Man pages for ggmlR
'GGML' Tensor Operations for Machine Learning
| abind_first | Bind Two Arrays Along the First Dimension |
| ag_add | Element-wise addition with broadcasting |
| ag_batch_norm | Create a Batch Normalisation layer |
| ag_clamp | Element-wise clamp |
| ag_cross_entropy_loss | Categorical Cross-Entropy loss |
| ag_dataloader | Create a mini-batch data loader |
| ag_default_device | Return the current default compute device |
| ag_default_dtype | Return the current default dtype for GPU operations |
| ag_device | Set the default compute device for ag_* operations |
| ag_dropout | Create a Dropout layer |
| ag_dtype | Set the default floating-point precision for ag_* GPU... |
| ag_embedding | Create an Embedding layer |
| ag_eval | Switch a layer or sequential model to eval mode |
| ag_exp | Element-wise exponential |
| ag_gradcheck | Numerical gradient check (like torch.autograd.gradcheck) |
| ag_linear | Create a dense layer with learnable parameters |
| ag_load_model | Load an autograd module from a saved state |
| ag_log | Element-wise natural logarithm |
| ag_matmul | Matrix multiplication |
| ag_mean | Mean of elements (or along a dim) |
| ag_mse_loss | Mean Squared Error loss |
| ag_mul | Element-wise multiplication |
| ag_multihead_attention | Create a Multi-Head Attention layer |
| ag_param | Create a parameter tensor (gradient tracked) |
| ag_pow | Element-wise power |
| ag_relu | ReLU activation |
| ag_reshape | Reshape tensor |
| ag_save_model | Save an autograd module's state to disk |
| ag_scale | Scale tensor by a scalar constant |
| ag_sequential | Create a sequential container of layers |
| ag_sigmoid | Sigmoid activation |
| ag_softmax | Softmax activation (column-wise) |
| ag_softmax_cross_entropy_loss | Fused softmax + cross-entropy loss (numerically stable) |
| ag_sub | Element-wise subtraction |
| ag_sum | Sum all elements (or along a dim): out = sum(x) |
| ag_tanh | Tanh activation |
| ag_tensor | Create a dynamic tensor (no gradient tracking) |
| ag_to_device | Move a tensor to the specified device |
| ag_train | Switch a layer or sequential model to training mode |
| ag_transpose | Transpose a tensor |
| as_gpu_matrix | Wrap a matrix so multiplies run on the GPU |
| as.matrix.ggml_matrix | Extract the underlying matrix from a ggml_matrix |
| augment.ggmlr_parsnip_model | Augment new data with predictions from a fitted ggml parsnip... |
| backward | Run backward pass from a scalar loss tensor |
| clip_grad_norm | Clip gradients by global L2 norm |
| compile | Compile a Model |
| dequantize_row_iq2_xxs | Dequantize Row (IQ) |
| dequantize_row_mxfp4 | Dequantize Row (MXFP4) |
| dequantize_row_nvfp4 | Dequantize NVFP4 Data |
| dequantize_row_q1_0 | Dequantize Q1_0 Data |
| dequantize_row_q2_K | Dequantize Row (K-quants) |
| dequantize_row_q4_0 | Dequantize Row (Q4_0) |
| dequantize_row_tq1_0 | Dequantize Row (Ternary) |
| dot-ggmlr_batch_shards | Split a batch index range into (near) equal contiguous shards |
| dot-ggmlr_largest_gene | Highest-expressed gene per cell (op = "largest_gene") |
| dot-ggmlr_neighbors_gpu | kNN + shared-nearest-neighbour graphs (op = "neighbors") |
| dot-ggmlr_normalize_gpu | GPU-accelerated LogNormalize (op = "normalize") |
| dot-ggmlr_pca_gpu | GPU-accelerated PCA on a dense expression matrix |
| dot-ggmlr_scale_gpu | GPU-accelerated ScaleData / z-score (op = "scale") |
| dot-ggmlr_umap_gpu | GPU-bound UMAP embedding (op = "umap") |
| dp_train | Data-parallel training across multiple GPUs |
| evaluate | Evaluate a Model |
| fit | Train a Model |
| ggml_abort_is_r_enabled | Check if R Abort Handler is Enabled |
| ggml_abs | Absolute Value (Graph) |
| ggml_abs_inplace | Absolute Value In-place (Graph) |
| ggml_add | Add tensors |
| ggml_add1 | Add Scalar to Tensor (Graph) |
| ggml_add_inplace | Element-wise Addition In-place (Graph) |
| ggml_add_rel_pos | Add Relative Position Bias (Graph) |
| ggml_apply | Apply a Layer Object to a Tensor Node |
| ggml_arange | Arange (Graph) |
| ggml_are_same_layout | Check if Two Tensors Have the Same Layout |
| ggml_are_same_shape | Compare Tensor Shapes |
| ggml_are_same_stride | Compare Tensor Strides |
| ggml_argmax | Argmax (Graph) |
| ggml_argsort | Argsort - Get Sorting Indices (Graph) |
| ggml_backend_alloc_ctx_tensors | Allocate Context Tensors to Backend |
| ggml_backend_buffer_clear | Clear buffer memory |
