# HELIA operator inventory Source: v1.21.2, commit 70611e608778a432566e30965e8b4bc7ea75aa36. Types below describe inputs. Support depends on shapes, output and weight types, backend and build settings. A listed type can use optimized, reference or storage-only execution. ## ABS Family: Arithmetic Input types: int8, int16, float32 Reference implementation in the HELIA adapter source. Quantized scaling requirements apply. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/elementwise.cc) ## ADD Family: Arithmetic Input types: int8, int16, float32, float16, int32 Float identical-shape inputs support arbitrary rank. Optimized broadcasting supports rank <= 4; higher-rank FP32 uses Reference and FP16 is rejected. int32 follows the adapter reference path. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/add.cc) ## ARG_MAX Family: Reduction Input types: int8, float32, float16 FP16 requires CORE >= 7.35.0, rank 1..4, INT32 axis/output and positive reduced extent. FP32/int8 use Reference. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/arg_min_max.cc) ## ARG_MIN Family: Reduction Input types: int8, float32, float16 FP16 requires CORE >= 7.35.0, rank 1..4, INT32 axis/output and positive reduced extent. FP32/int8 use Reference. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/arg_min_max.cc) ## AVERAGE_POOL_2D Family: Compute Input types: int8, int16, float32, float16 Four-dimensional tensors; check filter/stride/padding and output dimensions. Kernel failure is returned to the application. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/pooling.cc) ## BATCH_MATMUL Family: Compute Input types: int8, int16, float32, float16 Check batch broadcasting, adjoint flags and quantized weight/activation combinations against the adapter. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/batch_matmul.cc) ## CONCATENATION Family: Data movement Input types: int8, int16, float32, float16, int32, int64, bool Optimized float paths support rank <= 4; higher-rank float inputs use Reference. Input/output type and quantization must agree. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/concatenation.cc) ## CONV_2D Family: Compute Input types: int8, int16, float32, float16 Quantized weight/bias combinations are constrained. int4 weights use int8 activations. Float grouped convolution uses Reference for FP32 and is rejected for FP16. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/conv.cc) ## COS Family: Arithmetic Input types: float32 Reference implementation in the HELIA adapter source; Float32 input/output. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/elementwise.cc) ## DEPTHWISE_CONV_2D Family: Compute Input types: int8, int16, float32, float16 Quantized weight/bias combinations and depth multiplier are constrained. int4 is a weight type with int8 activations; it is not a general activation type. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/depthwise_conv.cc) ## DEQUANTIZE Family: Quantization Input types: int8, int16, uint8, float16 Output is FP32. FP16 means weight-storage widening, not FP16 computation; it works without ARM_NN_ENABLE_F16. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/dequantize.cc) ## EQUAL Family: Comparison Input types: bool, float32, int8, int16, int32, int64 Boolean output. Supported input types depend on the comparison; input shapes and broadcast rules must match. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/comparisons.cc) ## FILL Family: Data movement Input types: int8, int16, float32, float16, int32, int64 Constant dimensions and scalar fill value. Float APIs require CORE >= 7.33.0 and matching enabled feature. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/fill.cc) ## FULLY_CONNECTED Family: Compute Input types: int8, int16, float32, float16 Quantized activation/weight/bias types must match a supported combination; includes int4 weights with int8 activations. Float paths require matching CORE features. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/fully_connected.cc) ## GATHER Family: Data movement Input types: int8, int16, float32, float16 CORE >= 7.34.0. Native rank <= 4. Scalar indices use float native paths or int8 Reference; int16 rejects them. Higher-rank FP32/int8 use Reference; FP16/int16 reject. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/gather.cc) ## GATHER_ND Family: Data movement Input types: int8, int16, float32, float16 CORE >= 7.34.0. Native ranks <= 4 and positive index tuple width. Higher ranks and zero-width tuples use FP32/int8 Reference; FP16/int16 reject. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/gather_nd.cc) ## GREATER Family: Comparison Input types: float32, int8, int16, int32, int64 Boolean output. Check broadcast and operand types; no FP16 comparison branch. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/comparisons.cc) ## GREATER_EQUAL Family: Comparison Input types: float32, int8, int16, int32, int64 Boolean output. Check broadcast and operand types; no FP16 comparison branch. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/comparisons.cc) ## HARD_SWISH Family: Activation Input types: int8, int16, float32, float16 FP16 requires the CORE float feature; without it preparation fails. FP32 can use Reference. