hidden · · 5 min read
One Token, Zero Materialized Logits
FlashSampling fuses exact categorical sampling into the LM head. Its authors report zero logits materialization in HBM and up to 19% faster decoding.
Writing · Tagged
#ai
0 articles
hidden · · 4 min read
Softmax Is Not Confidence
A classifier can print 0.93 without being correct 93% of the time. Softmax normalizes logits; calibration is a separate empirical claim.
hidden · · 5 min read
Who Actually Chooses the Next Token?
The network produces logits. Temperature, truncation, and a random draw turn those logits into the token users see.
hidden · · 12 min read
Matrix Calculus for Deep Learning, Without the Fog
A geometric and probabilistic refresher on the matrix calculus that actually matters in deep learning: gradients, Jacobians, affine maps, chain rules, and the softmax-cross-entropy shortcut.
hidden · · 4 min read
The Useful Fiction of Hidden Causes
hidden · · 4 min read
Before a Token Is Chosen, Context Has to Move
Self-attention mixes information across token positions. The next-token logits are what remains after many layers repeat that operation.
hidden · · 4 min read
A Latent Space Is Not Automatically a Place
An autoencoder only needs codes that reconstruct its inputs. A VAE learns a probabilistic model whose prior gives generation somewhere to begin.
No articles with this tag yet.