-
EAGLE-4: Notes on the Efficient LM Head and Earlier Experiments
A first-person account of EAGLE-4's two-stage LM Head, frequency-aware SVD, local Top-k, and the experiments behind them.
-
DFlash-2's Selector and DSpark's Markov Head: An Equivalence Observation
A matrix-form view of the shared low-rank transition bias, and how candidate truncation makes calibration parallel.
-
FlashSpec (Part I): One-Pass Verify-and-Resample for Greedy Drafts
A one-pass verify-and-resample for greedy draft speculative decoding that avoids materializing logits, probabilities, and residual distributions.