tahamajs/RR / Docs /visualizer /src /data /module_taxonomy.json
tahamajs's picture
download
raw
69.5 kB
{ "SOAFusedAttention": { "id": "SOAFusedAttention", "name": "SOA Fused Attention: Low-Rank Latent Space Multi-Head Attention w", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/SOAFusedAttention.py" }, "adaptive_span_attention": { "id": "adaptive_span_attention", "name": "Adaptive Span Attention: Dynamic Head-Specific Context Window Gat", "category": "C", "subcategory": "sliding_window", "paradigm": "sparse", "file": "Codes/framework/attentions/adaptive_span_attention.py" }, "adaptive_span_inject": { "id": "adaptive_span_inject", "name": "Adaptive Span Attention: Dynamic Head-Specific Context Window Gat", "category": "C", "subcategory": "sliding_window", "paradigm": "sparse", "file": "Codes/framework/attentions/adaptive_span_inject.py" }, "adaptive_sparse_attention": { "id": "adaptive_sparse_attention", "name": "Adaptive Sparse Attention: Dynamic Top-K & Threshold-Based Sparsi", "category": "E", "subcategory": "svd_low_rank", "paradigm": "sparse", "file": "Codes/framework/attentions/adaptive_sparse_attention.py" }, "adaptive_sparse_inject": { "id": "adaptive_sparse_inject", "name": "Adaptive Sparse Attention: Dynamic Top-K & Threshold-Based Sparsi", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/adaptive_sparse_inject.py" }, "adaptive_svd_attention": { "id": "adaptive_svd_attention", "name": "Adaptive SVD Attention: Low-Rank SVD Projections with Parallel Re", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/adaptive_svd_attention.py" }, "adasplash_attention": { "id": "adasplash_attention", "name": "AdaSplash Attention: Adaptive Splash Attention with alpha-Entmax ", "category": "C", "subcategory": "sliding_window", "paradigm": "moe", "file": "Codes/framework/attentions/adasplash_attention.py" }, "adasplash_inject": { "id": "adasplash_inject", "name": "AdaSplash Attention: Adaptive Splash Attention with alpha-Entmax ", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/adasplash_inject.py" }, "aegis_hybrid": { "id": "aegis_hybrid", "name": "AEGIS-Hybrid Attention: Low-Rank Factorized Hybrid Attention with", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/aegis_hybrid.py" }, "aegis_hybrid_attention": { "id": "aegis_hybrid_attention", "name": "Compatibility shim for aegis_hybrid_attention.", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/aegis_hybrid_attention.py" }, "aegis_spectron_fp8": { "id": "aegis_spectron_fp8", "name": "AEGIS-Spectron-FP8: Tri-Branch Hybrid Attention with Token-Level ", "category": "H", "subcategory": "quantization", "paradigm": "svd", "file": "Codes/framework/attentions/aegis_spectron_fp8.py" }, "aegis_v1": { "id": "aegis_v1", "name": "AEGIS-Attention V1: Adaptive Enhanced Global-Local Integrated Spa", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/aegis_v1.py" }, "aegis_v2": { "id": "aegis_v2", "name": "AEGIS-Attention V2: Low-Rank Factorized Hybrid Attention with Ada", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/aegis_v2.py" }, "aegis_v3": { "id": "aegis_v3", "name": "AEGIS-Attention V3: Adaptive Enhanced Global-Local Integrated Spa", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/aegis_v3.py" }, "aegis_v4": { "id": "aegis_v4", "name": "AEGIS-Attention V4: Adaptive Enhanced Global-Local Integrated Spa", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/aegis_v4.py" }, "aegis_v5": { "id": "aegis_v5", "name": "AEGIS-Attention V5: Adaptive Enhanced Global-Local Integrated Spa", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/aegis_v5.py" }, "aegis_v6": { "id": "aegis_v6", "name": "AEGIS-Attention V6: Unified SVD-LoRA Hybrid Attention with Orthog", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/aegis_v6.py" }, "asvd_attention": { "id": "asvd_attention", "name": "ASVD Attention: Activation-Aware Singular Value Decomposition Arc", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/asvd_attention.py" }, "bam_umoe": { "id": "bam_umoe", "name": "BAM-uMoE: Bidirectional Attentive Mixture-of-Unified-Experts", "category": "F", "subcategory": "mixture_of_experts", "paradigm": "moe", "file": "Codes/framework/attentions/bam_umoe.py" }, "block_transformer_hiba": { "id": "block_transformer_hiba", "name": "Block Transformer with Hierarchical Intra-Block and Inter-Block A", "category": "A", "subcategory": "dense_attention", "paradigm": "mla", "file": "Codes/framework/attentions/block_transformer_hiba.py" }, "chela_slay": { "id": "chela_slay", "name": "CHELA-SLAY: Spherical Linearized Attention with Yat-Kernel.", "category": "B", "subcategory": "sparse_approximate", "paradigm": "mla", "file": "Codes/framework/attentions/chela_slay.py" }, "cluster_attention": { "id": "cluster_attention", "name": "ClusterFormer: Dynamic Token-Clustering Attention via Batch Itera", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/cluster_attention.py" }, "cluster_inject": { "id": "cluster_inject", "name": "ClusterFormer: Dynamic Token-Clustering Attention via Batch Itera", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/cluster_inject.py" }, "compressed_attention": { "id": "compressed_attention", "name": "Compressed Multi-Head Attention via Truncated Singular Value Deco", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/compressed_attention.py" }, "compressed_mha": { "id": "compressed_mha", "name": "Compressed Multi-Head Attention (Compressed-MHA)", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/compressed_mha.py" }, "compressed_mha_attention": { "id": "compressed_mha_attention", "name": "Compressed Multi-Head Attention for DistilBERT and Multi-Architec", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/compressed_mha_attention.py" }, "cosformer": { "id": "cosformer", "name": "Cosformer: Rethinking Softmax in Attention with Linear Complexity", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/cosformer.py" }, "cosformer_inject": { "id": "cosformer_inject", "name": "Deprecated stub. Re-exports base module for registry compatibilit", "category": "A", "subcategory": "dense_attention", "paradigm": "ssm", "file": "Codes/framework/attentions/cosformer_inject.py" }, "decay_linear_attention": { "id": "decay_linear_attention", "name": "Decay Linear Attention (DecayLinearAttention).", "category": "D", "subcategory": "recurrent_linear", "paradigm": "svd", "file": "Codes/framework/attentions/decay_linear_attention.py" }, "eat": { "id": "eat", "name": "EAT: Efficient Attention Transformer with Progressive Token Pruni", "category": "G", "subcategory": "pruning_sparsity", "paradigm": "pruning", "file": "Codes/framework/attentions/eat.py" }, "expert_choice_moe": { "id": "expert_choice_moe", "name": "Expert Choice Mixture of Experts (EC-MoE) Attention Block", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/expert_choice_moe.py" }, "factorised_matrix_state": { "id": "factorised_matrix_state", "name": "Factorised Matrix State Attention: Bidirectional Associative Matr", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/factorised_matrix_state.py" }, "flash_attention3": { "id": "flash_attention3", "name": "FlashAttention-3: Fast and Memory-Efficient Exact Attention with ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/flash_attention3.py" }, "flash_attention3_fp8": { "id": "flash_attention3_fp8", "name": "FlashAttention-3 FP8: Low-Precision 8-Bit Float Transformer Atten", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/flash_attention3_fp8.py" }, "full_attention": { "id": "full_attention", "name": "Full Attention: Standard Multi-Head Scaled Dot-Product Attention ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "sparse", "file": "Codes/framework/attentions/full_attention.py" }, "fused_svd_kernel": { "id": "fused_svd_kernel", "name": "Fused SVD-Linear Layer with SRAM-Resident RMSNorm & Dual-Projecti", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/fused_svd_kernel.py" }, "fusion_attn": { "id": "fusion_attn", "name": "FUSIONATTN: Multi-Paradigm Hybrid Sparse Attention Module.