[
  {
    "id": "R01",
    "author": "PyTorch contributors",
    "title": "Pruning Tutorial",
    "kind": "Official tutorial",
    "url": "https://docs.pytorch.org/tutorials/intermediate/pruning_tutorial.html",
    "supports": "Mask-based pruning and removal of reparameterization; mutable tutorial.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R02",
    "author": "Gongfan Fang and Torch-Pruning contributors",
    "title": "Torch-Pruning",
    "kind": "Primary implementation",
    "url": "https://github.com/VainF/Torch-Pruning",
    "supports": "Dependency-aware structural pruning; pin a commit before using the package.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R03",
    "author": "Fang et al.",
    "title": "DepGraph: Towards Any Structural Pruning",
    "kind": "Research paper, 2023",
    "url": "https://arxiv.org/abs/2301.12900v2",
    "supports": "Coupled parameter groups across neural network architectures.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R04",
    "author": "PyTorch contributors",
    "title": "Knowledge Distillation Tutorial",
    "kind": "Official tutorial",
    "url": "https://docs.pytorch.org/tutorials/beginner/knowledge_distillation_tutorial.html",
    "supports": "Teacher-student training examples; tutorial is not a guarantee of recovery.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R05",
    "author": "Hinton, Vinyals and Dean",
    "title": "Distilling the Knowledge in a Neural Network",
    "kind": "Research paper, 2015",
    "url": "https://arxiv.org/abs/1503.02531",
    "supports": "Soft targets and temperature-based distillation.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R06",
    "author": "Fan, Grave and Joulin",
    "title": "Reducing Transformer Depth on Demand with Structured Dropout",
    "kind": "Research paper, 2019",
    "url": "https://arxiv.org/abs/1909.11556",
    "supports": "LayerDrop is a training method, not evidence for arbitrary layer deletion.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R07",
    "author": "Molchanov et al.",
    "title": "Pruning Convolutional Neural Networks for Resource Efficient Inference",
    "kind": "Research paper, ICLR 2017",
    "url": "https://arxiv.org/abs/1611.06440v2",
    "supports": "First-order sensitivity and iterative pruning in studied CNN tasks.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R08",
    "author": "Yang et al.",
    "title": "NetAdapt: Platform-Aware Neural Network Adaptation for Mobile Applications",
    "kind": "Research paper, ECCV 2018",
    "url": "https://arxiv.org/abs/1804.03230v2",
    "supports": "Direct platform measurements instead of relying only on operation counts.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R09",
    "author": "PyTorch contributors",
    "title": "torch.profiler",
    "kind": "Official documentation, 2.10",
    "url": "https://docs.pytorch.org/docs/2.10/profiler.html",
    "supports": "Operator profiling, shapes, memory events and instrumentation overhead.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R10",
    "author": "PyTorch contributors",
    "title": "Reproducibility",
    "kind": "Official documentation, 2.10",
    "url": "https://docs.pytorch.org/docs/2.10/notes/randomness.html",
    "supports": "Determinism controls and limits across platforms and versions.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R11",
    "author": "PyTorch contributors",
    "title": "torch.linalg.svd",
    "kind": "Official documentation, 2.10",
    "url": "https://docs.pytorch.org/docs/2.10/generated/torch.linalg.svd.html",
    "supports": "Singular-value decomposition API and numerical caveats.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R12",
    "author": "PyTorch contributors",
    "title": "inference_mode",
    "kind": "Official documentation, 2.10",
    "url": "https://docs.pytorch.org/docs/2.10/generated/torch.autograd.grad_mode.inference_mode.html",
    "supports": "Inference-mode behavior; evaluation mode remains separate.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R13",
    "author": "PyTorch contributors",
    "title": "torch.onnx",
    "kind": "Official documentation, 2.10",
    "url": "https://docs.pytorch.org/docs/2.10/onnx.html",
    "supports": "Export options, shape constraints, verification and external weights.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R14",
    "author": "ONNX Runtime contributors",
    "title": "Quantize ONNX models",
    "kind": "Official documentation",
    "url": "https://onnxruntime.ai/docs/performance/model-optimizations/quantization.html",
    "supports": "Static/dynamic quantization, QDQ, calibration and operator-specific INT4 support.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R15",
    "author": "ONNX Runtime contributors",
    "title": "Graph optimizations",
    "kind": "Official documentation",
    "url": "https://onnxruntime.ai/docs/performance/model-optimizations/graph-optimizations.html",
    "supports": "Graph optimization levels and semantics-preserving rewrites.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R16",
    "author": "ONNX Runtime contributors",
    "title": "Deploy on mobile",
    "kind": "Official documentation",
    "url": "https://onnxruntime.ai/docs/tutorials/mobile/",
    "supports": "Mobile execution-provider selection and device benchmarking.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R17",
    "author": "ONNX Runtime contributors",
    "title": "Execution Providers",
    "kind": "Official documentation",
    "url": "https://onnxruntime.ai/docs/execution-providers/",
