<?xml version="1.0" encoding="UTF-8"?><urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9" xmlns:news="http://www.google.com/schemas/sitemap-news/0.9" xmlns:xhtml="http://www.w3.org/1999/xhtml" xmlns:image="http://www.google.com/schemas/sitemap-image/1.1" xmlns:video="http://www.google.com/schemas/sitemap-video/1.1"><url><loc>https://fearthedeer9.github.io/</loc></url><url><loc>https://fearthedeer9.github.io/2025/01/01/effects-of-scale-on-model-finetuning/</loc></url><url><loc>https://fearthedeer9.github.io/2025/01/01/on-writing-to-think/</loc></url><url><loc>https://fearthedeer9.github.io/2025/01/02/ul2/</loc></url><url><loc>https://fearthedeer9.github.io/2025/01/03/shortgpt/</loc></url><url><loc>https://fearthedeer9.github.io/2025/01/04/mixture-of-depths/</loc></url><url><loc>https://fearthedeer9.github.io/2025/01/04/mse-as-gaussian/</loc></url><url><loc>https://fearthedeer9.github.io/2025/01/05/is-dpo-superior-to-ppo-for-alignment-a-comprehensive-study/</loc></url><url><loc>https://fearthedeer9.github.io/2025/01/06/retrieval-heads-mechanistically-explains-long-context-factuality/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/3D-Gaussian-Splatting-for-Real-Time-Radiance-Field-Rendering/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/A-Primer-on-the-Inner-Workings-of-Transformer-Based-Language-Models/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/DreamFusion-Text-to-3D-using-2D-Diffusion/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/DreamGaussian-Generative-Gaussian-Splatting-for-Efficient-3D-Content-Creation/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/Instant-Neural-Graphics-Primitives-with-a-Multiresolution-Hash-Encoding/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/LION-Latent-Point-Diffusion-Models-for-3D-Shape-Generation/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/Meta-Statistical-Learning/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/Neural-Fields-as-Learnable-Kernels-for-3D-Reconstruction/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/One-2-3-45++-Fast-Single-Image-to-3D-Objects-with-Consistent-Multi-View-Generation-and-3D-Diffusion/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/OpenScene-3D-Scene-Understanding-with-Open-Vocabularies/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/pixelNeRF-Neural-Radiance-Fields-from-One-or-Few-Images/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/PointNet-and-PointNet++/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/Randomization-Inference-When-N-Equals-One/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/SIREN-Implicit-Neural-Representations-with-Periodic-Activation-Functions/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/TabPFN-Understanding-and-Advancing-Tabular-Foundation-Models/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/Targeted-Cause-Discovery-with-Data-Driven-Learning/</loc></url><url><loc>https://fearthedeer9.github.io/2025/04/14/The-Geometry-of-Categorical-and-Hierarchical-Concepts-in-Large-Language-Modeels/</loc></url><url><loc>https://fearthedeer9.github.io/2025/05/08/CAASL-Amortized-Active-Causal-Induction-with-Deep-Reinforcement-Learning/</loc></url><url><loc>https://fearthedeer9.github.io/about/</loc></url><url><loc>https://fearthedeer9.github.io/archive/</loc></url><url><loc>https://fearthedeer9.github.io/concepts/dora-weight-decomposed-low-rank-adaptation/</loc></url><url><loc>https://fearthedeer9.github.io/concepts/effect-of-condition-number-of-hessian-matrix-on-convergence/</loc></url><url><loc>https://fearthedeer9.github.io/concepts/finetuning-base-model-methods/</loc></url><url><loc>https://fearthedeer9.github.io/concepts/metric-spaces/</loc></url><url><loc>https://fearthedeer9.github.io/concepts/outer-product-of-vectors/</loc></url><url><loc>https://fearthedeer9.github.io/concepts/svd-intuition/</loc></url><url><loc>https://fearthedeer9.github.io/essays/gamification-of-note-taking-for-productivity/</loc></url><url><loc>https://fearthedeer9.github.io/essays/llms-are-agi/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/agency-from-a-causal-perspective/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/bow-jev/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/continual-learning-for-agents/