List of all items

Structs

• attn_v41::AttnV41
• attn_v41::AttnV41Cfg
• attn_v41::AttnV41LayerState
• attn_v41::AttnV41LayerWeights
• attn_v41::AttnV41Run
• attn_v41::CompressorWeightsGpu
• attn_v41::IndexerWeightsGpu
• attn_v41::LayerRole
• attn_v41::SharedV41
• deepseek_v41_layer::DeepSeekV41Layer
• deepseek_v41_layer::SegState
• deepseek_v41_layer::V41LayerState
• deepseek_v41_layer::V41Runtime
• deepseek_v41_ref::Golden
• deepseek_v41_ref::GoldenTensor
• deepseek_v41_ref::WeightMeta
• deepseek_v41_ref::attn::AttnCfg
• deepseek_v41_ref::attn::AttnRun
• deepseek_v41_ref::attn::AttnWeights
• deepseek_v41_ref::attn::WindowCache
• deepseek_v41_ref::compress::CompAttnCfg
• deepseek_v41_ref::compress::CompAttnRun
• deepseek_v41_ref::compress::CompressorState
• deepseek_v41_ref::compress::CompressorWeights
• deepseek_v41_ref::compress::IndexerCfg
• deepseek_v41_ref::compress::IndexerWeights
• deepseek_v41_ref::compress::LayerAttnState
• deepseek_v41_ref::compress::SharedRuntime
• deepseek_v41_ref::engram::EngramTables
• deepseek_v41_ref::engram::NgramHashState
• deepseek_v41_ref::model::EngramWeights
• deepseek_v41_ref::model::LayerTrace
• deepseek_v41_ref::model::LayerWeights
• deepseek_v41_ref::model::ModelCfg
• deepseek_v41_ref::model::ModelState
• deepseek_v41_ref::model::ModelWeights
• deepseek_v41_ref::model::StepTrace
• deepseek_v41_ref::moe::MoeCfg
• deepseek_v41_ref::moe::MoeWeights
• deepseek_v4_mtp::DeepseekV4MtpHead
• deepseek_v4_mtp::DeepseekV4MtpProposerState
• dflash_head::BlockDiffusionDraftHead
• dflash_head::DflashKernels
• dflash_head::DflashLayer
• dflash_head::DflashProposerState
• dflash_head::DflashScratch
• dflash_head::ProposeGraphs
• dflash_head::levers::DFlashLevers
• engram_v41::EngramHashTables
• engram_v41::EngramHasher
• engram_v41::EngramLayerWeights
• engram_v41::EngramV41
• glm5next_dsa::Glm5NextDsaConfig
• glm5next_dsa::Glm5NextDsaKernels
• glm5next_dsa::attend::DsaDecodeInputs
• glm5next_dsa::attend::DsaDecodePaging
• glm5next_dsa::attend::Glm5NextDsaDecodeKernel
• glm5next_dsa::binding::DsaBindReport
• glm5next_dsa::binding::DsaSpec
• glm5next_dsa::layer::Glm5NextDsaLayer
• glm5next_dsa::layer::Glm5NextDsaLayerKernels
• glm5next_dsa::layer::Glm5NextDsaWeights
• glm5next_dsa::layer::Glm5NextDsaWorkspace
• glm5next_dsa::select::DsaSelectGeometry
• glm5next_dsa::select::DsaSelectInputs
• glm5next_dsa::select::DsaSelectScratch
• glm5next_dsa::state::Glm5NextDsaState
• glm5next_dsa::tp::DsaTensorPlan
• glm5next_dsa::tp::DsaTpPlan
• glm5next_dsa_ref::DsaDims
• glm5next_dsa_ref::Pools
• glm5next_kda::Glm5NextKdaConfig
• glm5next_kda::Glm5NextKdaKernels
• glm5next_kda::Glm5NextKdaLayer
• glm5next_kda::Glm5NextKdaWeights
• glm5next_kda::Glm5NextKdaWorkspace
• glm5next_kda::KdaSeqState
• glm5next_kda::binding::KdaBindReport
• glm5next_kda::binding::RawTensor
• glm5next_kda::binding::TensorSpec
• glm5next_kda::tp::KdaTensorPlan
• glm5next_kda::tp::KdaTpPlan
• glm5next_kda::tp_bind::KdaShardedSource
• glm5next_kda_ref::KdaDims
• glm5next_kda_ref::KdaWeights
• glm5next_layer::Glm5NextLayer
• glm5next_layer::Glm5NextMhc
• glm5next_mhc::Glm5NextMhcKernels