| ggml_backend_buffer_free | Free Backend Buffer |
| ggml_backend_buffer_get_size | Get Backend Buffer Size |
| ggml_backend_buffer_get_usage | Get buffer usage |
| ggml_backend_buffer_is_host | Check if buffer is host memory |
| ggml_backend_buffer_is_multi_buffer | Check if buffer is a multi-buffer |
| ggml_backend_buffer_name | Get Backend Buffer Name |
| ggml_backend_buffer_reset | Reset buffer |
| ggml_backend_buffer_set_usage | Set buffer usage hint |
| ggml_backend_buffer_usage_any | Buffer usage: Any |
| ggml_backend_buffer_usage_compute | Buffer usage: Compute |
| ggml_backend_buffer_usage_weights | Buffer usage: Weights |
| ggml_backend_cpu_init | Initialize CPU Backend |
| ggml_backend_cpu_set_n_threads | Set CPU Backend Threads |
| ggml_backend_dev_by_name | Get device by name |
| ggml_backend_dev_by_type | Get device by type |
| ggml_backend_dev_count | Get number of available devices |
| ggml_backend_dev_description | Get device description |
| ggml_backend_dev_get | Get device by index |
| ggml_backend_dev_get_props | Get device properties |
| ggml_backend_device_register | Register a device |
| ggml_backend_device_type_accel | Device type: Accelerator |
| ggml_backend_device_type_cpu | Device type: CPU |
| ggml_backend_device_type_gpu | Device type: GPU |
| ggml_backend_device_type_igpu | Device type: Integrated GPU |
| ggml_backend_dev_init | Initialize backend from device |
| ggml_backend_dev_memory | Get device memory |
| ggml_backend_dev_name | Get device name |
| ggml_backend_dev_offload_op | Check if device should offload operation |
| ggml_backend_dev_supports_buft | Check if device supports buffer type |
| ggml_backend_dev_supports_op | Check if device supports operation |
| ggml_backend_dev_type | Get device type |
| ggml_backend_event_free | Free event |
| ggml_backend_event_new | Create new event |
| ggml_backend_event_record | Record event |
| ggml_backend_event_synchronize | Synchronize event |
| ggml_backend_event_wait | Wait for event |
| ggml_backend_free | Free Backend |
| ggml_backend_get_device | Get device from backend |
| ggml_backend_graph_compute | Compute Graph with Backend |
| ggml_backend_graph_compute_async | Compute graph asynchronously |
| ggml_backend_graph_plan_compute | Execute graph plan |
| ggml_backend_graph_plan_create | Create graph execution plan |
| ggml_backend_graph_plan_free | Free graph execution plan |
| ggml_backend_init_best | Initialize best available backend |
| ggml_backend_init_by_name | Initialize backend by name |
| ggml_backend_init_by_type | Initialize backend by type |
| ggml_backend_load | Load backend from dynamic library |
| ggml_backend_load_all | Load all available backends |
| ggml_backend_meta_device | Create a Meta Backend Device |
| ggml_backend_multi_buffer_alloc_buffer | Allocate multi-buffer |
| ggml_backend_multi_buffer_set_usage | Set usage for all buffers in a multi-buffer |
| ggml_backend_name | Get Backend Name |
| ggml_backend_reg_by_name | Get backend registry by name |
| ggml_backend_reg_count | Get number of registered backends |
| ggml_backend_reg_dev_count | Get number of devices in registry |
| ggml_backend_reg_dev_get | Get device from registry |
| ggml_backend_reg_get | Get backend registry by index |
| ggml_backend_register | Register a backend |
| ggml_backend_reg_name | Get registry name |
| ggml_backend_sched_alloc_graph | Allocate graph on scheduler |
| ggml_backend_sched_free | Free backend scheduler |
| ggml_backend_sched_get_backend | Get backend from scheduler |
| ggml_backend_sched_get_n_backends | Get number of backends in scheduler |
| ggml_backend_sched_get_n_copies | Get number of tensor copies |
| ggml_backend_sched_get_n_splits | Get number of graph splits |
| ggml_backend_sched_get_tensor_backend | Get tensor backend assignment |
| ggml_backend_sched_graph_compute | Compute graph using scheduler |
| ggml_backend_sched_graph_compute_async | Compute graph asynchronously |
| ggml_backend_sched_new | Create a new backend scheduler |
| ggml_backend_sched_reserve | Reserve memory for scheduler |
| ggml_backend_sched_reset | Reset scheduler |
| ggml_backend_sched_set_tensor_backend | Set tensor backend assignment |
| ggml_backend_sched_synchronize | Synchronize scheduler |
| ggml_backend_synchronize | Synchronize backend |
| ggml_backend_tensor_copy_async | Copy tensor asynchronously between backends |
| ggml_backend_tensor_get_and_sync | Backend Tensor Get and Sync |