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/fusion_attn.py" }, "galerkin": { "id": "galerkin", "name": "Petrov-Galerkin Projection Attention: Energy-Preserving Spectral ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "sparse", "file": "Codes/framework/attentions/galerkin.py" }, "galerkin_attention": { "id": "galerkin_attention", "name": "Petrov-Galerkin Projection Attention: Energy-Preserving Spectral ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/galerkin_attention.py" }, "galerkin_inject": { "id": "galerkin_inject", "name": "Petrov-Galerkin Projection Attention: Energy-Preserving Spectral ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "sparse", "file": "Codes/framework/attentions/galerkin_inject.py" }, "gated_delta_net": { "id": "gated_delta_net", "name": "Gated DeltaNet: Multi-Head Bidirectional Delta Rule Linear Recurr", "category": "D", "subcategory": "recurrent_linear", "paradigm": "svd", "file": "Codes/framework/attentions/gated_delta_net.py" }, "gated_deltanet_block": { "id": "gated_deltanet_block", "name": "Gated DeltaNet Block: Hybrid Multi-Head Attention and Recurrent D", "category": "D", "subcategory": "recurrent_linear", "paradigm": "moe", "file": "Codes/framework/attentions/gated_deltanet_block.py" }, "gated_linear_attention": { "id": "gated_linear_attention", "name": "Gated Linear Attention (GatedLinearAttention).", "category": "D", "subcategory": "recurrent_linear", "paradigm": "ssm", "file": "Codes/framework/attentions/gated_linear_attention.py" }, "goldfinch_ahn": { "id": "goldfinch_ahn", "name": "GoldFinch-AHN: Adaptive Hybrid Network Combining Sliding Window a", "category": "D", "subcategory": "recurrent_linear", "paradigm": "ssm", "file": "Codes/framework/attentions/goldfinch_ahn.py" }, "gqa": { "id": "gqa", "name": "Grouped-Query Attention (GQA): Memory-Efficient Shared Key-Value ", "category": "A", "subcategory": "grouped_query", "paradigm": "sparse", "file": "Codes/framework/attentions/gqa.py" }, "gqa_attention": { "id": "gqa_attention", "name": "Grouped-Query Attention (GQA): Memory-Efficient Shared Key-Value ", "category": "A", "subcategory": "dense_attention", "paradigm": "mla", "file": "Codes/framework/attentions/gqa_attention.py" }, "granite_hybrid": { "id": "granite_hybrid", "name": "Granite-Hybrid Attention: Interleaved Mamba-2 State Space Archite", "category": "I", "subcategory": "hybrid_composite", "paradigm": "moe", "file": "Codes/framework/attentions/granite_hybrid.py" }, "griffin_ssd": { "id": "griffin_ssd", "name": "Griffin SSD: Hybrid Local Sliding-Window Attention and Global Sta", "category": "E", "subcategory": "svd_low_rank", "paradigm": "ssm", "file": "Codes/framework/attentions/griffin_ssd.py" }, "halo_regla": { "id": "halo_regla", "name": "HALO-ReGLA: Hardware-Aware Linear Optics & Rectified Gated Linear", "category": "B", "subcategory": "sparse_approximate", "paradigm": "mla", "file": "Codes/framework/attentions/halo_regla.py" }, "hamming_attention": { "id": "hamming_attention", "name": "Hamming Attention Distillation (HAD): Sign-Binarized Attention wi", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/hamming_attention.py" }, "hgrn2_gated_deltanet": { "id": "hgrn2_gated_deltanet", "name": "HGRN-2 Gated DeltaNet Hybrid Attention Module.", "category": "D", "subcategory": "recurrent_linear", "paradigm": "ssm", "file": "Codes/framework/attentions/hgrn2_gated_deltanet.py" }, "hgrn2_layer": { "id": "hgrn2_layer", "name": "Hierarchically Gated Recurrent Network V2 (HGRN-2) Layer.", "category": "D", "subcategory": "recurrent_linear", "paradigm": "ssm", "file": "Codes/framework/attentions/hgrn2_layer.py" }, "hgrn2_outer": { "id": "hgrn2_outer", "name": "HGRN-2 Outer-Product Recurrent Linear Attention Module.", "category": "D", "subcategory": "recurrent_linear", "paradigm": "ssm", "file": "Codes/framework/attentions/hgrn2_outer.py" }, "hla_attention": { "id": "hla_attention", "name": "Hadamard Linear Attention (HLA).", "category": "D", "subcategory": "recurrent_linear", "paradigm": "ssm", "file": "Codes/framework/attentions/hla_attention.py" }, "hla_inject": { "id": "hla_inject", "name": "Deprecated stub. Re-exports base module for registry compatibilit", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/hla_inject.py" }, "hymba_transx": { "id": "hymba_transx", "name": "HymbaTransX: Hybrid Differential Transformer & State Space Model ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "mla", "file": "Codes/framework/attentions/hymba_transx.py" }, "infini_attention": { "id": "infini_attention", "name": "Infini-Attention with Compressive Memory and Sliding Window (Prod", "category": "E", "subcategory": "svd_low_rank", "paradigm": "mla", "file": "Codes/framework/attentions/infini_attention.py" }, "jamba": { "id": "jamba", "name": "Jamba-2 Hybrid Transformer: SVD-Mamba + MHA + MoE Attention Block", "category": "I", "subcategory": "hybrid_composite", "paradigm": "svd", "file": "Codes/framework/attentions/jamba.py" }, "jamba_info": { "id": "jamba_info", "name": "Information-Bottlenecked Jamba Architecture with Dual Operational", "category": "I", "subcategory": "hybrid_composite", "paradigm": "ssm", "file": "Codes/framework/attentions/jamba_info.py" }, "kda_attention": { "id": "kda_attention", "name": "Kernelized Decay Attention (KDA) / Recurrent Linear Attention (Ha", "category": "D", "subcategory": "recurrent_linear", "paradigm": "ssm", "file": "Codes/framework/attentions/kda_attention.py" }, "kda_inject": { "id": "kda_inject", "name": "Deprecated stub. Re-exports base module for registry compatibilit", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/kda_inject.py" }, "leka_v1": { "id": "leka_v1", "name": "LEKAv1: Linear-time Efficient Kernel Attention (Production-Harden", "category": "E", "subcategory": "svd_low_rank", "paradigm": "mla", "file": "Codes/framework/attentions/leka_v1.py" }, "leka_v1_inject": { "id": "leka_v1_inject", "name": "Deprecated stub. Re-exports base module for registry compatibilit", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/leka_v1_inject.py" }, "leka_v2": { "id": "leka_v2", "name": "LEKAv2: Linear-time Efficient Kernel Attention with Landmark Dist", "category": "E", "subcategory": "svd_low_rank", "paradigm": "mla", "file": "Codes/framework/attentions/leka_v2.py" }, "leka_v2_inject": { "id": "leka_v2_inject", "name": "LEKA v2: Landmark-Enhanced Kernel Attention with Dual-Scale