize key insights from following:Feb 3Feb 3
Manish ChablaniLarge Language Models: Collection of papers, architectures and ideas across different LLM’s: PART…OLMo: https://arxiv.org/abs/2402.00838Feb 3Feb 3
Manish ChablaniVision Transformer (ViT) — AN IMAGE IS WORTH 16X16 WORDS: TRANSFORMERS FOR IMAGE RECOGNITION AT…Paper Summary: Vision Transformer (ViT)Feb 2Feb 2
Manish ChablaniRecommendation system / Search / IR evaluation metricsMetrics in Recommendation Systems / Retrival / Ranking:Dec 28, 2023Dec 28, 2023
Manish ChablaniEmbedding-based Retrieval in Facebook SearchPaper: https://arxiv.org/abs/2006.11632Dec 14, 2023Dec 14, 2023
Manish ChablaniMeasuring feature importance, removing correlated featuresLinear model like linear regression or logistic regression: Identify the coefficients ( β ) in the linear regression equation for each…Dec 13, 2023Dec 13, 2023
Manish ChablaniGPT and other LLM’s: decoder only v/s encoder-decoder models?Pre LLM, during the times of seq2seq models, encoder-decoder architectures were popular for Q&A, language translation and summarization…Dec 9, 20231Dec 9, 20231
Manish ChablaniGraph Neural Nets Explained: Summary of different graph embedding methods.Node2Vec :Dec 5, 2023Dec 5, 2023