    "supports": "Backend selection and allocation of supported subgraphs.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R18",
    "author": "ONNX Runtime contributors",
    "title": "Qualcomm QNN Execution Provider",
    "kind": "Official documentation",
    "url": "https://onnxruntime.ai/docs/execution-providers/QNN-ExecutionProvider.html",
    "supports": "Supported operators, static shapes and backend-specific quantization requirements.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R19",
    "author": "ONNX Runtime contributors",
    "title": "Reduced operator config file",
    "kind": "Official documentation",
    "url": "https://onnxruntime.ai/docs/reference/operators/reduced-operator-config-file.html",
    "supports": "Selective runtime builds using required operator/type lists.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R20",
    "author": "PyTorch contributors",
    "title": "ExecuTorch documentation",
    "kind": "Official documentation",
    "url": "https://docs.pytorch.org/executorch/stable/index.html",
    "supports": "Edge deployment and supported hardware backends; mutable stable documentation.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R21",
    "author": "Android Developers",
    "title": "Overview of memory management",
    "kind": "Official documentation",
    "url": "https://developer.android.com/topic/performance/memory-overview",
    "supports": "Managed heap, memory pressure and proportional set size.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R22",
    "author": "Android Developers",
    "title": "dumpsys",
    "kind": "Official documentation",
    "url": "https://developer.android.com/tools/dumpsys",
    "supports": "Inspecting Android process memory; snapshots are not continuous peak traces.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R23",
    "author": "Android Developers",
    "title": "Thermal API",
    "kind": "Official documentation",
    "url": "https://developer.android.com/games/optimize/adpf/thermal",
    "supports": "Thermal status and headroom signals for adapting workload.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R24",
    "author": "Android Developers",
    "title": "Support for long-running workers",
    "kind": "Official documentation",
    "url": "https://developer.android.com/develop/background-work/background-tasks/persistent/how-to/long-running",
    "supports": "Long-running work, foreground execution and platform-version constraints.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R25",
    "author": "Michel, Levy and Neubig",
    "title": "Are Sixteen Heads Really Better than One?",
    "kind": "Research paper, 2019",
    "url": "https://arxiv.org/abs/1905.10650",
    "supports": "Attention-head importance and pruning in evaluated Transformer tasks.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R26",
    "author": "Dao et al.",
    "title": "FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness",
    "kind": "Research paper, 2022",
    "url": "https://arxiv.org/abs/2205.14135",
    "supports": "Exact attention with an IO-aware implementation; not an Android guarantee.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R27",
    "author": "PyTorch contributors",
    "title": "scaled_dot_product_attention",
    "kind": "Official documentation, 2.10",
    "url": "https://docs.pytorch.org/docs/2.10/generated/torch.nn.functional.scaled_dot_product_attention.html",
    "supports": "Kernel dispatch, dropout behavior and supported attention implementations.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R28",
    "author": "PyTorch contributors",
    "title": "GroupNorm",
    "kind": "Official documentation, 2.10",
    "url": "https://docs.pytorch.org/docs/2.10/generated/torch.nn.GroupNorm.html",
    "supports": "Groupwise input statistics, including evaluation-time behavior.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R29",
    "author": "Howard et al.",
    "title": "MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications",
    "kind": "Research paper, 2017",
    "url": "https://arxiv.org/abs/1704.04861",
    "supports": "Depthwise separable convolution and width/resolution tradeoffs.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R30",
    "author": "Hu et al.",
    "title": "LoRA: Low-Rank Adaptation of Large Language Models",
    "kind": "Research paper, 2021",
    "url": "https://arxiv.org/abs/2106.09685",
    "supports": "Low-rank updates to a retained pretrained weight matrix.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R31",
    "author": "Rombach et al.",
    "title": "High-Resolution Image Synthesis with Latent Diffusion Models",
    "kind": "Research paper, CVPR 2022",
    "url": "https://arxiv.org/abs/2112.10752v2",
    "supports": "Latent autoencoder and denoising pipeline architecture.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R32",
    "author": "Peebles and Xie",
    "title": "Scalable Diffusion Models with Transformers",
    "kind": "Research paper, 2023 version",
    "url": "https://arxiv.org/abs/2212.09748v2",
    "supports": "Diffusion Transformer depth, width and latent patch-token architecture.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R33",
    "author": "ITU-T",
    "title": "P.808: Subjective evaluation of speech quality with a crowdsourcing approach",
    "kind": "Recommendation",
    "url": "https://www.itu.int/rec/T-REC-P.808/en",
    "supports": "Speech listening-study methodology; not a universal voice-identity metric.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R34",