• glm5next_mhc::Glm5NextMhcSiteWeights
• glm5next_mlp::Glm5NextMlpConfig
• glm5next_mlp::Glm5NextMlpKernels
• glm5next_mlp::forward::Glm5NextMlpWorkspace
• glm5next_mlp::weights::Glm5NextDenseMlpWeights
• glm5next_mlp::weights::Glm5NextExpertPtrTable
• glm5next_mlp::weights::Glm5NextExpertWeights
• glm5next_mlp::weights::Glm5NextMoePtrTables
• glm5next_mlp::weights::Glm5NextMoeWeights
• glm5next_mlp::weights::Nvfp4Proj
• glm5next_mtp_head::Glm5NextMtpHead
• glm5next_mtp_head::Glm5NextMtpProposerState
• glm5next_skeleton::Glm5NextTextSkeleton
• glm5next_skeleton::SkeletonLayer
• glm5next_skeleton::StateBudget
• glm5next_skeleton::StructuralAccounting
• kimi_k3::Ablation
• kimi_k3::AttnResHub
• kimi_k3::HybridCache
• kimi_k3::K3CpuModel
• kimi_k3::K3Graph
• kimi_k3::K3LayerSpec
• kimi_k3::KdaConfig
• kimi_k3::KdaState
• kimi_k3::LatentMoeConfig
• kimi_k3::MlaConfig
• kimi_k3::MlaKv
• kimi_k3::bound::K3BoundLayer
• kimi_k3::bound::K3HostShared
• kimi_k3::bound::WeightMeta
• kimi_k3::device_cache::DeviceHybridCache
• kimi_k3::device_cache::DeviceLayerCache
• kimi_k3::kda::KdaConfig
• kimi_k3::kda::KdaState
• kimi_k3::kda_cuda::K3KdaDecodeKernels
• kimi_k3::kda_cuda::KdaDeviceState
• kimi_k3::latent_moe::LatentMoeConfig
• kimi_k3::mla::MlaConfig
• kimi_k3::mla_cuda::K3MlaDecodeKernels
• kimi_k3::mla_cuda::MlaDeviceKv
• kimi_k3::moe_cuda::K3MoeGemmKernels
• kimi_k3::state::K3CpuFallbackState
• mistral_loader::MistralWeightLoader
• moe_v41::MoeV41
• moe_v41::MoeV41Cfg
• moe_v41::MoeV41LayerWeights
• moe_v41::MoeV41Stage
• moe_v41::MoeV41Timing
• moe_v41::RouterWeights
• nemotron_mamba2::NemotronMamba2Layer
• nemotron_moe::NemotronMoeLayer
• precision_schedule::PrecisionSchedule
• seq_state_reserve::PerSequenceState
• tp_shard::TpAttentionDims
• tp_shard::TpGdnDims
• tp_shard::TpMoeDims
• weight_loader::Gemma4WeightLoader
• weight_loader::LagunaWeightLoader
• weight_loader::LongcatWeightLoader
• weight_loader::MinimaxM2WeightLoader
• weight_loader::NemotronHWeightLoader
• weight_loader::NllbWeightLoader
• weight_loader::Qwen35DenseWeightLoader
• weight_loader::Qwen35WeightLoader
• weight_loader::Qwen3VLWeightLoader
• weight_loader::Qwen3WeightLoader
• weight_loader::Qwen4ExpWeightLoader
• weight_loader::Step3p7WeightLoader
• weight_loader::deepseek_v41::DeepSeekV41WeightLoader
• weight_loader::deepseek_v4::DeepSeekV4WeightLoader
• weight_loader::deepseek_v4::mtp::DeepseekV4MtpModule
• weight_loader::dflash_loader::DflashConfig
• weight_loader::dflash_loader::DflashLayerWeights
• weight_loader::dflash_loader::DflashRopeScaling
• weight_loader::dflash_loader::DflashSubConfig
• weight_loader::dflash_loader::DflashWeights
• weight_loader::fp8_residency::DenseFp8Inputs
• weight_loader::fp8_residency::DenseFp8Plan
• weight_loader::fp8_residency::DerivedResidency
• weight_loader::fp8_residency::RouteEnv
• weight_loader::fp8_residency::TwinsBuilt
• weight_loader::glm5_next::Accounting
• weight_loader::glm5_next_load::Glm5NextWeightLoader
• weight_loader::glm5_next_mtp::Glm5NextMtpModule
• weight_loader::predicted_residency::Fp8RouteInputs
• weight_loader::predicted_residency::PredictedDerived