Disti", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/leka_v2_inject.py" }, "leka_v3": { "id": "leka_v3", "name": "LEKAv3: Linear-time Efficient Kernel Attention with Value-Scored ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "mla", "file": "Codes/framework/attentions/leka_v3.py" }, "leka_v3_inject": { "id": "leka_v3_inject", "name": "LEKA v3: Value-Aware Landmark Kernel Attention with Padding Invar", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/leka_v3_inject.py" }, "linformer_attention": { "id": "linformer_attention", "name": "Linformer Attention: Linear-Complexity Sequence Projection via Lo", "category": "B", "subcategory": "linformer", "paradigm": "mla", "file": "Codes/framework/attentions/linformer_attention.py" }, "lola_attention": { "id": "lola_attention", "name": "Lola Attention: Low-Rank Localized Attention via Three-Memory Sys", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/lola_attention.py" }, "lora_attention": { "id": "lora_attention", "name": "LoRA Attention: Parameter-Efficient Low-Rank Adaptation", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/lora_attention.py" }, "lsh_attention": { "id": "lsh_attention", "name": "Locality-Sensitive Hashing Attention: Angular Random Hyperplane B", "category": "E", "subcategory": "svd_low_rank", "paradigm": "pruning", "file": "Codes/framework/attentions/lsh_attention.py" }, "mamba": { "id": "mamba", "name": "Mamba: Gated State Space Model (Mamba) linear attention replaceme", "category": "D", "subcategory": "recurrent_linear", "paradigm": "ssm", "file": "Codes/framework/attentions/mamba.py" }, "mamba_inject": { "id": "mamba_inject", "name": "Mamba Injection Module.", "category": "D", "subcategory": "recurrent_linear", "paradigm": "ssm", "file": "Codes/framework/attentions/mamba_inject.py" }, "mamba_svd_flash": { "id": "mamba_svd_flash", "name": "MambaSVD-Flash: Hybrid Bidirectional SSM-FlashAttention Module (P", "category": "D", "subcategory": "recurrent_linear", "paradigm": "svd", "file": "Codes/framework/attentions/mamba_svd_flash.py" }, "minimal_aurora": { "id": "minimal_aurora", "name": "MinimalAURORA Attention: Energy-Preserving SVD-Factorized Attenti", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/minimal_aurora.py" }, "minimal_nova": { "id": "minimal_nova", "name": "MinimalNOVA Attention: SVD-Factorized Low-Rank Projection Attenti", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/minimal_nova.py" }, "minimal_prism": { "id": "minimal_prism", "name": "MinimalPRISM Attention: Singular Spectrum Energy-Factorized Low-R", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/minimal_prism.py" }, "minimal_svd": { "id": "minimal_svd", "name": "MinimalSVD: Cumulative Energy-Budgeted SVD Factorization Attentio", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/minimal_svd.py" }, "minimal_svd_v2": { "id": "minimal_svd_v2", "name": "MinimalSVD v2: Refined SVD Low-Rank Attention with Flexible Atten", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/minimal_svd_v2.py" }, "mixattention_moska": { "id": "mixattention_moska", "name": "MixAttention-MoSKA: Frequency/Variance-Aware Sparse Attention (FA", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/mixattention_moska.py" }, "mla_attention": { "id": "mla_attention", "name": "Multi-Head Latent Attention (MLA): Low-Rank Latent Bottleneck Com", "category": "E", "subcategory": "svd_low_rank", "paradigm": "mla", "file": "Codes/framework/attentions/mla_attention.py" }, "moba": { "id": "moba", "name": "MoBA: Mixture of Block Attention for Long-Context Modeling.", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/moba.py" }, "moba_attention": { "id": "moba_attention", "name": "MoBA Attention: Mixture-of-Block Attention via Centroid Cosine Ro", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/moba_attention.py" }, "moe_deltakv_paged": { "id": "moe_deltakv_paged", "name": "MoE-DeltaKV Paged: Quantized & Paged KV-Cache Mixture-of-Experts ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/moe_deltakv_paged.py" }, "nasc_attention": { "id": "nasc_attention", "name": "NASC: NEXUS-AEGIS Spectral Cascade Attention.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/nasc_attention.py" }, "nexus": { "id": "nexus", "name": "NEXUS: Neural Efficient X-scoring Unified Sparse Attention.", "category": "B", "subcategory": "sparse_approximate", "paradigm": "moe", "file": "Codes/framework/attentions/nexus.py" }, "nha_attention": { "id": "nha_attention", "name": "Native Hybrid Attention (NHA) – Production-Hardened.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "mla", "file": "Codes/framework/attentions/nha_attention.py" }, "nha_compressor": { "id": "nha_compressor", "name": "Native Hybrid Attention (NHA) Compressor.", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/nha_compressor.py" }, "nsa_attention": { "id": "nsa_attention", "name": "Native Sparse Attention: Hybrid Local Sliding-Window and Dynamic ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "moe", "file": "Codes/framework/attentions/nsa_attention.py" }, "nystrom_attention": { "id": "nystrom_attention", "name": "Nyström Attention: Low-Rank Landmark Approximation via Matrix Pse", "category": "B", "subcategory": "nystrom", "paradigm": "svd", "file": "Codes/framework/attentions/nystrom_attention.py" }, "optimized_hybrid_attention": { "id": "optimized_hybrid_attention", "name": "Optimized Hybrid Attention (OHA): Fused Sliding-Window Local Self", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/optimized_hybrid_attention.py" }, "orion": { "id": "orion", "name": "ORION: Optimized Rectified Input-Aware Orthogonal Network with En", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/orion.py" }, "performer_attention": { "id": "performer_attention", "name": "Performer Attention – Fast Attention via Positive Orthogonal Rand", "category": "B", "subcategory": "performer", "paradigm": "mla", "file": "Codes/framework/attentions/performer_attention.py" }, "phantom": { "id": "phantom", "name": "PHANTOM: Parameter-Hashed AttentioN with Token-wise Orthogonal Ma", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/phantom.py" }, "primal_attention": { "id": "primal_attention", "name": "Primal Attention: Self-Attention through Asymmetric Kernel SVD in", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/primal_attention.py" }, "progressive_token_pruning": { "id": "progressive_token_pruning", "name": "Progressive Token Pruning (PTP) – Production-Hardened.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "mla", "file": "Codes/framework/attentions/progressive_token_pruning.py" }, "projection_factorized_attn": { "id": "projection_factorized_attn", "name": "Projection Factorized SVD Linear Attention.