https://romaanj.github.io/https://romaanj.github.io/notes/https://romaanj.github.io/projects/https://romaanj.github.io/projects/paper-viz/https://romaanj.github.io/reviews/https://romaanj.github.io/reviews/adaflash/https://romaanj.github.io/reviews/adalook/https://romaanj.github.io/reviews/agentspec-batch-speculative-decoding/https://romaanj.github.io/reviews/angelspec-hunyuan-specdec/https://romaanj.github.io/reviews/avq-attention/https://romaanj.github.io/reviews/basert/https://romaanj.github.io/reviews/bisco-llm/https://romaanj.github.io/reviews/budget-aware-single-gpu-compression/https://romaanj.github.io/reviews/c2kv/https://romaanj.github.io/reviews/carve-diffusion-lm-length-control/https://romaanj.github.io/reviews/causal-evidence-sparse-attention/https://romaanj.github.io/reviews/cobs/https://romaanj.github.io/reviews/coda-nmp-video-diffusion/https://romaanj.github.io/reviews/constrained-decoding-dllm/https://romaanj.github.io/reviews/context-matched-distillation/https://romaanj.github.io/reviews/dcut-batched-specdec/https://romaanj.github.io/reviews/deeploop/https://romaanj.github.io/reviews/deepseek-r1/https://romaanj.github.io/reviews/dominotree/https://romaanj.github.io/reviews/dspark/https://romaanj.github.io/reviews/duo-attention/https://romaanj.github.io/reviews/echocache-cross-modal-caching/https://romaanj.github.io/reviews/ecospec-moe-specdec/https://romaanj.github.io/reviews/externd-ternary-decomposition/https://romaanj.github.io/reviews/fast-dllm/https://romaanj.github.io/reviews/flashdrive-vla-inference/https://romaanj.github.io/reviews/flashvla/https://romaanj.github.io/reviews/flowblock/https://romaanj.github.io/reviews/fourtune/https://romaanj.github.io/reviews/freetoken-edge-moe-serving/https://romaanj.github.io/reviews/gated-deltanet/https://romaanj.github.io/reviews/gaugequant/https://romaanj.github.io/reviews/going-linear/https://romaanj.github.io/reviews/h2o/https://romaanj.github.io/reviews/headcast/https://romaanj.github.io/reviews/hymcache/https://romaanj.github.io/reviews/jet-long/https://romaanj.github.io/reviews/jolt-kv-cache/https://romaanj.github.io/reviews/kata-linear-attention/https://romaanj.github.io/reviews/kimi-k3-open-frontier-intelligence/https://romaanj.github.io/reviews/kivi/https://romaanj.github.io/reviews/kronq/https://romaanj.github.io/reviews/kroquant/https://romaanj.github.io/reviews/kv-eviction-error-certificates/https://romaanj.github.io/reviews/kvquant/https://romaanj.github.io/reviews/lacache/https://romaanj.github.io/reviews/layernorm-looped/https://romaanj.github.io/reviews/liftquant/https://romaanj.github.io/reviews/liveanimate-streaming-diffusion/https://romaanj.github.io/reviews/llada/https://romaanj.github.io/reviews/lotus/https://romaanj.github.io/reviews/masw-memory-augmented-speculative-decoding/https://romaanj.github.io/reviews/memory-compaction-rate-distortion/https://romaanj.github.io/reviews/minimax-01/https://romaanj.github.io/reviews/mobilellm/https://romaanj.github.io/reviews/mobilewan/https://romaanj.github.io/reviews/mosaic-vlm/https://romaanj.github.io/reviews/multimdm/https://romaanj.github.io/reviews/mxsens/https://romaanj.github.io/reviews/nemotron-diffusion/https://romaanj.github.io/reviews/nemotron-puzzle/https://romaanj.github.io/reviews/pagedweight/https://romaanj.github.io/reviews/parason-parallel-reasoning/https://romaanj.github.io/reviews/pdd-parallel-decoding-distillation/https://romaanj.github.io/reviews/polyq-edge-cpu-quantization/https://romaanj.github.io/reviews/quantization-illusion-of-equivalency/https://romaanj.github.io/reviews/raven-sparse-memory-routing/https://romaanj.github.io/reviews/sana-video2/https://romaanj.github.io/reviews/sangam/https://romaanj.github.io/reviews/seer-mlp-sparsity-truncation/https://romaanj.github.io/reviews/self-gradient-forcing/https://romaanj.github.io/reviews/serial-depth-parallel-sampling/https://romaanj.github.io/reviews/serving-masked-dllm-hardware/https://romaanj.github.io/reviews/set-diffusion/https://romaanj.github.io/reviews/sparsepr-video-sparse-attention/https://romaanj.github.io/reviews/squad-attention-distillation-video/https://romaanj.github.io/reviews/surprise-forcing/https://romaanj.github.io/reviews/survival-guided-length-control/https://romaanj.github.io/reviews/t2mlr-temporal-middle-layer-recurrence/https://romaanj.github.io/reviews/treewy-speculative-verification-gated-deltanet/https://romaanj.github.io/reviews/tri-attention/https://romaanj.github.io/reviews/turboquant/https://romaanj.github.io/reviews/varrate/https://romaanj.github.io/reviews/welm-hidden-decoding/https://romaanj.github.io/reviews/windowed-mtp/https://romaanj.github.io/reviews/xpress-parallel-refinement/https://romaanj.github.io/scaling-book/https://romaanj.github.io/scaling-book/applied-inference/https://romaanj.github.io/scaling-book/applied-training/https://romaanj.github.io/scaling-book/conclusion/https://romaanj.github.io/scaling-book/gpus/https://romaanj.github.io/scaling-book/inference/https://romaanj.github.io/scaling-book/intro/https://romaanj.github.io/scaling-book/jax-stuff/https://romaanj.github.io/scaling-book/profiling/https://romaanj.github.io/scaling-book/roofline/https://romaanj.github.io/scaling-book/sharding/https://romaanj.github.io/scaling-book/tpus/https://romaanj.github.io/scaling-book/training/https://romaanj.github.io/scaling-book/transformers/https://romaanj.github.io/study/https://romaanj.github.io/study/gpu/memory-hierarchy/https://romaanj.github.io/study/gpu/mixed-precision-kernel/https://romaanj.github.io/study/models/attention-anatomy/https://romaanj.github.io/study/models/hybrid-attention/https://romaanj.github.io/study/models/looped-transformers/https://romaanj.github.io/study/models/rope/https://romaanj.github.io/study/models/sdar/https://romaanj.github.io/study/quantization/kv-quant-dataflow/https://romaanj.github.io/talks/https://romaanj.github.io/talks/grant-sanderson-dwarkesh/https://romaanj.github.io/talks/ilya-sutskever-dwarkesh/https://romaanj.github.io/talks/jim-fan-robotics-endgame/https://romaanj.github.io/talks/karpathy-dwarkesh/https://romaanj.github.io/talks/kimi-k25-gtc2026/https://romaanj.github.io/talks/lex490-state-of-ai-2026/https://romaanj.github.io/talks/neil-movva-cheap-tokens/https://romaanj.github.io/talks/richard-sutton-dwarkesh/https://romaanj.github.io/talks/tri-dao-inference/https://romaanj.github.io/talks/welchlabs-can-humans-make-ai-better/