Enums

• attn_v41::AttnMat
• deepseek_v41_layer::GraphMode
• dflash_head::DflashQuantization
• glm5next_dsa::select::DsaSelectLaunch
• glm5next_dsa::tp::DsaShard
• glm5next_kda::binding::AttnBlockKind
• glm5next_kda::binding::KdaDtype
• glm5next_kda::tp::KdaShard
• glm5next_layer::Glm5NextMixer
• glm5next_layer::Glm5NextMlpSite
• glm5next_mlp::Glm5NextMlpKind
• glm5next_skeleton::FinalStep
• glm5next_skeleton::Mixer
• glm5next_skeleton::Mlp
• glm5next_skeleton::ResidualStep
• glm5next_skeleton::Site
• glm5next_skeleton::StateKind
• kimi_k3::LayerCache
• kimi_k3::MixerKind
• kimi_k3::MlpKind
• precision_schedule::Dtype
• precision_schedule::Role
• tp_shard::TpShardKind
• weight_loader::QuantFormat
• weight_loader::WeightFormat
• weight_loader::glm5_next::TensorRole
• weight_loader::predicted_residency::DerivedBytesEstimate

Traits

• glm5next_kda::binding::KdaTensorSource
• weight_loader::ModelWeightLoader

Functions

• deepseek_v41_layer::graph_mode
• deepseek_v41_layer::step_graph_on
• deepseek_v41_ref::assert_close
• deepseek_v41_ref::attn::act_quant_inplace
• deepseek_v41_ref::attn::apply_rotary
• deepseek_v41_ref::attn::attention
• deepseek_v41_ref::attn::freqs_cis
• deepseek_v41_ref::attn::pow2_ceil
• deepseek_v41_ref::attn::sparse_attn
• deepseek_v41_ref::attn::window_topk_idxs
• deepseek_v41_ref::checksum
• deepseek_v41_ref::compress::attention_any
• deepseek_v41_ref::compress::compressor
• deepseek_v41_ref::compress::fp4_quant_e4m3_inplace
• deepseek_v41_ref::compress::fp4_quant_e8m0_inplace
• deepseek_v41_ref::compress::indexer
• deepseek_v41_ref::compress::linear_f32
• deepseek_v41_ref::compress::select_candidate_blocks
• deepseek_v41_ref::compress::to_e2m1_rne
• deepseek_v41_ref::compress::torch_cpu_topk_set
• deepseek_v41_ref::compress::yarn_freqs_cis
• deepseek_v41_ref::engram::e4m3_to_f32
• deepseek_v41_ref::engram::embed_rows
• deepseek_v41_ref::engram::f32_to_e4m3_rne
• deepseek_v41_ref::engram::f32_to_e8m0_rne
• deepseek_v41_ref::engram::gate_and_add
• deepseek_v41_ref::engram::linear_bf16
• deepseek_v41_ref::engram::regen_bf16_matrix
• deepseek_v41_ref::engram::regen_bf16_qk
• deepseek_v41_ref::engram::regen_table
• deepseek_v41_ref::fixed_int
• deepseek_v41_ref::fixed_raw
• deepseek_v41_ref::fixed_value
• deepseek_v41_ref::fnv1a64
• deepseek_v41_ref::hc::hc_mixes
• deepseek_v41_ref::hc::hc_post
• deepseek_v41_ref::hc::hc_pre
• deepseek_v41_ref::hc::rms_norm
• deepseek_v41_ref::hc::split_sinkhorn
• deepseek_v41_ref::model::forward
• deepseek_v41_ref::moe::expert
• deepseek_v41_ref::moe::gate
• deepseek_v41_ref::moe::linear_bf16
• deepseek_v41_ref::moe::moe
• deepseek_v41_ref::regen_param
• deepseek_v41_ref::splitmix64
• deepseek_v41_ref::to_bf16_rne
• dflash_head::dflash_ctx_cap
• glm5next_dsa::attend::decode_attention
• glm5next_dsa::binding::dsa_tensor_specs
• glm5next_dsa::binding::verify_dsa_block
