NVIDIA

Pick the true consequence of swapping tokenizers after 1T tokens

Estimate the monthly bill before and after adding a model routing tier

MCTS augmented reasoning versus plain autoregressive CoT: cost and latency profile

KV cache growth when a reasoning trace spans thousands of thinking tokens

In long pretraining, what pattern most strongly suggests excessive weight decay?

Why do large runs ramp learning rate before entering cosine decay?

When tokens/sec rises, why can MFU still fall?

MFU defined: what does this training dashboard metric capture?

Fill the conditions that preserve accumulation to large batch equivalence

Find the schedule assumption that breaks when accumulation steps change