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/projection_factorized_attn.py" }, "qcqa_attention": { "id": "qcqa_attention", "name": "Quality and Capacity-Aware Grouped Query Attention (QCQA): Dynami", "category": "A", "subcategory": "dense_attention", "paradigm": "moe", "file": "Codes/framework/attentions/qcqa_attention.py" }, "rala_attention": { "id": "rala_attention", "name": "Rank-Augmented Linear Attention (RALA): Low-Rank Parametric Head ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/rala_attention.py" }, "rala_compressor": { "id": "rala_compressor", "name": "RALA Compressor: Rank-Augmented Linear Attention Layer Architectu", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/rala_compressor.py" }, "reap_moe": { "id": "reap_moe", "name": "REAP-MoE: Robust Expert Allocation and Pruning Mixture of Experts", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/reap_moe.py" }, "regla_attention": { "id": "regla_attention", "name": "Refined Gated Linear Attention (ReGLA): Element-Wise Key Gating &", "category": "E", "subcategory": "svd_low_rank", "paradigm": "mla", "file": "Codes/framework/attentions/regla_attention.py" }, "replace": { "id": "replace", "name": "Model Attention Replacement and Dynamic Graph Surgery Utilities.", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/replace.py" }, "res_svd": { "id": "res_svd", "name": "Spectral Residual Low-Rank SVD Attention (Res-SVD Attention) – Pr", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/res_svd.py" }, "routing_attention": { "id": "routing_attention", "name": "Routing Attention: Content-Adaptive Query-Centric Clustering via ", "category": "B", "subcategory": "sparse_approximate", "paradigm": "moe", "file": "Codes/framework/attentions/routing_attention.py" }, "rpc": { "id": "rpc", "name": "Random Projection Compressed (RPC) Attention.", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/rpc.py" }, "rpc_attention": { "id": "rpc_attention", "name": "RPC Attention: Randomized Phase Cluster Attention via Fixed Rando", "category": "A", "subcategory": "dense_attention", "paradigm": "mla", "file": "Codes/framework/attentions/rpc_attention.py" }, "salt": { "id": "salt", "name": "SALT Attention: Singular Value Adaptation with Low-Rank Tuning.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/salt.py" }, "salt_attention": { "id": "salt_attention", "name": "SALT Attention: Singular Value Adaptation with Low-Rank Tuning fo", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/salt_attention.py" }, "samba_m3": { "id": "samba_m3", "name": "SAMBA-M3: Hybrid Sliding-Window Attention and Spectral Mamba Bloc", "category": "D", "subcategory": "recurrent_linear", "paradigm": "ssm", "file": "Codes/framework/attentions/samba_m3.py" }, "samba_moe": { "id": "samba_moe", "name": "Samba-MoE: Selective State Space (SAMBA-M3) + REAP-MoE Hybrid Arc", "category": "D", "subcategory": "recurrent_linear", "paradigm": "moe", "file": "Codes/framework/attentions/samba_moe.py" }, "simple_ssm": { "id": "simple_ssm", "name": "Simple State Space Model (SimpleSSM).", "category": "A", "subcategory": "dense_attention", "paradigm": "ssm", "file": "Codes/framework/attentions/simple_ssm.py" }, "smart_weighted_ensemble": { "id": "smart_weighted_ensemble", "name": "Smart Weighted Ensemble: Task-Aware Attention & Dynamic Confidenc", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/smart_weighted_ensemble.py" }, "soa_attn": { "id": "soa_attn", "name": "Shared Orthogonal Attention (SOA) with Per-Projection Latent Bott", "category": "E", "subcategory": "svd_low_rank", "paradigm": "mla", "file": "Codes/framework/attentions/soa_attn.py" }, "soa_fused_module": { "id": "soa_fused_module", "name": "SOA-Fused Attention: Low-Rank Latent Space Multi-Head Attention w", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/soa_fused_module.py" }, "soa_module": { "id": "soa_module", "name": "SOA Attention: Shared Orthogonal Attention with Latent Bottleneck", "category": "E", "subcategory": "svd_low_rank", "paradigm": "sparse", "file": "Codes/framework/attentions/soa_module.py" }, "span_expanded_attention": { "id": "span_expanded_attention", "name": "Span-Expanded Attention: Receptive Field Extension with Parallel ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/span_expanded_attention.py" }, "spectra_v2": { "id": "spectra_v2", "name": "SPECTRA-v2: Sparse Per-Head Efficient Context Transformer with Re", "category": "B", "subcategory": "sparse_approximate", "paradigm": "moe", "file": "Codes/framework/attentions/spectra_v2.py" }, "spectra_v2_attention": { "id": "spectra_v2_attention", "name": "Re-export module for SpectraV2Attention.", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/spectra_v2_attention.py" }, "spectral_mamba": { "id": "spectral_mamba", "name": "Spectral-Mamba Attention Module: Hybrid Attention-State Space Mod", "category": "D", "subcategory": "recurrent_linear", "paradigm": "ssm", "file": "Codes/framework/attentions/spectral_mamba.py" }, "spectral_mamba_attention": { "id": "spectral_mamba_attention", "name": "Re-export module for SpectralMambaAttention.", "category": "D", "subcategory": "recurrent_linear", "paradigm": "ssm", "file": "Codes/framework/attentions/spectral_mamba_attention.py" }, "spectron": { "id": "spectron", "name": "SpectronAttention: Importance-Aware Low-Rank + Sparse Attention.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/spectron.py" }, "spectron_attention": { "id": "spectron_attention", "name": "Spectron Attention Re-export Module.", "category": "H", "subcategory": "quantization", "paradigm": "sparse", "file": "Codes/framework/attentions/spectron_attention.py" }, "squeezing_heads": { "id": "squeezing_heads", "name": "Squeezing Heads Attention Re-export Module.", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/squeezing_heads.py" }, "squeezing_heads_distill": { "id": "squeezing_heads_distill", "name": "Squeezing Heads Distillation: Multi-Head Receptive Field Distilla", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/squeezing_heads_distill.py" }, "squeezing_heads_distiller": { "id": "squeezing_heads_distiller", "name": "Squeezing Heads Distiller: Progressive Multi-Head Knowledge Disti", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/squeezing_heads_distiller.py" }, "svd_attention": { "id": "svd_attention", "name": "SVD Attention: Sequence-Dimension Singular Subspace Projection Ar", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/svd_attention.py" }, "svd_compressed": { "id": "svd_compressed", "name": "SVD-Compressed Attention: Importance-Aware Lagrangian Rank Alloca", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/svd_compressed.py" }, "svd_compressed_attention": { "id": "svd_compressed_attention", "name": "SVD-Compressed Attention Re-export Module.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/svd_compressed_attention.py" }, "svd_linear_attention": { "id": "svd_linear_attention", "name": "SVD Factorized Linear Attention & Kernelized Linear Attention Mod", "category": "D", "subcategory": "recurrent_linear", "paradigm": "svd", "file": "Codes/framework/attentions/svd_linear_attention.py" }, "svd_linear_decay": { "id": "svd_linear_decay", "name": "SVD Linear Decay Gated Attention Alias.", "category": "D", "subcategory": "recurrent_linear", "paradigm": "svd", "file": "Codes/framework/attentions/svd_linear_decay.py" }, "svd_llm_v2": { "id": "svd_llm_v2", "name": "SVD-LLM V2 Attention Alias.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/svd_llm_v2.py" }, "svd_llm_v2_attention": { "id": "svd_llm_v2_attention", "name": "SVD-LLM V2 Attention: Truncation-Aware Explicit Factor Buffer Arc", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/svd_llm_v2_attention.py" }, "svd_mamba": { "id": "svd_mamba", "name": "SVD-Mamba Attention: Hardware-Efficient State-Space Recurrence vi", "category": "D", "subcategory": "recurrent_linear", "paradigm": "svd", "file": "Codes/framework/attentions/svd_mamba.py" }, "svd_proj_attn": { "id": "svd_proj_attn", "name": "SVD Projection Attention Alias.