| ggml_backend_tensor_get_async | Get tensor data asynchronously |
| ggml_backend_tensor_get_data | Get Tensor Data via Backend |
| ggml_backend_tensor_get_f32_first | Get First Float from Backend Tensor |
| ggml_backend_tensor_set_async | Set tensor data asynchronously |
| ggml_backend_tensor_set_data | Set Tensor Data via Backend |
| ggml_backend_unload | Unload backend |
| ggml_batch_norm | Create a Batch Normalization Layer Object |
| ggml_blck_size | Get Block Size |
| ggml_build_forward_expand | Build forward expand |
| ggml_callback_early_stopping | Early stopping callback |
| ggml_can_repeat | Check If Tensor Can Be Repeated |
| ggml_ceil | Ceiling (Graph) |
| ggml_ceil_inplace | Ceiling In-place (Graph) |
| ggml_clamp | Clamp (Graph) |
| ggml_compile | Compile a Sequential Model |
| ggml_concat | Concatenate Tensors (Graph) |
| ggml_cont | Make Contiguous (Graph) |
| ggml_conv_1d | 1D Convolution (Graph) |
| ggml_conv_1d_dw | Depthwise 1D Convolution (Graph) |
| ggml_conv_2d | 2D Convolution (Graph) |
| ggml_conv_2d_direct | Direct 2D Convolution (Graph) |
| ggml_conv_2d_dw | Depthwise 2D Convolution (Graph) |
| ggml_conv_2d_dw_direct | Depthwise 2D Convolution, direct (Graph) |
| ggml_conv_transpose_1d | Transposed 1D Convolution (Graph) |
| ggml_conv_transpose_2d_p0 | Transposed 2D Convolution, zero padding (Graph) |
| ggml_cos | Cosine (Graph) |
| ggml_count_equal | Count Equal Elements (Graph) |
| ggml_cpu_add | Element-wise Addition (CPU Direct) |
| ggml_cpu_features | Get All CPU Features |
| ggml_cpu_get_rvv_vlen | Get RISC-V Vector Length |
| ggml_cpu_get_sve_cnt | Get SVE Vector Length (ARM) |
| ggml_cpu_has_amx_int8 | CPU Feature Detection - AMX INT8 |
| ggml_cpu_has_arm_fma | CPU Feature Detection - ARM FMA |
| ggml_cpu_has_avx | CPU Feature Detection - AVX |
| ggml_cpu_has_avx2 | CPU Feature Detection - AVX2 |
| ggml_cpu_has_avx512 | CPU Feature Detection - AVX-512 |
| ggml_cpu_has_avx512_bf16 | CPU Feature Detection - AVX-512 BF16 |
| ggml_cpu_has_avx512_vbmi | CPU Feature Detection - AVX-512 VBMI |
| ggml_cpu_has_avx512_vnni | CPU Feature Detection - AVX-512 VNNI |
| ggml_cpu_has_avx_vnni | CPU Feature Detection - AVX-VNNI |
| ggml_cpu_has_bmi2 | CPU Feature Detection - BMI2 |
| ggml_cpu_has_dotprod | CPU Feature Detection - Dot Product (ARM) |
| ggml_cpu_has_f16c | CPU Feature Detection - F16C |
| ggml_cpu_has_fma | CPU Feature Detection - FMA |
| ggml_cpu_has_fp16_va | CPU Feature Detection - FP16 Vector Arithmetic (ARM) |
| ggml_cpu_has_llamafile | CPU Feature Detection - Llamafile |
| ggml_cpu_has_matmul_int8 | CPU Feature Detection - INT8 Matrix Multiply (ARM) |
| ggml_cpu_has_neon | CPU Feature Detection - NEON (ARM) |
| ggml_cpu_has_riscv_v | CPU Feature Detection - RISC-V Vector |
| ggml_cpu_has_sme | CPU Feature Detection - SME (ARM) |
| ggml_cpu_has_sse3 | CPU Feature Detection - SSE3 |
| ggml_cpu_has_ssse3 | CPU Feature Detection - SSSE3 |
| ggml_cpu_has_sve | CPU Feature Detection - SVE (ARM) |
| ggml_cpu_has_vsx | CPU Feature Detection - VSX (PowerPC) |
| ggml_cpu_has_vxe | CPU Feature Detection - VXE (IBM z/Architecture) |
| ggml_cpu_has_wasm_simd | CPU Feature Detection - WebAssembly SIMD |
| ggml_cpu_mul | Element-wise Multiplication (CPU Direct) |
| ggml_cpy | Copy Tensor with Type Conversion (Graph) |
| ggml_crossprod | GPU cross product (drop-in for 'crossprod') |
| ggml_cycles | Get CPU Cycles |
| ggml_cycles_per_ms | Get CPU Cycles per Millisecond |
| ggml_default_mlp | Default MLP builder for classification and regression |
| ggml_dense | Create a Dense Layer Object |
| ggml_diag | Diagonal Matrix (Graph) |
| ggml_diag_mask_inf | Diagonal Mask with -Inf (Graph) |
| ggml_diag_mask_inf_inplace | Diagonal Mask with -Inf In-place (Graph) |
| ggml_diag_mask_zero | Diagonal Mask with Zero (Graph) |
| ggml_div | Element-wise Division (Graph) |
| ggml_div_inplace | Element-wise Division In-place (Graph) |
| ggml_dup | Duplicate Tensor (Graph) |
| ggml_dup_inplace | Duplicate Tensor In-place (Graph) |
| ggml_dup_tensor | Duplicate Tensor |
| ggml_element_size | Get Element Size |
| ggml_elu | ELU Activation (Graph) |
| ggml_elu_inplace | ELU Activation In-place (Graph) |
| ggml_embedding | Create an Embedding Layer Object |
| ggml_estimate_memory | Estimate Required Memory |
| ggml_evaluate | Evaluate a Trained Model |
| ggml_exp | Exponential (Graph) |
| ggml_exp_inplace | Exponential In-place (Graph) |