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/hard_swish.cc) ## LEAKY_RELU Family: Activation Input types: int8, int16, float32 Float32 uses the reference path. No FP16 execution branch in this adapter. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/leaky_relu.cc) ## LESS Family: Comparison Input types: float32, int8, int16, int32, int64 Boolean output. Check broadcast and operand types; no FP16 comparison branch. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/comparisons.cc) ## LESS_EQUAL Family: Comparison Input types: float32, int8, int16, int32, int64 Boolean output. Check broadcast and operand types; no FP16 comparison branch. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/comparisons.cc) ## LOG Family: Arithmetic Input types: float32 Reference implementation in the HELIA adapter source; Float32 input/output. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/elementwise.cc) ## LOGICAL_NOT Family: Comparison Input types: bool Reference boolean operation. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/elementwise.cc) ## LOGISTIC Family: Activation Input types: int8, int16, float32, float16 int8 uses a lookup table; int16 and enabled floats have HELIA paths. NaN is not supported by the optimized float activation path. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/logistic.cc) ## MAXIMUM Family: Arithmetic Input types: int8, int16, float32, float16, int32, int64 Optimized paths support rank <= 4. Higher-rank FP32/int8/int16 use Reference; FP16 is rejected. int32/int64 use Reference. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/maximum_minimum.cc) ## MAX_POOL_2D Family: Compute Input types: int8, int16, float32, float16 Four-dimensional tensors; check filter/stride/padding and output dimensions. Kernel failure is returned to the application. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/pooling.cc) ## MEAN Family: Reduction Input types: int8, int16, float32, float16 Float optimized paths support rank <= 4 and axis sets; higher-rank FP32 uses Reference, FP16 is rejected. See reduce_common.cc. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/reduce.cc) ## MINIMUM Family: Arithmetic Input types: int8, int16, float32, float16, int32, int64 Optimized paths support rank <= 4. Higher-rank FP32/int8/int16 use Reference; FP16 is rejected. int32/int64 use Reference. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/maximum_minimum.cc) ## MUL Family: Arithmetic Input types: int8, int16, float32, float16, int32 Float identical-shape inputs support arbitrary rank. Optimized broadcasting supports rank <= 4; higher-rank FP32 uses Reference and FP16 is rejected. int32 follows the adapter reference path. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/mul.cc) ## NOT_EQUAL Family: Comparison Input types: bool, float32, int8, int16, int32, int64 Boolean output. Supported input types depend on the comparison; input shapes and broadcast rules must match. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/comparisons.cc) ## PACK Family: Data movement Input types: int8, int16, float32, float16, int32, int64 Equal input shapes, including scalars. Float APIs require CORE >= 7.33.0 and matching enabled feature. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/pack.cc) ## PAD Family: Data movement Input types: int8, int16, float32, float16, int32 FP16 requires four-dimensional tensors and the float feature. Padding values and output shape must match. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/pad.cc) ## PADV2 Family: Data movement Input types: int8, int16, float32, float16, int32 FP16 requires four-dimensional tensors and the float feature. Padding values and output shape must match. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/pad.cc) ## QUANTIZE Family: Quantization Input types: int8, int16, int32, uint8, float32 Supported input/output pairs are specific: this type list does not imply all pairings. Registration is supplied by kernels/quantize.cc. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/quantize_common.cc) ## REDUCE_ALL Family: Reduction Input types: bool Boolean reduction using Reference in reduce_common.cc. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/reduce.cc) ## REDUCE_MAX Family: Reduction Input types: int8, int16, float32 No FP16 execution path. Axis, output shape and quantization checks apply; implementation is in reduce_common.cc. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/reduce.cc) ## REDUCE_MIN Family: Reduction Input types: int8, int16, float32 No FP16 execution path. Axis, output shape and quantization checks apply; implementation is in reduce_common.cc. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/reduce.cc) ## RELU Family: Activation Input types: int8, int16, float32, float16 Check activation bounds, tensor shape and quantization. Optimized float paths require the matching CORE feature. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/activations.cc) ## RELU6 Family: Activation Input types: int8, int16, float32, float16 Check activation bounds, tensor shape and quantization. Optimized float paths require the matching CORE feature. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/activations.cc) ## RESHAPE Family: Data movement Input types: int8, int16, float32, float16, int32, int64, uint8, bool Preserves storage and element count. FP16 storage works without float arithmetic enabled. Types listed here are representative fixed-width storage types. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/reshape.cc) ## RSQRT Family: Arithmetic Input types: int8, int16, float32, float16 FP32 and integer paths use Reference. FP16 requires CORE >= 7.33.0 and ARM_NN_ENABLE_F16; input/output shapes and types must match. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/elementwise.cc) ## SIN Family: Arithmetic Input types: float32 Reference implementation in the HELIA adapter source; Float32 input/output. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/elementwise.cc) ## SOFTMAX Family: Activation Input types: int8, int16, float32, float16 Optimized float path requires beta = 1. Other beta values use FP32 Reference or fail FP16 preparation. int8-to-int16 output has a dedicated registration. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/softmax.cc) ## SPLIT Family: Data movement Input types: int8, int16, float32, float16, int32 Constant axis and equal output extents. Float APIs require CORE >= 7.33.0 and corresponding enabled feature. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/split.cc) ## SPLIT_V Family: Data movement Input types: int8, int16, float32, float16, int32 Constant axis and split sizes; one inferred -1 size and zero-length pieces are allowed. Shapes/types/counts are checked during preparation. Float APIs require CORE >= 7.33.0. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/split_v.cc) ## SQRT Family: Arithmetic Input types: float32, float16 FP32 uses Reference. FP16 requires CORE >= 7.33.0 and ARM_NN_ENABLE_F16; input/output shapes and types must match. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/elementwise.cc) ## SQUARE Family: Arithmetic Input types: float32 Reference implementation in the HELIA adapter source; Float32 input/output. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/elementwise.cc) ## SQUEEZE Family: Data movement Input types: int8, int16, float32, float16, int32, int64, uint8, bool Bitwise storage copy; removes unit dimensions, checks matching byte counts and rejects strings. Input rank <= 8. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/squeeze.cc) ## STRIDED_SLICE Family: Data movement Input types: int8, int16, int32, float32, bool Check slicing masks, strides and shape constraints in the adapter; no FP16 execution branch. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/strided_slice.cc) ## SUB Family: Arithmetic Input types: int8, int16, float32, float16, int32 Float identical-shape inputs support arbitrary rank. Optimized broadcasting supports rank <= 4; higher-rank FP32 uses Reference and FP16 is rejected. int32 follows the adapter reference path. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/sub.cc) ## SUM Family: Reduction Input types: int8, int16, float32, float16 SUM is LiteRT REDUCE_SUM. Float optimized paths support rank <= 4 and axis sets; higher-rank FP32 uses Reference, FP16 is rejected. See reduce_common.cc. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/reduce.cc) ## SVDF Family: Recurrent Input types: int8, float32, float16 Data input is int8, FP32 or enabled FP16. int16 is a time-weight/state type, not a supported data input type. Variable state persists across invocations; weight types, rank and activation configuration are constrained. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/svdf.cc) ## TANH Family: Activation Input types: int8, int16, float32, float16 NaN is not supported by the optimized float activation path. Float support is build-gated. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/tanh.cc) ## TRANSPOSE Family: Data movement Input types: int8, int16, float32, float16 Optimized rank <= 4. Higher ranks use Reference; FP16 has a bitwise storage fallback even with its float feature disabled. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/transpose.cc) ## TRANSPOSE_CONV Family: Compute Input types: int8, int16, float32, float16 Check output shape, padding and quantized weight/bias combinations. Float features are build-gated. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/transpose_conv.cc) ## UNIDIRECTIONAL_SEQUENCE_LSTM Family: Recurrent Input types: int8, int16, float32, float16 Standard four-gate LSTM only. Preparation rejects peephole, projection, layer normalization and missing gates (CIFG), including FP32. Quantized inputs use int8 weights and int16 cell state. Preserve variable state across invocations. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/unidirectional_sequence_lstm.cc) ## UNPACK Family: Data movement Input types: int8, int16, float32, float16, int32 One output per selected axis element. Float APIs require CORE >= 7.33.0 and matching enabled feature. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/unpack.cc) ## ZEROS_LIKE Family: Data movement Input types: int8, int16, int32, int64, float32 Fills output storage with zero; no FP16 execution branch. [Adapter source](https://github.com/AmbiqAI/helia-rt/blob/70611e608778a432566e30965e8b4bc7ea75aa36/tensorflow/lite/micro/kernels/helia/zeros_like.cc)