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/svd_proj_attn.py" }, "svd_projection_attention": { "id": "svd_projection_attention", "name": "SVD Projection Attention: Explicit Low-Rank Matrix Reconstruction", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/svd_projection_attention.py" }, "svd_sequence_attention": { "id": "svd_sequence_attention", "name": "Low-Rank SVD Sequence Attention Re-export Module.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/svd_sequence_attention.py" }, "template_attention": { "id": "template_attention", "name": "Template Attention: Reference Implementation & Standardization Ba", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/template_attention.py" }, "token_compressed": { "id": "token_compressed", "name": "Token-Compressed Attention Wrapper: Structural Abstraction & Pass", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/token_compressed.py" }, "token_compressed_attention": { "id": "token_compressed_attention", "name": "Token Compressed Attention: Saliency-Based Key-Value Pruning", "category": "G", "subcategory": "pruning_sparsity", "paradigm": "sparse", "file": "Codes/framework/attentions/token_compressed_attention.py" }, "tpa_attention": { "id": "tpa_attention", "name": "TPAttention: Tensor Product Attention with Factorised Query-Key P", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/attentions/tpa_attention.py" }, "tt": { "id": "tt", "name": "TokenCompressedAttention: A No‑Compression Baseline Wrapper for A", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/tt.py" }, "unified_rope": { "id": "unified_rope", "name": "Universal Rotary Position Embedding (UnifiedRoPE): Positional Inv", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/unified_rope.py" }, "upi_mamba": { "id": "upi_mamba", "name": "UPI-Stuffed Mamba: Uncertainty-Penalized Interpolated Mamba SSM", "category": "D", "subcategory": "recurrent_linear", "paradigm": "ssm", "file": "Codes/framework/attentions/upi_mamba.py" }, "upi_stuffed_mamba": { "id": "upi_stuffed_mamba", "name": "UPI-Stuffed Mamba: Uncertainty-Penalized Interpolated Mamba SSM B", "category": "D", "subcategory": "recurrent_linear", "paradigm": "ssm", "file": "Codes/framework/attentions/upi_stuffed_mamba.py" }, "vig_attention": { "id": "vig_attention", "name": "ViG Attention: Vision GNN-inspired 2D Grid Linear Attention with ", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/vig_attention.py" }, "vspec": { "id": "vspec", "name": "VSpec++ Attention: Variational Spectral Token-Adaptive Attention ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "sparse", "file": "Codes/framework/attentions/vspec.py" }, "weighted_ensemble": { "id": "weighted_ensemble", "name": "Smart Weighted Ensemble with Task-Aware Attention and Hybrid Comb", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/attentions/weighted_ensemble.py" }, "wgqa_attention": { "id": "wgqa_attention", "name": "Weighted Grouped-Query Attention (WGQA).", "category": "A", "subcategory": "dense_attention", "paradigm": "mla", "file": "Codes/framework/attentions/wgqa_attention.py" }, "xattention": { "id": "xattention", "name": "XAttention: Block-Sparse Coarse-to-Fine Routing with Unified Soft", "category": "E", "subcategory": "svd_low_rank", "paradigm": "moe", "file": "Codes/framework/attentions/xattention.py" }, "zeros_attention": { "id": "zeros_attention", "name": "ZeroS Attention: Zero-Mean Normalized Linear Attention", "category": "E", "subcategory": "svd_low_rank", "paradigm": "sparse", "file": "Codes/framework/attentions/zeros_attention.py" }, "a3_compressor": { "id": "a3_compressor", "name": "A3 Compressor: Activation-Aware Factorized SVD Compression with F", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/a3_compressor.py" }, "aa_svd_compressor": { "id": "aa_svd_compressor", "name": "Activation-Aware Singular Value Decomposition (AASVD) Compressor.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/aa_svd_compressor.py" }, "aasvd": { "id": "aasvd", "name": "Activation-Aware Singular Value Decomposition (AASVD): Streaming ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/aasvd.py" }, "adaptive_compressor": { "id": "adaptive_compressor", "name": "AdaptiveTransformerCompressor: Dynamic Energy-Threshold Attention", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/adaptive_compressor.py" }, "adaptive_ranks": { "id": "adaptive_ranks", "name": "Adaptive Ranks SVD Compressor: dynamic rank selection per layer b", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/adaptive_ranks.py" }, "adasvd_compressor": { "id": "adasvd_compressor", "name": "AdaSVDCompressor: Adaptive SVD via Nuclear-Norm Layer Importance ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/adasvd_compressor.py" }, "admm_compressor": { "id": "admm_compressor", "name": "ADMM Compressor: Alternating Direction Method of Multipliers for ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "pruning", "file": "Codes/framework/compression_engine/admm_compressor.py" }, "admm_pruner": { "id": "admm_pruner", "name": "ADMM Pruner: Alternating Direction Method of Multipliers with Dyn", "category": "G", "subcategory": "pruning_sparsity", "paradigm": "pruning", "file": "Codes/framework/compression_engine/admm_pruner.py" }, "armor_pruner": { "id": "armor_pruner", "name": "ARMOR Pruner: Adaptive Resilient Matrix Orthogonal Rank Pruning (", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/armor_pruner.py" }, "ast_pruner": { "id": "ast_pruner", "name": "AST: Adaptive Semi-Structured 2:4 and N:M Pruning with Empirical ", "category": "G", "subcategory": "pruning_sparsity", "paradigm": "pruning", "file": "Codes/framework/compression_engine/ast_pruner.py" }, "astro_quantizer": { "id": "astro_quantizer", "name": "AstroQuantizer: Dynamic Uniform Range Quantization Engine with St", "category": "H", "subcategory": "quantization", "paradigm": "quant", "file": "Codes/framework/compression_engine/astro_quantizer.py" }, "bonsai_pruner": { "id": "bonsai_pruner", "name": "Bonsai: Iterative Magnitude and Taylor Pruning (IMP) with Gradien", "category": "G", "subcategory": "pruning_sparsity", "paradigm": "moe", "file": "Codes/framework/compression_engine/bonsai_pruner.py" }, "cachenotes_compressor": { "id": "cachenotes_compressor", "name": "CacheNotes Compressor: Task-Aware CPT Guidance for Static SVD & D", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/cachenotes_compressor.py" }, "come": { "id": "come", "name": "CoMe: Compress LLMs through Layer Concatenation and Shedding with", "category": "A", "subcategory": "dense_attention", "paradigm": "pruning", "file": "Codes/framework/compression_engine/come.py" }, "compot_compressor": { "id": "compot_compressor", "name": "COMPOT: Compressed Orthogonal Dictionary Learning Compressor.