• glm5next_dsa::build::absorb_o
• glm5next_dsa::build::absorb_q
• glm5next_dsa::build::build_dsa_weights
• glm5next_dsa::build::shard_host
• glm5next_dsa::select::contiguous_pool_count
• glm5next_dsa::select::select_tokens
• glm5next_dsa::select::topk_smem_for_tile
• glm5next_dsa::select::topk_tile
• glm5next_dsa::state::dsa_capacity
• glm5next_dsa::state::indexer_state_bytes
• glm5next_dsa::state::max_dsa_context
• glm5next_dsa_ref::expand_kv
• glm5next_dsa_ref::expand_selection
• glm5next_dsa_ref::index_scores
• glm5next_dsa_ref::kept_pools
• glm5next_dsa_ref::layer_norm
• glm5next_dsa_ref::linear
• glm5next_dsa_ref::mla_masked_attention
• glm5next_dsa_ref::pool_states
• glm5next_dsa_ref::rms_norm
• glm5next_dsa_ref::sigmoid_f32
• glm5next_dsa_ref::topk_pools
• glm5next_dsa_ref::topk_to_mask
• glm5next_dsa_ref::visible
• glm5next_kda::binding::bind_kda_weights
• glm5next_kda::binding::classify_attn_block
• glm5next_kda::tp_bind::shard_bytes
• glm5next_kda_ref::bounded_gate
• glm5next_kda_ref::kda_chunked
• glm5next_kda_ref::kda_chunked_prenorm
• glm5next_kda_ref::kda_recurrent
• glm5next_kda_ref::kda_recurrent_prenorm
• glm5next_kda_ref::kda_reference_layer
• glm5next_kda_ref::l2norm_rows
• glm5next_kda_ref::linear
• glm5next_kda_ref::rms_norm_gated
• glm5next_kda_ref::unbounded_gdn_gate
• glm5next_layer::prefill_rows
• glm5next_layer::profile::end
• glm5next_layer::profile::end_nosync
• glm5next_layer::profile::end_us
• glm5next_layer::profile::full
• glm5next_layer::profile::on
• glm5next_layer::profile::probe_buf
• glm5next_layer::profile::start
• glm5next_layer::profile::start_hot
• glm5next_layer::profile::stash_route
• glm5next_layer::profile::step
• glm5next_layer::profile::trace_bar
• glm5next_layer::profile::trace_on
• glm5next_layer::state::alloc_kda_ssm_state
• glm5next_mhc::glm_hc_expand
• glm5next_mhc::glm_hc_post
• glm5next_mhc::glm_hc_pre
• glm5next_mhc::hc_head_mean
• glm5next_mhc::mhc_mix_max_tokens
• glm5next_mhc::mix_hc
• glm5next_mlp::build::build_dense_mlp
• glm5next_mlp::build::build_moe
• glm5next_mlp::forward::forward_dense
• glm5next_mlp::forward::forward_moe
• glm5next_mlp::forward::mlp_ws_bytes
• glm5next_mlp::forward::mlp_ws_shared
• glm5next_mlp::forward::mlp_ws_total_bytes
• glm5next_mlp::forward::row_batch_max
• kimi_k3::attnres_blend
• kimi_k3::attnres_mix
• kimi_k3::attnres_softmax_mix
• kimi_k3::cuda_kda_enabled
• kimi_k3::cuda_mla_enabled
• kimi_k3::dense_cuda::launch_dense_matrices
• kimi_k3::dense_cuda::launch_dense_mlp
• kimi_k3::gated_mla_attend
• kimi_k3::greedy_decode
• kimi_k3::kda::bounded_gate
• kimi_k3::kda::conv_update
• kimi_k3::kda::cuda_kda_enabled
• kimi_k3::kda::full_rank_output_gate
• kimi_k3::kda::kda_decode_token
• kimi_k3::kda::kda_from
• kimi_k3::kda::kda_recurrent_step
• kimi_k3::kda_cuda::launch_k3_kda_decode_token