| ggml_extract | Extract a feature-by-cell matrix from a single-cell container |
| ggml_fit | Train a Model (dispatcher) |
| ggml_fit_opt | Fit model with R-side epoch loop and callbacks |
| ggml_flash_attn_back | Flash Attention Backward (Graph) |
| ggml_flash_attn_ext | Flash Attention (Graph) |
| ggml_floor | Floor (Graph) |
| ggml_floor_inplace | Floor In-place (Graph) |
| ggml_free | Free GGML context |
| ggml_freeze_weights | Freeze Layer Weights |
| ggml_ftype_to_ggml_type | Convert ftype to ggml_type |
| ggml_gallocr_alloc_graph | Allocate Memory for Graph |
| ggml_gallocr_free | Free Graph Allocator |
| ggml_gallocr_get_buffer_size | Get Graph Allocator Buffer Size |
| ggml_gallocr_new | Create Graph Allocator |
| ggml_gallocr_reserve | Reserve Memory for Graph |
| ggml_geglu | GeGLU (GELU Gated Linear Unit) (Graph) |
| ggml_geglu_quick | GeGLU Quick (Fast GeGLU) (Graph) |
| ggml_geglu_split | GeGLU Split (Graph) |
| ggml_gelu | GELU Activation (Graph) |
| ggml_gelu_erf | Exact GELU Activation (Graph) |
| ggml_gelu_inplace | GELU Activation In-place (Graph) |
| ggml_gelu_quick | GELU Quick Activation (Graph) |
| ggml_get_f32 | Get F32 data |
| ggml_get_f32_nd | Get Single Float Value by N-D Index |
| ggml_get_first_tensor | Get First Tensor from Context |
| ggml_get_i32 | Get I32 Data |
| ggml_get_i32_nd | Get Single Int32 Value by N-D Index |
| ggml_get_layer | Get a Layer from a Sequential Model |
| ggml_get_max_tensor_size | Get Maximum Tensor Size |
| ggml_get_mem_size | Get Context Memory Size |
| ggml_get_name | Get Tensor Name |
| ggml_get_next_tensor | Get Next Tensor from Context |
| ggml_get_no_alloc | Get No Allocation Mode |
| ggml_get_n_threads | Get Number of Threads |
| ggml_get_op_params | Get Tensor Operation Parameters |
| ggml_get_op_params_f32 | Get Float Op Parameter |
| ggml_get_op_params_i32 | Get Integer Op Parameter |
| ggml_get_rel_pos | Get Relative Position (Graph) |
| ggml_get_rows | Get Rows by Indices (Graph) |
| ggml_get_rows_back | Get Rows Backward (Graph) |
| ggml_get_unary_op | Get Unary Operation from Tensor |
| ggml_glu | Generic GLU (Gated Linear Unit) (Graph) |
| GGML_GLU_OP_REGLU | GLU Operation Types |
| ggml_glu_split | Generic GLU Split (Graph) |
| ggml_graph_compute | Compute graph |
| ggml_graph_compute_with_ctx | Compute Graph with Context (Alternative Method) |
| ggml_graph_dump_dot | Export Graph to DOT Format |
| ggml_graph_get_tensor | Get Tensor from Graph by Name |
| ggml_graph_n_nodes | Get Number of Nodes in Graph |
| ggml_graph_node | Get Graph Node |
| ggml_graph_overhead | Get Graph Overhead |
| ggml_graph_print | Print Graph Information |
| ggml_graph_reset | Reset Graph (for backpropagation) |
| ggml_graph_view | Create a View of a Subgraph |
| ggml_group_norm | Group Normalization (Graph) |
| ggml_group_norm_inplace | Group Normalization In-place (Graph) |
| ggml_gru | Create a GRU Layer Object |
| ggml_hardsigmoid | Hard Sigmoid Activation (Graph) |
| ggml_hardswish | Hard Swish Activation (Graph) |
| ggml_im2col | Image to Column (Graph) |
| ggml_init | Initialize GGML context |
| ggml_init_auto | Create Context with Auto-sizing |
| ggml_inject | Inject a single-cell result back into its container |
| ggml_input | Declare a Functional API Input Tensor |
| ggml_is_available | Check if GGML is available |
| ggml_is_contiguous | Check if Tensor is Contiguous |
| ggml_is_contiguous_0 | Check Tensor Contiguity (Dimension 0) |
| ggml_is_contiguous_1 | Check Tensor Contiguity (Dimensions >= 1) |
| ggml_is_contiguous_2 | Check Tensor Contiguity (Dimensions >= 2) |
| ggml_is_contiguous_channels | Check Channel-wise Contiguity |
| ggml_is_contiguously_allocated | Check If Tensor is Contiguously Allocated |
| ggml_is_contiguous_rows | Check Row-wise Contiguity |
| ggml_is_permuted | Check if Tensor is Permuted |
| ggml_is_quantized | Check If Type is Quantized |
| ggml_is_transposed | Check if Tensor is Transposed |
| ggml_l2_norm | L2 Normalization (Graph) |
| ggml_l2_norm_inplace | L2 Normalization In-place (Graph) |
| ggml_layer_add | Element-wise Addition of Two Tensor Nodes |
| ggml_layer_batch_norm | Add Batch Normalization Layer |
| ggml_layer_concatenate | Concatenate Tensor Nodes Along an Axis |
| ggml_layer_conv_1d | Create a Conv1D Layer Object |
| ggml_layer_conv_2d | Create a Conv2D Layer Object |
| ggml_layer_dense | Add Dense (Fully Connected) Layer |