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/compot_compressor.py" }, "coopq_quantizer": { "id": "coopq_quantizer", "name": "CoopQ: Cooperative Game-Theoretic Bit Allocation Quantizer with S", "category": "H", "subcategory": "quantization", "paradigm": "quant", "file": "Codes/framework/compression_engine/coopq_quantizer.py" }, "cr_compressor": { "id": "cr_compressor", "name": "CR Compressor: Interpolative Decomposition for Weight Compression", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/compression_engine/cr_compressor.py" }, "cur_compressor": { "id": "cur_compressor", "name": "CUR Compressor: replaces Linear and Conv1D layers with CUR decomp", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/compression_engine/cur_compressor.py" }, "cur_linear": { "id": "cur_linear", "name": "CUR Linear Layer: low-rank cross-approximation using column and r", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/cur_linear.py" }, "cur_moe": { "id": "cur_moe", "name": "CUR-MoE Compressor alias module.", "category": "F", "subcategory": "mixture_of_experts", "paradigm": "moe", "file": "Codes/framework/compression_engine/cur_moe.py" }, "cur_moe_compressor": { "id": "cur_moe_compressor", "name": "CUR-MoE Compressor: apply CUR cross-approximation to Mixture of E", "category": "F", "subcategory": "mixture_of_experts", "paradigm": "moe", "file": "Codes/framework/compression_engine/cur_moe_compressor.py" }, "cure": { "id": "cure", "name": "CURE: CUR Expert Compression Engine with Randomized Subspace Appr", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/cure.py" }, "delta_kv": { "id": "delta_kv", "name": "DeltaKV: Residual-Based Key-Value Cache Compression via Long-Rang", "category": "A", "subcategory": "dense_attention", "paradigm": "mla", "file": "Codes/framework/compression_engine/delta_kv.py" }, "direct_svd": { "id": "direct_svd", "name": "Direct SVD Compressor: Truncated SVD low-rank factorized layer re", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/direct_svd.py" }, "dobi": { "id": "dobi", "name": "DobiSVDLinearLayer & Compressor: Differentiable SVD with Learnabl", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/dobi.py" }, "dobi_svd": { "id": "dobi_svd", "name": "DobiSVDLinearLayer & Compressor: Differentiable SVD with Learnabl", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/dobi_svd.py" }, "dynamic_entropy": { "id": "dynamic_entropy", "name": "Dynamic Entropy (`dynamic_entropy`)", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/dynamic_entropy.py" }, "dynamic_rank": { "id": "dynamic_rank", "name": "Dynamic Entropy Singular Value Decomposition (Dynamic Entropy SVD", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/dynamic_rank.py" }, "enlighten_compressor": { "id": "enlighten_compressor", "name": "ENLighten: Low-Rank SVD + 2:4 Structured Sparsity (Photonic-Align", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/enlighten_compressor.py" }, "entropy_budget_allocation": { "id": "entropy_budget_allocation", "name": "EntropyBudgetAllocation: global rank assignment via whitened acti", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/entropy_budget_allocation.py" }, "erc_svd": { "id": "erc_svd", "name": "ERCCompressor: Error-Controlled Singular Value Energy Thresholdin", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/erc_svd.py" }, "faq_quantizer": { "id": "faq_quantizer", "name": "FAQ (Feature-Activation-Aware Quantization): adjusts quantization", "category": "H", "subcategory": "quantization", "paradigm": "quant", "file": "Codes/framework/compression_engine/faq_quantizer.py" }, "flap": { "id": "flap", "name": "FLAP: Fluctuation-Based Adaptive Structured Pruning.", "category": "G", "subcategory": "pruning_sparsity", "paradigm": "pruning", "file": "Codes/framework/compression_engine/flap.py" }, "flat_llm": { "id": "flat_llm", "name": "FLAT-LLM: Activation-Aware Structured Dimension & Neuron Pruning ", "category": "G", "subcategory": "pruning_sparsity", "paradigm": "pruning", "file": "Codes/framework/compression_engine/flat_llm.py" }, "frandi": { "id": "frandi", "name": "FRanDI Compressor alias module.", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/compression_engine/frandi.py" }, "frandi_compressor": { "id": "frandi_compressor", "name": "FRanDI: Feature-Regression Aligned Singular Value Decomposition C", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/compression_engine/frandi_compressor.py" }, "gfws_d_compressor": { "id": "gfws_d_compressor", "name": "GFWSVD: Fisher-weighted / Covariance-weighted SVD. Registered as ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/gfws_d_compressor.py" }, "grecodec": { "id": "grecodec", "name": "GreCoDeC Compressor alias module.", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/compression_engine/grecodec.py" }, "grecodec_compressor": { "id": "grecodec_compressor", "name": "GreCoDeC: Greedy Coordinate Descent Low-Rank Compressor.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/grecodec_compressor.py" }, "hashed_linear": { "id": "hashed_linear", "name": "HashedNets: Vectorized Weight Sharing via Hash Tables with Sign H", "category": "E", "subcategory": "svd_low_rank", "paradigm": "sparse", "file": "Codes/framework/compression_engine/hashed_linear.py" }, "hashed_net_compressor": { "id": "hashed_net_compressor", "name": "HashedNet Compressor: replaces nn.Linear and Conv1D layers with H", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/compression_engine/hashed_net_compressor.py" }, "huffman_coder": { "id": "huffman_coder", "name": "Huffman Coder & Entropy Compression Engine for Weight and KV Cach", "category": "E", "subcategory": "svd_low_rank", "paradigm": "mla", "file": "Codes/framework/compression_engine/huffman_coder.py" }, "hybrid_svd_layer": { "id": "hybrid_svd_layer", "name": "Hybrid SVD Layer: Single‑Head SVD Approximation with Positional E", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/hybrid_svd_layer.py" }, "hybridkv": { "id": "hybridkv", "name": "HybridKV: Hybrid KV Cache Compression for Multimodal & Long-Conte", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/hybridkv.py" }, "hybridkv_compressor": { "id": "hybridkv_compressor", "name": "HybridKV Compressor: Dual Static-Dynamic Head-Specific Hierarchic", "category": "I", "subcategory": "hybrid_composite", "paradigm": "mla", "file": "Codes/framework/compression_engine/hybridkv_compressor.py" }, "kmeans_quantize": { "id": "kmeans_quantize", "name": "KMeans-based weight clustering quantizer with STE. Registered as ", "category": "H", "subcategory": "quantization", "paradigm": "quant", "file": "Codes/framework/compression_engine/kmeans_quantize.py" }, "kronecker_compressor": { "id": "kronecker_compressor", "name": "Kronecker Product Compressor: replaces nn.Linear and Conv1D layer", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/kronecker_compressor.py" }, "kronecker_linear": { "id": "kronecker_linear", "name": "Kronecker Linear Layer: weight matrix represented as Kronecker pr", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/kronecker_linear.py" }, "kronecker_tt_lora": { "id": "kronecker_tt_lora", "name": "Kronecker-TT-LoRA: Hybrid Compression Engine for Parameter-Effici", "category": "E", "subcategory": "svd_low_rank", "paradigm": "sparse", "file": "Codes/framework/compression_engine/kronecker_tt_lora.py" }, "kvtc": { "id": "kvtc", "name": "KVTC: KV Cache Transform Coding - PCA