• kimi_k3::kda_cuda::launch_k3_kda_decode_token_on_device
• kimi_k3::kda_decode_token
• kimi_k3::kda_from
• kimi_k3::latent_moe::expert_situ
• kimi_k3::latent_moe::latent_moe_forward
• kimi_k3::latent_moe::mix_routed_experts
• kimi_k3::latent_moe::moe_from
• kimi_k3::latent_moe::sigmoid_topk
• kimi_k3::latent_moe_forward
• kimi_k3::mix_routed_experts
• kimi_k3::mla::apply_output_gate
• kimi_k3::mla::cuda_mla_enabled
• kimi_k3::mla::gated_mla_attend
• kimi_k3::mla::maybe_rope
• kimi_k3::mla::mla_decode_token
• kimi_k3::mla::mla_from
• kimi_k3::mla::sdpa
• kimi_k3::mla::sdpa_one
• kimi_k3::mla_cuda::launch_k3_mla_decode_token
• kimi_k3::mla_cuda::launch_k3_mla_decode_token_on_device
• kimi_k3::mla_decode_token
• kimi_k3::mla_from
• kimi_k3::moe_cuda::launch_k3_latent_moe_experts
• kimi_k3::moe_cuda::launch_k3_moe_e8m0_ptrtable
• kimi_k3::moe_from
• kimi_k3::sigmoid_topk
• kimi_k3::situ_glu
• kimi_k3::situ_glu_vec
• kimi_k3::softcap
• kimi_k3::tp::supports_tp
• kimi_k3::tp::tensor_plan
• moe_v41::bf16_bytes
• moe_v41::prefetch_k
• moe_v41::route_from_logits
• moe_v41::shared_side
• nemotron_moe::set_shared_expert_e4m3_from_cli
• nemotron_moe::shared_expert_e4m3
• seq_state_reserve::per_sequence_state_bytes
• tp_shard::load_qk_norms_tp
• tp_shard::load_qkvo_tp
• tp_shard::shard_dense_1d_bf16
• tp_shard::shard_dense_bf16
• tp_shard::shard_dense_weight
• tp_shard::shard_fp8_block_scaled
• tp_shard::shard_gdn_ba_rows
• tp_shard::shard_gdn_conv_rows
• tp_shard::shard_gdn_out_proj_row_parallel
• tp_shard::shard_gdn_qkv_rows
• tp_shard::shard_gdn_qkvz_rows
• tp_shard::shard_gdn_value_vector
• tp_shard::shard_quantized_nvfp4
• weight_loader::deepseek_v41::bf16_ptr
• weight_loader::deepseek_v41::download_f32
• weight_loader::deepseek_v41::f32_ptr
• weight_loader::deepseek_v41::load_layers::load_layers
• weight_loader::deepseek_v41::upload_f32
• weight_loader::deepseek_v4::mtp::load_v4_mtp_module
• weight_loader::dflash_loader::load_dflash_weights
• weight_loader::dflash_loader::parse_dflash_config
• weight_loader::dflash_loader::store_has_dflash_weights
• weight_loader::fp8_residency::attn_fp8_twin_bytes
• weight_loader::fp8_residency::attn_nvfp4_bytes
• weight_loader::fp8_residency::dense_ffn_nvfp4_bytes
• weight_loader::fp8_residency::ffn_gateup_fused_bytes
• weight_loader::fp8_residency::ffn_gateup_fused_parts
• weight_loader::fp8_residency::fp8_twin_bytes
• weight_loader::fp8_residency::keep_nvfp4_fallback
• weight_loader::fp8_residency::nvfp4_bytes
• weight_loader::glm5_next::account
• weight_loader::glm5_next::classify
• weight_loader::glm5_next::is_mtp_only_name
• weight_loader::glm5_next::layer_index
• weight_loader::glm5_next_load::upload_bf16
• weight_loader::glm5_next_mtp::load_glm5next_mtp_module
• weight_loader::glm5_next_vision::load_glm5_next_vision
• weight_loader::predicted_residency::predicted_derived_bytes