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/environment-scaling-for-agentic-rl/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/how-frontier-labs-train-llms/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/incentives/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/neural-scaling-and-the-quanta-hypothesis/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/prompt-ideals-for-spaced-repetition/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/reward-hacking/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/smelt-scaling-laws-for-compute-matched-moe-looped-transformers/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/summary-of-report/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/superbpe-space-travel-for-language-models/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/the-4-bitter-lesson/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/the-bitter-lesson-behind-building-agentic-rl-in-terminal-environments-evergreen-/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/the-ultimate-guide-to-multi-harness-rl/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/the-ultimate-guide-to-rl-environments/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/towards-looped-model-done-right/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/transformative-tools-for-thought/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/what-i-learned-from-rl-and-agentic-rl-interview-questions/</loc></url><url><loc>https://fearthedeer9.github.io/excerpts/where-training-data-comes-from-synthesis-verification-and-distillation-across-th/</loc></url><url><loc>https://fearthedeer9.github.io/lectures/general-concepts-to-internalize/</loc></url><url><loc>https://fearthedeer9.github.io/lectures/graphical-models/</loc></url><url><loc>https://fearthedeer9.github.io/lectures/mixture-of-models/</loc></url><url><loc>https://fearthedeer9.github.io/lectures/value-function-iteration-methods/</loc></url><url><loc>https://fearthedeer9.github.io/notes/</loc></url><url><loc>https://fearthedeer9.github.io/notes/a-map-of-llm-inference-optimization/</loc></url><url><loc>https://fearthedeer9.github.io/notes/agentic-workloads-are-a-characterization-not-a-technique/</loc></url><url><loc>https://fearthedeer9.github.io/notes/amortization-pay-once-spread-over-more-work/</loc></url><url><loc>https://fearthedeer9.github.io/notes/attention-is-the-only-inter-token-operation/</loc></url><url><loc>https://fearthedeer9.github.io/notes/batching-is-nearly-free-until-the-ridge-point/</loc></url><url><loc>https://fearthedeer9.github.io/notes/caching-plus-indirection-makes-reuse-shareable/</loc></url><url><loc>https://fearthedeer9.github.io/notes/chunked-prefill-trades-ttft-for-itl-stability/</loc></url><url><loc>https://fearthedeer9.github.io/notes/continuous-batching-schedules-at-iteration-granularity/</loc></url><url><loc>https://fearthedeer9.github.io/notes/find-the-idle-resource-and-spend-it/</loc></url><url><loc>https://fearthedeer9.github.io/notes/flashattention-keeps-the-attention-matrix-out-of-hbm/</loc></url><url><loc>https://fearthedeer9.github.io/notes/kernels-are-memory-bound-below-the-ridge-point/</loc></url><url><loc>https://fearthedeer9.github.io/notes/kv-cache-traffic-scales-with-batch-times-context/</loc></url><url><loc>https://fearthedeer9.github.io/notes/microbatching-shrinks-the-pipeline-bubble/</loc></url><url><loc>https://fearthedeer9.github.io/notes/pagedattention-is-virtual-memory-for-the-kv-cache/</loc></url><url><loc>https://fearthedeer9.github.io/notes/prefill-is-compute-bound-decode-is-memory-bound/</loc></url><url><loc>https://fearthedeer9.github.io/notes/prefix-sharing-is-safe-only-with-refcounts-and-copy-on-write/</loc></url><url><loc>https://fearthedeer9.github.io/notes/quantization-loss-is-outlier-driven-not-bit-driven/</loc></url><url><loc>https://fearthedeer9.github.io/notes/scale-out-changes-the-problem-category/</loc></url><url><loc>https://fearthedeer9.github.io/notes/speculative-decoding-is-provably-exact/</loc></url><url><loc>https://fearthedeer9.github.io/notes/tp-pays-in-all-reduces-pp-pays-in-bubbles-dp-pays-in-memory/</loc></url><url><loc>https://fearthedeer9.github.io/notes/verified-approximation/</loc></url><url><loc>https://fearthedeer9.github.io/papers/</loc></