| ggml_layer_dropout | Add Dropout Layer |
| ggml_layer_embedding | Add Embedding Layer |
| ggml_layer_flatten | Add Flatten Layer |
| ggml_layer_global_average_pooling_2d | Global Average Pooling for 2D Feature Maps |
| ggml_layer_global_max_pooling_2d | Global Max Pooling for 2D Feature Maps |
| ggml_layer_gru | Add a GRU Layer |
| ggml_layer_lstm | Add an LSTM Layer |
| ggml_layer_max_pooling_2d | Add 2D Max Pooling Layer |
| ggml_leaky_relu | Leaky ReLU Activation (Graph) |
| ggml_load_model | Load a Full Model (Architecture + Weights) |
| ggml_load_weights | Load Model Weights from File |
| ggml_log | Natural Logarithm (Graph) |
| ggml_log_inplace | Natural Logarithm In-place (Graph) |
| ggml_log_is_r_enabled | Check if R Logging is Enabled |
| ggml_log_set_default | Restore Default GGML Logging |
| ggml_log_set_r | Enable R-compatible GGML Logging |
| ggml_lstm | Create an LSTM Layer Object |
| ggml_marshal_model | Marshal a ggmlR model to an in-memory container |
| ggml_matmul | GPU matrix multiply (drop-in for '%*%') |
| ggml_matmul_f64 | GPU double-precision matrix multiply |
| ggml_matrix-class | A GPU-backed matrix wrapper |
| ggml_mean | Mean (Graph) |
| ggml_model | Create a Functional Model |
| ggml_model_backend | Backend a fitted ggml model actually ran on |
| ggml_model_sequential | Create a Sequential Neural Network Model |
| ggml_mul | Multiply tensors |
| ggml_mul_inplace | Element-wise Multiplication In-place (Graph) |
| ggml_mul_mat | Matrix Multiplication (Graph) |
| ggml_mul_mat_id | Matrix Multiplication with Expert Selection (Graph) |
| ggml_nbytes | Get number of bytes |
| ggml_n_dims | Get Number of Dimensions |
| ggml_neg | Negation (Graph) |
| ggml_neg_inplace | Negation In-place (Graph) |
| ggml_nelements | Get number of elements |
| ggml_new_f32 | Create Scalar F32 Tensor |
| ggml_new_i32 | Create Scalar I32 Tensor |
| ggml_new_tensor | Create Tensor with Arbitrary Dimensions |
| ggml_new_tensor_1d | Create 1D tensor |
| ggml_new_tensor_2d | Create 2D tensor |
| ggml_new_tensor_3d | Create 3D Tensor |
| ggml_new_tensor_4d | Create 4D Tensor |
| ggml_norm | Layer Normalization (Graph) |
| ggml_norm_inplace | Layer Normalization In-place (Graph) |
| ggml_nrows | Get Number of Rows |
| ggml_op_can_inplace | Check if Operation Can Be Done In-place |
| ggml_op_desc | Get Operation Description from Tensor |
| ggml_op_name | Get Operation Name |
| ggml_ops_registry | Supported single-cell operations |
| ggml_op_symbol | Get Operation Symbol |
| ggml_opt_alloc | Allocate graph for evaluation |
| ggml_opt_context_optimizer_type | Get optimizer type from context |
| ggml_opt_dataset_data | Get data tensor from dataset |
| ggml_opt_dataset_free | Free optimization dataset |
| ggml_opt_dataset_get_batch | Get batch from dataset |
| ggml_opt_dataset_init | Create a new optimization dataset |
| ggml_opt_dataset_labels | Get labels tensor from dataset |
| ggml_opt_dataset_ndata | Get number of datapoints in dataset |
| ggml_opt_dataset_shuffle | Shuffle dataset |
| ggml_opt_dataset_weights | Get dataset per-datapoint weights tensor |
| ggml_opt_default_params | Get default optimizer parameters |
| ggml_opt_epoch | Run one training epoch |
| ggml_opt_eval | Evaluate model |
| ggml_opt_fit | Fit model to dataset |
| ggml_opt_free | Free optimizer context |
| ggml_opt_get_lr | Get current learning rate from optimizer context |
| ggml_opt_grad_acc | Get gradient accumulator for a tensor |
| ggml_opt_init | Initialize optimizer context |
| ggml_opt_init_for_fit | Initialize optimizer context for R-side epoch loop |
| ggml_opt_inputs | Get inputs tensor from optimizer context |
| ggml_opt_labels | Get labels tensor from optimizer context |
| ggml_opt_loss | Get loss tensor from optimizer context |
| ggml_opt_loss_type_cross_entropy | Loss type: Cross Entropy |
| ggml_opt_loss_type_mean | Loss type: Mean |
| ggml_opt_loss_type_mse | Loss type: Mean Squared Error |
| ggml_opt_loss_type_sum | Loss type: Sum |
| ggml_opt_loss_type_weighted_mse | Loss type: Weighted Mean Squared Error |
| ggml_opt_ncorrect | Get number of correct predictions tensor |
| ggml_opt_optimizer_name | Get optimizer name |
| ggml_opt_optimizer_type_adamw | Optimizer type: AdamW |
| ggml_opt_optimizer_type_sgd | Optimizer type: SGD |
| ggml_opt_outputs | Get outputs tensor from optimizer context |
| ggml_opt_pred | Get predictions tensor from optimizer context |