Type Aliases

• glm5next_dsa::build::LoadFn
• glm5next_mlp::build::ExpertFn
• glm5next_mlp::build::LoadFn

Constants

• deepseek_v41_ref::GOLDEN_JSON
• deepseek_v41_ref::attn::FP8_BLOCK
• deepseek_v41_ref::compress::FP4_BLOCK
• deepseek_v41_ref::compress::LATENT_BLOCK
• engram_v41::ENGRAM_ROW_BYTES
• glm5next_dsa::DSA_MODULE
• glm5next_dsa::KERNEL_KV_LORA_DIM
• glm5next_dsa::KERNEL_MAX_KPOOL
• glm5next_dsa::LAYERNORM_MODULE
• glm5next_dsa::MASKED_ATTN_MAX_KEYS
• glm5next_dsa::attend::DSA_DECODE_MODULE
• glm5next_dsa::select::TOPK_SMEM_CEILING
• glm5next_dsa_ref::INVALID
• glm5next_kda::SMEM_CEILING
• glm5next_kda::binding::DSA_MARKERS
• glm5next_kda::binding::KDA_TENSORS
• glm5next_layer::profile::DSA_ATTEND
• glm5next_layer::profile::DSA_INDEXER
• glm5next_layer::profile::DSA_PROJ
• glm5next_layer::profile::DSA_SELECT
• glm5next_layer::profile::KDA
• glm5next_layer::profile::KDA_BACK
• glm5next_layer::profile::KDA_FRONT
• glm5next_layer::profile::KDA_RECUR
• glm5next_layer::profile::MHC
• glm5next_layer::profile::MHC_POST
• glm5next_layer::profile::MLP_DENSE
• glm5next_layer::profile::MOE_COMBINE
• glm5next_layer::profile::MOE_EXPERTS
• glm5next_layer::profile::MOE_HOSTSYNC
• glm5next_layer::profile::MOE_ROUTER
• glm5next_layer::profile::MOE_SHARED
• glm5next_layer::profile::NORM
• glm5next_layer::profile::REDUCE_ATTN
• glm5next_layer::profile::REDUCE_ATTN_BAR
• glm5next_layer::profile::REDUCE_ATTN_ENQ
• glm5next_layer::profile::REDUCE_MLP
• glm5next_layer::profile::REDUCE_MLP_BAR
• glm5next_layer::profile::REDUCE_MLP_ENQ
• glm5next_mhc::GLM5NEXT_MHC_MODULE
• glm5next_mhc::MHC_MIX_MAX_TOKENS
• glm5next_mlp::FFN_MODULE
• glm5next_mlp::GEMM_MODULE
• glm5next_mlp::KERNEL_MAX_TOP_K
• glm5next_mlp::MOE_GROUPED_MODULE
• glm5next_mlp::MOE_MODULE
• glm5next_mlp::W4A16_GEMV_MODULE
• glm5next_mlp::W4A16_MODULE
• glm5next_mlp::forward::MOE_ROW_BATCH_MAX_ROWS
• glm5next_mlp::forward::MOE_ROW_UNION_MAX_IDS
• kimi_k3::KDA_L2_EPS
• kimi_k3::kda::KDA_L2_EPS
• kimi_k3::kda_cuda::CONV_ENTRY
• kimi_k3::kda_cuda::MODULE
• kimi_k3::kda_cuda::RECURRENT_ENTRY
• kimi_k3::mla_cuda::MODULE
• kimi_k3::mla_cuda::ROPE_ENTRY
• kimi_k3::mla_cuda::SDPA_ENTRY
• kimi_k3::moe_cuda::E8M0_ENTRY
• kimi_k3::moe_cuda::MODULE
• kimi_k3::moe_cuda::PTRTABLE_E8M0