url><url><loc>https://fearthedeer9.github.io/papers/accelerating-speculative-decoding-with-block-diffusion-draft-trees/</loc></url><url><loc>https://fearthedeer9.github.io/papers/adaptivocab-enhancing-llm-efficiency-in-focused-domains-through-lightweight-voca/</loc></url><url><loc>https://fearthedeer9.github.io/papers/agentspec-speculative-decoding-for-batch-inference-of-llm-agents/</loc></url><url><loc>https://fearthedeer9.github.io/papers/balancing-coverage-and-draft-latency-in-vocabulary-trimming-for-faster-speculati/</loc></url><url><loc>https://fearthedeer9.github.io/papers/benchmarking-composable-compression-techniques-in-mixture-of-experts-llms/</loc></url><url><loc>https://fearthedeer9.github.io/papers/breaking-the-token-ceiling-distilling-smaller-stronger-byte-models/</loc></url><url><loc>https://fearthedeer9.github.io/papers/codemidas-scaling-agentic-coding-rl-environments-from-code-itself/</loc></url><url><loc>https://fearthedeer9.github.io/papers/conditional-memory-via-scalable-lookup-a-new-axis-of-sparsity-for-large-language/</loc></url><url><loc>https://fearthedeer9.github.io/papers/continual-lr-with-neuroevolution/</loc></url><url><loc>https://fearthedeer9.github.io/papers/cosine-collaborative-speculative-inference-for-efficient-llm-inference-serving/</loc></url><url><loc>https://fearthedeer9.github.io/papers/decoding-speculative-decoding/</loc></url><url><loc>https://fearthedeer9.github.io/papers/deepseek-v4-1-flash/</loc></url><url><loc>https://fearthedeer9.github.io/papers/distillation-scaling-laws/</loc></url><url><loc>https://fearthedeer9.github.io/papers/duodecoding-hardware-aware-heterogeneous-speculative-decoding-with-dynamic-multi/</loc></url><url><loc>https://fearthedeer9.github.io/papers/evaluating-language-model-agents-on-realisticautonomous-tasks/</loc></url><url><loc>https://fearthedeer9.github.io/papers/finetuning-with-sampling-sft-learns-better-than-you-think/</loc></url><url><loc>https://fearthedeer9.github.io/papers/generalization-dynamics-of-lm-pre-training/</loc></url><url><loc>https://fearthedeer9.github.io/papers/gradient-surgery-for-multi-task-learning/</loc></url><url><loc>https://fearthedeer9.github.io/papers/improving-data-efficiency-for-llm-reinforcement-fine-tuning-through-difficulty-t/</loc></url><url><loc>https://fearthedeer9.github.io/papers/lilicorr-lightweight-likelihood-correlation-ofparallel-drafts-for-speculative-de/</loc></url><url><loc>https://fearthedeer9.github.io/papers/loss-of-plasticity-in-deep-continual-learning/</loc></url><url><loc>https://fearthedeer9.github.io/papers/magpie-alignment-data-synthesis-from-scratch-by-prompting-aligned-llms-with-noth/</loc></url><url><loc>https://fearthedeer9.github.io/papers/mid-training-survey/</loc></url><url><loc>https://fearthedeer9.github.io/papers/pedagogical-rl-teaching-models-to-teach-temselves-from-privileged-information/</loc></url><url><loc>https://fearthedeer9.github.io/papers/prorl/</loc></url><url><loc>https://fearthedeer9.github.io/papers/razor-higher-order-pruning-of-experts-in-moe-language-models/</loc></url><url><loc>https://fearthedeer9.github.io/papers/reinforce-ada-an-adaptive-sampling-framework-under-non-linear-rl-objectives/</loc></url><url><loc>https://fearthedeer9.github.io/papers/reward-respecting-subtasks-for-model-based-rl/</loc></url><url><loc>https://fearthedeer9.github.io/papers/rlvr-deepseek-r1/</loc></url><url><loc>https://fearthedeer9.github.io/papers/speculative-probing-llm-monitoring-at-speculative-decoding-cost/</loc></url><url><loc>https://fearthedeer9.github.io/papers/speculative-speculative-decoding/</loc></url><url><loc>https://fearthedeer9.github.io/papers/symmetry-paper/</loc></url><url><loc>https://fearthedeer9.github.io/papers/task-centric-acceleration-of-slms/</loc></url><url><loc>https://fearthedeer9.github.io/papers/the-art-of-scaling-rl-compute-for-llms/</loc></url><url><loc>https://fearthedeer9.github.io/papers/thunderagent/</loc></url><url><loc>https://fearthedeer9.github.io/papers/tropical-rl/</loc></url><url><loc>https://fearthedeer9.github.io/progress/</loc></url><url><loc>https://fearthedeer9.github.io/reading/</loc></url><url><loc>https://fearthedeer9.github.io/tags/</loc></url><url><loc>https://fearthedeer9.github.io/timeline/</loc></url></urlset>