decorrelation + adaptive qu", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/kvtc.py" }, "kvtc_compressor": { "id": "kvtc_compressor", "name": "KVTC Compressor: Key-Value Transform Coding via PCA Decorrelation", "category": "H", "subcategory": "quantization", "paradigm": "mla", "file": "Codes/framework/compression_engine/kvtc_compressor.py" }, "liftr": { "id": "liftr", "name": "LIFTR: Low-Rank In-Flight Truncated Reconstruction Compressor", "category": "A", "subcategory": "dense_attention", "paradigm": "pruning", "file": "Codes/framework/compression_engine/liftr.py" }, "lillama_compressor": { "id": "lillama_compressor", "name": "LiLLaMA: Local Distillation via Activation Cross-Covariance Singu", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/lillama_compressor.py" }, "lsc_compressor": { "id": "lsc_compressor", "name": "LSC Compressor: Low-Rank Spectral Fourier Compression Engine", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/lsc_compressor.py" }, "mamba_shedder": { "id": "mamba_shedder", "name": "Mamba-Shedder: Sensitivity-Based Structured Pruning and Layer She", "category": "D", "subcategory": "recurrent_linear", "paradigm": "ssm", "file": "Codes/framework/compression_engine/mamba_shedder.py" }, "mgie_svd": { "id": "mgie_svd", "name": "MGIE-SVD: Maximum Information Entropy SVD Rank Selection.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/mgie_svd.py" }, "mgie_svd_compressor": { "id": "mgie_svd_compressor", "name": "MGIESVDBudgetCompressor: Information Entropy SVD Rank Allocation ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/mgie_svd_compressor.py" }, "mlorq_compressor": { "id": "mlorq_compressor", "name": "MLoRQ Compressor: Multi-Layer Optimal Rank SVD and Joint Low-Bit ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/mlorq_compressor.py" }, "modegpt": { "id": "modegpt", "name": "MoDeGPT: Modular Decomposition of Transformer Blocks.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/modegpt.py" }, "modegpt_compressor": { "id": "modegpt_compressor", "name": "MoDeGPT: Modular Decomposition of Transformer Blocks.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/modegpt_compressor.py" }, "nli_softmax": { "id": "nli_softmax", "name": "NLI Softmax approximation – a fast, quantisation‑friendly softmax", "category": "H", "subcategory": "quantization", "paradigm": "pruning", "file": "Codes/framework/compression_engine/nli_softmax.py" }, "nmf_compressor": { "id": "nmf_compressor", "name": "NMFCompressor: Non-Negative Matrix Factorization Compression Engi", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/nmf_compressor.py" }, "oats": { "id": "oats", "name": "OATS: Outlier-Aware Training Sparsification. Registered as 'oats'", "category": "B", "subcategory": "sparse_approximate", "paradigm": "sparse", "file": "Codes/framework/compression_engine/oats.py" }, "oats_compressor": { "id": "oats_compressor", "name": "OATS: Outlier-Aware Training & Saliency-Based Sparsification with", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/oats_compressor.py" }, "optimal_subspace_projection": { "id": "optimal_subspace_projection", "name": "Optimal Subspace Projection: Activation-Aware Subspace Projection", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/optimal_subspace_projection.py" }, "padre_attention": { "id": "padre_attention", "name": "PADRe: Polynomial Attention with Dropout Regularization.", "category": "B", "subcategory": "sparse_approximate", "paradigm": "mla", "file": "Codes/framework/compression_engine/padre_attention.py" }, "proxsparse": { "id": "proxsparse", "name": "ProxSparse: Proximal Gradient Regularized Structured & Semi-Struc", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/proxsparse.py" }, "proxsparse_pruner": { "id": "proxsparse_pruner", "name": "ProxSparse: regularized learning of semi-structured sparsity. Reg", "category": "G", "subcategory": "pruning_sparsity", "paradigm": "pruning", "file": "Codes/framework/compression_engine/proxsparse_pruner.py" }, "pruning_utils": { "id": "pruning_utils", "name": "Pruning Utilities and Saliency Estimation for Structural Pruning ", "category": "G", "subcategory": "pruning_sparsity", "paradigm": "pruning", "file": "Codes/framework/compression_engine/pruning_utils.py" }, "replaceme_compressor": { "id": "replaceme_compressor", "name": "ReplaceMe: Layer Removal and Replacement with Linear Residual Blo", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/replaceme_compressor.py" }, "roast_hashed_linear": { "id": "roast_hashed_linear", "name": "ROAST: Hardware-Aware Weight Sharing with Tiling.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/roast_hashed_linear.py" }, "rotational_slicegpt_int4": { "id": "rotational_slicegpt_int4", "name": "Rotational SliceGPT INT4 Compressor: Rotational Invariance, Subsp", "category": "H", "subcategory": "quantization", "paradigm": "mla", "file": "Codes/framework/compression_engine/rotational_slicegpt_int4.py" }, "rpca": { "id": "rpca", "name": "RPCA Compressor alias module.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/rpca.py" }, "rpca_compressor": { "id": "rpca_compressor", "name": "RPCACompressor: Robust Principal Component Analysis Low-Rank & Sp", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/rpca_compressor.py" }, "saes_svd": { "id": "saes_svd", "name": "SAES-SVD v2: error suppression with fixed rank ratio. Registered ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/saes_svd.py" }, "saes_svd_compressor": { "id": "saes_svd_compressor", "name": "SAES-SVD Compressor: Joint intra- and inter-layer error compensat", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/saes_svd_compressor.py" }, "saes_svd_v2": { "id": "saes_svd_v2", "name": "SAES-SVD v2: Spectrum-Aware Error-Suppressed Singular Value Decom", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/saes_svd_v2.py" }, "salr": { "id": "salr", "name": "SALR: Structure-Aware / Sparse Low-Rank compression combining SVD", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/salr.py" }, "sdrm": { "id": "sdrm", "name": "SDRM: Scaled Dimension Rank Modulation Compressor. Registered as ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/sdrm.py" }, "sdrm_compressor": { "id": "sdrm_compressor", "name": "SDRM: Rank Assignment Proportional to Parameter Budget for Edge I", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/sdrm_compressor.py" }, "slicegpt": { "id": "slicegpt", "name": "SliceGPT: Post-Training Slicing with Orthogonal Transformation & ", "category": "A", "subcategory": "dense_attention", "paradigm": "pruning", "file": "Codes/framework/compression_engine/slicegpt.py" }, "slicegpt_compressor": { "id": "slicegpt_compressor", "name": "SliceGPT PCA-based slicing compressor. Registered as 'slicegpt_pc", "category": "A", "subcategory": "dense_attention", "paradigm": "pruning", "file": "Codes/framework/compression_engine/slicegpt_compressor.py" }, "slicegpt_pca": { "id": "slicegpt_pca", "name": "SliceGPT PCA Compressor: Principal Component Analysis based Dimen", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/slicegpt_pca.py" }, "slim_compressor": { "id": "slim_compressor", "name": "SLiM: Structured Low-Rank Integrated Matrix Compression (Quantiza", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/slim_compressor.py" }, "soka": { "id": "soka", "name": "SoKA: Second-Order Knowledge-Aware Low-Rank Linear Compressor. Re", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/soka.py" }, "soka_compressor": { "id": "soka_compressor", "name": "SoKA: Second-Order Knowledge-Aware Low-Rank Linear Compressor.