| ggml_opt_prepare_alloc | Prepare allocation for non-static graphs |
| ggml_opt_reset | Reset optimizer context |
| ggml_opt_result_accuracy | Get accuracy from result |
| ggml_opt_result_free | Free optimization result |
| ggml_opt_result_init | Initialize optimization result |
| ggml_opt_result_loss | Get loss from result |
| ggml_opt_result_ndata | Get number of datapoints from result |
| ggml_opt_result_pred | Get predictions from result |
| ggml_opt_result_reset | Reset optimization result |
| ggml_opt_set_lr | Set learning rate in optimizer context |
| ggml_opt_static_graphs | Check if using static graphs |
| ggml_out_prod | Outer Product (Graph) |
| ggml_pad | Pad Tensor with Zeros (Graph) |
| ggml_pad_reflect_1d | Reflective 1D Padding (Graph) |
| ggml_permute | Permute Tensor Dimensions (Graph) |
| ggml_pool_1d | 1D Pooling (Graph) |
| ggml_pool_2d | 2D Pooling (Graph) |
| ggml_pop_layer | Remove the Last Layer from a Sequential Model |
| ggml_pp_dp_forward | Pipeline + data parallel forward pass (PP x DP) |
| ggml_pp_forward | Pipeline-parallel forward pass across per-device layer stages |
| ggml_predict | Get Predictions from a Trained Model |
| ggml_predict_classes | Predict Classes from a Trained Model |
| ggml_print_mem_status | Print Context Memory Status |
| ggml_print_objects | Print Objects in Context |
| ggml_quant_block_info | Get Quantization Block Info |
| ggml_quantize_chunk | Quantize Data Chunk |
| ggml_quantize_free | Free Quantization Resources |
| ggml_quantize_init | Initialize Quantization Tables |
| ggml_quantize_requires_imatrix | Check if Quantization Requires Importance Matrix |
| ggml_reglu | ReGLU (ReLU Gated Linear Unit) (Graph) |
| ggml_reglu_split | ReGLU Split (Graph) |
| ggml_relu | ReLU Activation (Graph) |
| ggml_relu_inplace | ReLU Activation In-place (Graph) |
| ggml_repeat | Repeat (Graph) |
| ggml_repeat_back | Repeat Backward (Graph) |
| ggml_reset | Reset GGML Context |
| ggml_reshape_1d | Reshape to 1D (Graph) |
| ggml_reshape_2d | Reshape to 2D (Graph) |
| ggml_reshape_3d | Reshape to 3D (Graph) |
| ggml_reshape_4d | Reshape to 4D (Graph) |
| ggml_result | Construct a single-cell result |
| ggml_rms_norm | RMS Normalization (Graph) |
| ggml_rms_norm_back | RMS Norm Backward (Graph) |
| ggml_rms_norm_inplace | RMS Normalization In-place (Graph) |
| ggml_roll | Roll (Graph) |
| ggml_rope | Rotary Position Embedding (Graph) |
| ggml_rope_ext | Extended RoPE with Frequency Scaling (Graph) |
| ggml_rope_ext_back | RoPE Extended Backward (Graph) |
| ggml_rope_ext_inplace | Extended RoPE Inplace (Graph) |
| ggml_rope_inplace | Rotary Position Embedding In-place (Graph) |
| ggml_rope_multi | Multi-RoPE for Vision Models (Graph) |
| ggml_rope_multi_inplace | Multi-RoPE Inplace (Graph) |
| ggml_round | Round (Graph) |
| ggml_round_inplace | Round In-place (Graph) |
| ggmlR-package | ggmlR: 'GGML' Tensor Operations for Machine Learning |
| ggmlr_parsnip_fit_classif | parsnip ggml engine: classification fit |
| ggmlr_parsnip_fit_regr | parsnip ggml engine: regression fit |
| ggml_run | Run a single-cell task on the GGML backend |
| ggml_save_model | Save a Full Model (Architecture + Weights) |
| ggml_save_weights | Save Model Weights to File |
| ggml_scale | Scale (Graph) |
| ggml_scale_inplace | Scale Tensor In-place (Graph) |
| ggml_schedule_cosine_decay | Cosine annealing LR scheduler |
| ggml_schedule_reduce_on_plateau | Reduce on plateau LR scheduler |
| ggml_schedule_step_decay | Step decay LR scheduler |
| ggml_set | Set Tensor Region (Graph) |
| ggml_set_1d | Set 1D Tensor Region (Graph) |
| ggml_set_2d | Set 2D Tensor Region (Graph) |
| ggml_set_abort_callback_default | Restore Default Abort Behavior |
| ggml_set_abort_callback_r | Enable R-compatible Abort Handling |
| ggml_set_f32 | Set F32 data |
| ggml_set_f32_nd | Set Single Float Value by N-D Index |
| ggml_set_i32 | Set I32 Data |
| ggml_set_i32_nd | Set Single Int32 Value by N-D Index |
| ggml_set_input | Mark Tensor as Input |
| ggml_set_name | Set Tensor Name |
| ggml_set_no_alloc | Set No Allocation Mode |
| ggml_set_n_threads | Set Number of Threads |
| ggml_set_omp_threads | Set OpenMP Thread Count |
| ggml_set_op_params | Set Tensor Operation Parameters |
| ggml_set_op_params_f32 | Set Float Op Parameter |
| ggml_set_op_params_i32 | Set Integer Op Parameter |
| ggml_set_output | Mark Tensor as Output |
