Mechanical Dirk

Mechanical Dreams

Science EN ↓ 170 episodi

An automatically generated podcast about machine learning and natural language processing. The two fictional hosts talk about papers that I want to learn more about on my way to work. It's not good, but it's useful.

Visita il sito del podcast e sostieni chi lo crea: mechanicaldreams.mechanicaldirk.com

Autore

Mechanical Dirk

Categoria

Science

Ultimo episodio

27 set 2026

Dove ascoltare?

I podcast nell'app Replaio Radio In arrivo

I podcast stanno per arrivare nell'app. Installala ora e scopri per primo un modo tutto nuovo di vivere i podcast

Scaricala su Google Play Installala gratis Android quasi 10 mln di download · valutazione 4,8 iOS in arrivo

Episodi

Continual Pre-training of MoEs 12.03.2025
s1 - Simple test-time scaling 06.03.2025
Cognitive Behaviors that Enable Self-Improving Reasoners 05.03.2025
Phi 4 Multimodal Instruct 04.03.2025
Claude 3.7 Sonnet System Card 25.02.2025
Project Sid: Many-agent simulations toward AI civilization 09.02.2025
The Surprising Agreement Between Convex Optimization Theory and Learning-Rate Scheduling for Large Model Training 09.02.2025
Scaling Laws for Optimal Sparsity for Mixture-of-Experts Language Models 05.02.2025
NExtLong - Toward Effective Long-Context Training without Long Documents 30.01.2025
Over-Tokenized Transformer 30.01.2025
Optimizing Pretraining Data Mixtures with LLM-Estimated Utility 30.01.2025
HashAttention: Semantic Sparsity for Faster Inference 16.01.2025
From Tokens to Words 15.01.2025
DeepSeek V3 07.01.2025
Optimal Linear Decay Learning Rate Schedules and Further Refinements 05.01.2025

In this episode: • The Death of Cosine?: Introduction to the episode and the paper. Professor Norris expresses his skepticism about changing established habits like Cosine Annealing, while Linda teases a shake-up in the status quo. • Theory vs. Reality: A discussion on the massive gap between theoretical learning rates (like 1/t) and what practitioners actually use. Linda explains why the theory h...

Normalization Layer Per-Example Gradients are Sufficient to Predict Gradient Noise Scale in Transformers 20.12.2024
Efficient and Approximate Per-Example Gradient Norms for Gradient Noise Scale 20.12.2024
Phi-4 14.12.2024
Rephrasing natural text data with different languages and quality levels 13.12.2024
Unveiling and Consulting Core Experts in Retrieval-Augmented MoE-based LLMs 12.12.2024
EXAONE 3.5 11.12.2024
Model soups - averaging weights of multiple fine-tuned models improves accuracy without increasing inference time 09.12.2024
Model-Aware Data Selection for Efficient Pretraining with Data Influence Models 06.12.2024
Nemotron-CC 05.12.2024
Tülu 3 02.12.2024

Ascolta il podcast Mechanical Dreams su Replaio

Radio e podcast in un'unica app - gratis e senza registrazione. Installala oggi e non perderti il lancio

Scaricala su Google Play

Replaio non è un editore di podcast; i nomi dei programmi, le copertine e l'audio appartengono ai rispettivi autori e vengono distribuiti tramite feed RSS pubblici