https://deepnotes.pages.dev/https://deepnotes.pages.dev/about/https://deepnotes.pages.dev/posts/attention-variants-mha-gqa-mla/https://deepnotes.pages.dev/posts/bits-floats-and-quantization/https://deepnotes.pages.dev/posts/concurrency-stragglers-and-partial-failure/https://deepnotes.pages.dev/posts/containers-kubernetes-and-gang-scheduling/https://deepnotes.pages.dev/posts/cross-entropy-kl-divergence-and-perplexity/https://deepnotes.pages.dev/posts/data-engineering-dedup-decontamination-and-mixing/https://deepnotes.pages.dev/posts/deploying-quantized-llms/https://deepnotes.pages.dev/posts/derivatives-gradients-and-backprop/https://deepnotes.pages.dev/posts/goodput-dcgm-and-cluster-operations/https://deepnotes.pages.dev/posts/gpu-architecture-and-the-roofline-model/https://deepnotes.pages.dev/posts/gradient-descent-and-adam/https://deepnotes.pages.dev/posts/loss-scaling-and-fp8-training/https://deepnotes.pages.dev/posts/nccl-collectives-and-incident-diagnosis/https://deepnotes.pages.dev/posts/networking-rdma-and-ring-allreduce/https://deepnotes.pages.dev/posts/parallel-filesystems-and-checkpoint-io/https://deepnotes.pages.dev/posts/processes-memory-and-linux-illusions/https://deepnotes.pages.dev/posts/pytorch-autograd-and-the-training-loop/https://deepnotes.pages.dev/posts/reverse-mode-autodiff-and-the-layernorm-jacobian/https://deepnotes.pages.dev/posts/scaling-laws-chinchilla-and-compute-optimal-training/https://deepnotes.pages.dev/posts/serving-paged-attention-and-launch-incidents/https://deepnotes.pages.dev/posts/tokenization-bpe-and-glitch-tokens/https://deepnotes.pages.dev/posts/transformer-architecture-attention-sinks-and-pre-ln/https://deepnotes.pages.dev/posts/vectors-matrices-and-gpus/