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/soka_compressor.py" }, "sparsegpt": { "id": "sparsegpt", "name": "SparseGPT: One-Shot Second-Order Hessian-Guided Weight Pruning fo", "category": "G", "subcategory": "pruning_sparsity", "paradigm": "svd", "file": "Codes/framework/compression_engine/sparsegpt.py" }, "spectraformer_attention": { "id": "spectraformer_attention", "name": "Spectraformer: Random Feature Linear Attention with Flexible Kern", "category": "B", "subcategory": "sparse_approximate", "paradigm": "mla", "file": "Codes/framework/compression_engine/spectraformer_attention.py" }, "subspace_projection": { "id": "subspace_projection", "name": "Subspace Projection Compressor: Low-Rank Subspace Projection and ", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/subspace_projection.py" }, "svd_linear": { "id": "svd_linear", "name": "SVD-based low-rank linear layer replacement with optional Fused S", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/svd_linear.py" }, "swift_svd": { "id": "swift_svd", "name": "Swift-SVD: fast closed-form truncated SVD with symmetric square-r", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/swift_svd.py" }, "swift_svd_act": { "id": "swift_svd_act", "name": "Swift-SVD Activation-Aware Compressor via cross-covariance / acti", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/swift_svd_act.py" }, "swift_svd_compressor": { "id": "swift_svd_compressor", "name": "Swift-SVD Activation-aware via cross-covariance. Registered as 's", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/swift_svd_compressor.py" }, "tlinformer_attention": { "id": "tlinformer_attention", "name": "TLinFormer: Exact linear‑time attention via low‑rank sequence pro", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/tlinformer_attention.py" }, "tomoe": { "id": "tomoe", "name": "ToMoE: Converting Dense MLP into Mixture of Experts via Dynamic S", "category": "F", "subcategory": "mixture_of_experts", "paradigm": "moe", "file": "Codes/framework/compression_engine/tomoe.py" }, "tomoe_compressor": { "id": "tomoe_compressor", "name": "ToMoE: converts dense linear and Conv1D layers (e.g. FFN and atte", "category": "F", "subcategory": "mixture_of_experts", "paradigm": "moe", "file": "Codes/framework/compression_engine/tomoe_compressor.py" }, "tt_compressor": { "id": "tt_compressor", "name": "Tensor Train Compressor: replaces eligible nn.Linear and Conv1D l", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/tt_compressor.py" }, "tt_linear": { "id": "tt_linear", "name": "Tensor-Train Linear Layer: parameterizes weight matrices as Tenso", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/tt_linear.py" }, "uniql_compressor": { "id": "uniql_compressor", "name": "UniQL: Joint low‑rank + quantisation using SVD and uniform quant.", "category": "E", "subcategory": "svd_low_rank", "paradigm": "svd", "file": "Codes/framework/compression_engine/uniql_compressor.py" }, "zerokv_neo": { "id": "zerokv_neo", "name": "ZeroKV-Neo: L2-norm based KV cache eviction with attention sink a", "category": "A", "subcategory": "dense_attention", "paradigm": "mla", "file": "Codes/framework/compression_engine/zerokv_neo.py" }, "zerokv_neo_compressor": { "id": "zerokv_neo_compressor", "name": "ZeroKV-Neo: L2-norm based KV cache eviction with attention sink a", "category": "A", "subcategory": "dense_attention", "paradigm": "mla", "file": "Codes/framework/compression_engine/zerokv_neo_compressor.py" } }
{
"SOAFusedAttention": {
"id": "SOAFusedAttention",
"name": "SOA Fused Attention: Low-Rank Latent Space Multi-Head Attention w",
"category": "E",
"subcategory": "svd_low_rank",
"paradigm": "svd",
"file": "Codes/framework/attentions/SOAFusedAttention.py"
},
"adaptive_span_attention": {
"id": "adaptive_span_attention",
"name": "Adaptive Span Attention: Dynamic Head-Specific Context Window Gat",
"category": "C",
"subcategory": "sliding_window",
"paradigm": "sparse",
"file": "Codes/framework/attentions/adaptive_span_attention.py"
},
"adaptive_span_inject": {
"id": "adaptive_span_inject",
"name": "Adaptive Span Attention: Dynamic Head-Specific Context Window Gat",
"category": "C",
"subcategory": "sliding_window",
"paradigm": "sparse",
"file": "Codes/framework/attentions/adaptive_span_inject.py"
},
"adaptive_sparse_attention": {
"id": "adaptive_sparse_attention",
"name": "Adaptive Sparse Attention: Dynamic Top-K & Threshold-Based Sparsi",
"category": "E",
"subcategory": "svd_low_rank",
"paradigm": "sparse",
"file": "Codes/framework/attentions/adaptive_sparse_attention.py"
},
"adaptive_sparse_inject": {
"id": "adaptive_sparse_inject",
"name": "Adaptive Sparse Attention: Dynamic Top-K & Threshold-Based Sparsi",
"category": "B",
"subcategory": "sparse_approximate",
"paradigm": "sparse",
"file": "Codes/framework/attentions/adaptive_sparse_inject.py"
},
"adaptive_svd_attention": {
"id": "adaptive_svd_attention",
"name": "Adaptive SVD Attention: Low-Rank SVD Projections with Parallel Re",
"category": "E",
"subcategory": "svd_low_rank",
"paradigm": "svd",
"file": "Codes/framework/attentions/adaptive_svd_attention.py"
},
"adasplash_attention": {
"id": "adasplash_attention",
"name": "AdaSplash Attention: Adaptive Splash Attention with alpha-Entmax ",
"category": "C",
"subcategory": "sliding_window",
"paradigm": "moe",
"file": "Codes/framework/attentions/adasplash_attention.py"
},
"adasplash_inject": {
"id": "adasplash_inject",
"name": "AdaSplash Attention: Adaptive Splash Attention with alpha-Entmax ",
"category": "B",
"subcategory": "sparse_approximate",
"paradigm": "sparse",
"file": "Codes/framework/attentions/adasplash_inject.py"
},
"aegis_hybrid": {
"id": "aegis_hybrid",
"name": "AEGIS-Hybrid Attention: Low-Rank Factorized Hybrid Attention with",
"category": "E",
"subcategory": "svd_low_rank",
"paradigm": "svd",
"file": "Codes/framework/attentions/aegis_hybrid.py"
},
"aegis_hybrid_attention": {
"id": "aegis_hybrid_attention",
"name": "Compatibility shim for aegis_hybrid_attention.",
"category": "B",
"subcategory": "sparse_approximate",
"paradigm": "sparse",
"file": "Codes/framework/attentions/aegis_hybrid_attention.py"
},
"aegis_spectron_fp8": {
"id": "aegis_spectron_fp8",
"name": "AEGIS-Spectron-FP8: Tri-Branch Hybrid Attention with Token-Level ",
"category": "H",
"subcategory": "quantization",
"paradigm": "svd",
"file": "Codes/framework/attentions/aegis_spectron_fp8.py"
},
"aegis_v1": {
"id": "aegis_v1",
"name": "AEGIS-Attention V1: Adaptive Enhanced Global-Local Integrated Spa",
"category": "E",
"subcategory": "svd_low_rank",
"paradigm": "svd",
"file": "Codes/framework/attentions/aegis_v1.py"
},
"aegis_v2": {
"id": "aegis_v2",
"name": "AEGIS-Attention V2: Low-Rank Factorized Hybrid Attention with Ada",

Xet Storage Details

Size:
69.5 kB
·
Xet hash:
8e88afd9ee874bd8b7d7770e6428ea31f42994a4f26187324c318299925fb1da

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.