| ggml_set_param | Set Tensor as Trainable Parameter |
| ggml_set_seed | Set the random seed for reproducible ggmlR runs |
| ggml_set_zero | Set Tensor to Zero |
| ggml_sgn | Sign Function (Graph) |
| ggml_sigmoid | Sigmoid Activation (Graph) |
| ggml_sigmoid_inplace | Sigmoid Activation In-place (Graph) |
| ggml_silu | SiLU Activation (Graph) |
| ggml_silu_back | SiLU Backward (Graph) |
| ggml_silu_inplace | SiLU Activation In-place (Graph) |
| ggml_sin | Sine (Graph) |
| ggml_soft_max | Softmax (Graph) |
| ggml_soft_max_ext | Extended Softmax with Masking and Scaling (Graph) |
| ggml_soft_max_ext_back | Softmax Backward Extended (Graph) |
| ggml_soft_max_ext_back_inplace | Extended Softmax Backward Inplace (Graph) |
| ggml_soft_max_ext_inplace | Extended Softmax Inplace (Graph) |
| ggml_soft_max_inplace | Softmax In-place (Graph) |
| ggml_softplus | Softplus Activation (Graph) |
| ggml_softplus_inplace | Softplus Activation In-place (Graph) |
| GGML_SORT_ORDER_ASC | Sort Order Constants |
| ggml_sqr | Square (Graph) |
| ggml_sqr_inplace | Square In-place (Graph) |
| ggml_sqrt | Square Root (Graph) |
| ggml_sqrt_inplace | Square Root In-place (Graph) |
| ggml_step | Step Function (Graph) |
| ggml_sub | Element-wise Subtraction (Graph) |
| ggml_sub_inplace | Element-wise Subtraction In-place (Graph) |
| ggml_sum | Sum (Graph) |
| ggml_sum_rows | Sum Rows (Graph) |
| ggml_swiglu | SwiGLU (Swish/SiLU Gated Linear Unit) (Graph) |
| ggml_swiglu_split | SwiGLU Split (Graph) |
| ggml_tanh | Tanh Activation (Graph) |
| ggml_tanh_inplace | Tanh Activation In-place (Graph) |
| ggml_task | Construct a single-cell compute task |
| ggml_tcrossprod | GPU transposed cross product (drop-in for 'tcrossprod') |
| ggml_tensor_copy | Copy Tensor Data |
| ggml_tensor_nb | Get Tensor Strides (nb) |
| ggml_tensor_num | Count Tensors in Context |
| ggml_tensor_overhead | Get Tensor Overhead |
| ggml_tensor_set_f32_scalar | Fill Tensor with Scalar |
| ggml_tensor_shape | Get Tensor Shape |
| ggml_tensor_type | Get Tensor Type |
| ggml_test | Test GGML |
| ggml_time_init | Initialize GGML Timer |
| ggml_time_ms | Get Time in Milliseconds |
| ggml_timestep_embedding | Timestep Embedding (Graph Operation) |
| ggml_time_us | Get Time in Microseconds |
| ggml_top_k | Top-K Indices (Graph) |
| ggml_tp_dp_forward | TPxDP hybrid matrix multiply across replicas of Vulkan device... |
| ggml_training_history | Training history of a fitted ggml model |
| ggml_transpose | Transpose (Graph) |
| GGML_TYPE_F32 | GGML Data Types |
| ggml_type_name | Get Type Name |
| ggml_type_size | Get Type Size in Bytes |
| ggml_type_sizef | Get Type Size as Float |
| ggml_unary_op_name | Get Unary Operation Name |
| ggml_unfreeze_weights | Unfreeze Layer Weights |
| ggml_unmarshal_model | Unmarshal a ggmlR model from an in-memory container |
| ggml_upscale | Upscale Tensor (Graph) |
| ggml_used_mem | Get Used Memory |
| ggml_version | Get GGML version |
| ggml_view_1d | 1D View with Byte Offset (Graph) |
| ggml_view_2d | 2D View with Byte Offset (Graph) |
| ggml_view_3d | 3D View with Byte Offset (Graph) |
| ggml_view_4d | 4D View with Byte Offset (Graph) |
| ggml_view_tensor | View Tensor |
| ggml_vulkan_available | Check if Vulkan support is available |
| ggml_vulkan_backend_name | Get Vulkan backend name |
| ggml_vulkan_device_caps | Get Vulkan device capabilities |
| ggml_vulkan_device_count | Get number of Vulkan devices |
| ggml_vulkan_device_description | Get Vulkan device description |
| ggml_vulkan_device_groups | Probe Vulkan device groups (NVLink / multi-GPU peer access) |
| ggml_vulkan_device_memory | Get Vulkan device memory |
| ggml_vulkan_free | Free Vulkan backend |
| ggml_vulkan_hard_exit_available | Is the Vulkan hard-exit path compiled in? |
| ggml_vulkan_init | Initialize Vulkan backend |
| ggml_vulkan_is_backend | Check if backend is Vulkan |
| ggml_vulkan_list_devices | List all Vulkan devices |
| ggml_vulkan_p2p_selftest | Opaque-fd device-to-device P2P self-test |
| ggml_vulkan_shutdown | Explicitly tear down the Vulkan instance and devices |
| ggml_vulkan_split_buffer_type | Create a Vulkan tensor-split buffer type |
| ggml_vulkan_split_mul_mat | Tensor-parallel matrix multiply across Vulkan devices |
| ggml_vulkan_split_row_ranges | Compute tensor-parallel row-split ranges |
| ggml_vulkan_stage_handoff | Hand a pipeline stage's activation tensor to the next stage's... |