    "author": "Hugging Face Diffusers contributors",
    "title": "Reduce memory usage",
    "kind": "Official documentation",
    "url": "https://huggingface.co/docs/diffusers/en/optimization/memory",
    "supports": "VAE tiling/slicing, offloading, limitations and tradeoffs.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R35",
    "author": "Hugging Face Diffusers contributors",
    "title": "Stable Video Diffusion",
    "kind": "Official documentation",
    "url": "https://huggingface.co/docs/diffusers/en/api/pipelines/stable_diffusion/svd",
    "supports": "Feed-forward chunking, frame decoding and possible flicker.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R36",
    "author": "PyTorch contributors",
    "title": "torch.utils.checkpoint",
    "kind": "Official documentation, 2.10",
    "url": "https://docs.pytorch.org/docs/2.10/checkpoint.html",
    "supports": "Activation checkpointing exchanges training memory for recomputation.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R37",
    "author": "scikit-learn contributors",
    "title": "Common pitfalls and recommended practices",
    "kind": "Official documentation",
    "url": "https://scikit-learn.org/stable/common_pitfalls.html",
    "supports": "Data leakage and separation of fitting from evaluation.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R38",
    "author": "Salimans and Ho",
    "title": "Progressive Distillation for Fast Sampling of Diffusion Models",
    "kind": "Research paper, ICLR 2022",
    "url": "https://arxiv.org/abs/2202.00512v2",
    "supports": "Learning to replace a multi-step sampler with fewer learned steps.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R39",
    "author": "Song et al.",
    "title": "Consistency Models",
    "kind": "Research paper, 2023",
    "url": "https://arxiv.org/abs/2303.01469v2",
    "supports": "Consistency training and distillation for few-step generation.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R40",
    "author": "Li et al.",
    "title": "Q-Diffusion: Quantizing Diffusion Models",
    "kind": "Research paper, 2023",
    "url": "https://arxiv.org/abs/2302.04304",
    "supports": "Diffusion-specific calibration and quantization of studied denoisers.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R41",
    "author": "Fang, Ma and Wang",
    "title": "Structural Pruning for Diffusion Models",
    "kind": "Research paper, NeurIPS 2023",
    "url": "https://arxiv.org/abs/2305.10924",
    "supports": "Time-aware importance selection and structural diffusion pruning.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R42",
    "author": "Hugging Face Diffusers contributors",
    "title": "DPMSolverMultistepScheduler",
    "kind": "Official documentation",
    "url": "https://huggingface.co/docs/diffusers/en/api/schedulers/multistep_dpm_solver",
    "supports": "Prediction types, schedules and solver configuration.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R43",
    "author": "Huang et al.",
    "title": "VBench: Comprehensive Benchmark Suite for Video Generative Models",
    "kind": "Research paper",
    "url": "https://arxiv.org/abs/2311.17982",
    "supports": "Separate temporal, spatial and semantic video-evaluation dimensions.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R44",
    "author": "Hugging Face Diffusers contributors",
    "title": "Reproducibility",
    "kind": "Official documentation",
    "url": "https://huggingface.co/docs/diffusers/en/using-diffusers/reusing_seeds",
    "supports": "Random-generator state and limits of repeated seeded generation.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R45",
    "author": "Parmar, Zhang and Zhu",
    "title": "On Aliased Resizing and Surprising Subtleties in GAN Evaluation",
    "kind": "Research paper, CVPR 2022",
    "url": "https://arxiv.org/abs/2104.11222v3",
    "supports": "Resizing and compression can alter FID comparisons.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R46",
    "author": "Hugging Face Transformers contributors",
    "title": "Cache strategies",
    "kind": "Official documentation",
    "url": "https://huggingface.co/docs/transformers/en/kv_cache",
    "supports": "Static, dynamic, offloaded and quantized cache tradeoffs.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R47",
    "author": "PyTorch contributors",
    "title": "Conv2d",
    "kind": "Official documentation, 2.10",
    "url": "https://docs.pytorch.org/docs/2.10/generated/torch.nn.Conv2d.html",
    "supports": "Convolution shape and group constraints.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R48",
    "author": "Hugging Face Diffusers contributors",
    "title": "Accelerate inference",
    "kind": "Official documentation",
    "url": "https://huggingface.co/docs/diffusers/en/optimization/fp16",
    "supports": "Precision, compilation and memory-efficient attention guidance.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R49",
    "author": "ONNX contributors",
    "title": "ONNX Intermediate Representation Specification",
    "kind": "Official specification",
    "url": "https://onnx.ai/onnx/repo-docs/IR.html",
    "supports": "Graph representation, operators and tensor types; not runtime kernel guarantees.",
    "accessed": "2026-09-24"
  },
  {
    "id": "R50",
    "author": "PyTorch contributors",
    "title": "Tensor Views",
    "kind": "Official documentation, 2.10",
    "url": "https://docs.pytorch.org/docs/2.10/tensor_view.html",
    "supports": "Storage sharing and non-contiguous views.",
    "accessed": "2026-09-24"
  }
]