| ggml_vulkan_status | Print Vulkan status |
| ggml_win_part | Window Partition (Graph) |
| ggml_win_unpart | Window Un-partition (Graph) |
| ggml_with_temp_ctx | Execute with Temporary Context |
| gguf_free | Free GGUF Resources |
| gguf_load | Load a GGUF File |
| gguf_metadata | Get GGUF Metadata |
| gguf_tensor_data | Extract Tensor Data |
| gguf_tensor_info | Get Tensor Info |
| gguf_tensor_names | List Tensor Names in a GGUF File |
| glance.ggmlr_parsnip_model | One-row summary of a fitted ggml parsnip model |
| iq2xs_free_impl | Free IQ2 Quantization Tables |
| iq2xs_init_impl | Initialize IQ2 Quantization Tables |
| iq3xs_free_impl | Free IQ3 Quantization Tables |
| iq3xs_init_impl | Initialize IQ3 Quantization Tables |
| is_ag_tensor | Check if object is an ag_tensor |
| lr_scheduler_cosine | Cosine-annealing learning rate scheduler |
| lr_scheduler_step | Step-decay learning rate scheduler |
| nn_apply_activation | Apply activation function |
| nn_build_batch_norm | Build batch_norm forward pass |
| nn_build_conv_1d | Build conv_1d forward pass |
| nn_build_conv_2d | Build conv_2d forward pass |
| nn_build_dense | Build dense forward pass |
| nn_build_dropout | Build dropout forward pass |
| nn_build_embedding | Build embedding forward pass |
| nn_build_flatten | Build flatten forward pass |
| nn_build_functional_graph | Build ggml computation graph for a functional model |
| nn_build_functional_node | Build a single ggml tensor for one functional node |
| nn_build_global_average_pooling_2d | Build global_average_pooling_2d forward pass |
| nn_build_global_max_pooling_2d | Build global_max_pooling_2d forward pass |
| nn_build_graph | Build computation graph with allocated weights and inputs |
| nn_build_gru | Build GRU forward pass for Sequential model |
| nn_build_layer | Build a layer's forward pass |
| nn_build_lstm | Build LSTM forward pass for Sequential model |
| nn_build_max_pooling_2d | Build max_pooling_2d forward pass |
| nn_count_layer_params | Count parameters for a single layer |
| nn_functional_output_shape | Infer output shape of a functional node given its parent... |
| nn_gru_step | Build one GRU step |
| nn_infer_shapes | Infer shapes for all layers in model |
| nn_init_glorot_uniform | Initialize weight tensor with Glorot uniform distribution |
| nn_init_he_uniform | Initialize weight tensor with He uniform distribution |
| nn_init_recurrent_uniform | Initialize recurrent weight tensor with small deterministic... |
| nn_init_zeros | Initialize bias tensor to zeros |
| nn_lstm_step | Build one LSTM step |
| nn_topo_sort | Topologically sort nodes reachable from output nodes |
| onnx_device_info | ONNX model device/scheduler diagnostics |
| onnx_inputs | List ONNX model inputs |
| onnx_load | Load an ONNX model |
| onnx_run | Run ONNX model inference |
| onnx_summary | ONNX model summary |
| optimizer_adam | Create an Adam optimizer |
| optimizer_sgd | Create an SGD optimizer |
| plot.ggml_history | Plot training history |
| predict.ggml_sequential_model | Predict with a Trained Model |
| print.ag_tensor | Print method for ag_tensor |
| print.ggml_functional_model | Print method for ggml_functional_model |
| print.ggml_history | Print method for ggml_history |
| print.ggml_sequential_model | Print method for ggml_sequential_model |
| print.onnx_model | Print ONNX model summary |
| quantize_iq2_xxs | Quantize Data (IQ) |
| quantize_mxfp4 | Quantize Data (MXFP4) |
| quantize_nvfp4 | Quantize Data (NVFP4) |
| quantize_q1_0 | Quantize Data (Q1_0) |
| quantize_q2_K | Quantize Data (K-quants) |
| quantize_q4_0 | Quantize Data (Q4_0) |
| quantize_row_iq3_xxs_ref | Quantize Row Reference (IQ) |
| quantize_row_mxfp4_ref | Quantize Row Reference (MXFP4) |
| quantize_row_q2_K_ref | Quantize Row Reference (K-quants) |
| quantize_row_q4_0_ref | Quantize Row Reference (Basic) |
| quantize_row_tq1_0_ref | Quantize Row Reference (Ternary) |
| quantize_tq1_0 | Quantize Data (Ternary) |
| reexports | Objects exported from other packages |
| rope_types | RoPE Mode Constants |
| RunGGML | Run a GGML GPU operation on a Seurat object |
| slice_first_dim | Slice an Array or Matrix Along Its First Dimension |
| summary.ggml_sequential_model | Summary method for ggml_sequential_model |
| tidy.ggmlr_parsnip_model | Tidy a fitted ggml parsnip model into a per-layer table |
| with_grad_tape | Run code with gradient tape enabled |
