Skip to main content

make_quantized

family

Declared in clika_rt/clika_rt_core.h.

Members​

NameDescription
make_quantizedClikaRT::make_quantized <- ClikaRT/compute/q_tensor.h consumes payload (contract §6) optional; NULL = absent; consumes global_scale when present (contract §6) C++ defaults (dropped at this surface): global_scale={}
make_quantized_affineClikaRT::make_quantized_affine <- ClikaRT/compute/q_tensor.h consumes codes (contract §6) consumes scales (contract §6) consumes biases (contract §6)
make_quantized_awqClikaRT::make_quantized_awq <- ClikaRT/compute/q_tensor.h consumes qweight (contract §6) consumes qzeros (contract §6) consumes scales (contract §6)
make_quantized_bnb4ClikaRT::make_quantized_bnb4 <- ClikaRT/compute/q_tensor.h consumes packed (contract §6) consumes absmax (contract §6) optional; NULL = absent; consumes nested_absmax when present (contract §6) optional; NULL = absent; consumes nested_quant_map when present (contract §6) optional; NULL = absent; consumes offset when present (contract §6) optional; NULL = absent; consumes quant_map when present (contract §6)
make_quantized_ct_packClikaRT::make_quantized_ct_pack <- ClikaRT/compute/q_tensor.h consumes packed (contract §6) consumes scales (contract §6)
make_quantized_fp8ClikaRT::make_quantized_fp8 <- ClikaRT/compute/q_tensor.h consumes codes (contract §6) consumes scale (contract §6)
make_quantized_fp8_blockedClikaRT::make_quantized_fp8_blocked <- ClikaRT/compute/q_tensor.h consumes codes (contract §6) consumes scale (contract §6)
make_quantized_gptqClikaRT::make_quantized_gptq <- ClikaRT/compute/q_tensor.h consumes qweight (contract §6) consumes qzeros (contract §6) consumes scales (contract §6) optional; NULL = absent; consumes g_idx when present (contract §6)
make_quantized_mxfp4ClikaRT::make_quantized_mxfp4 <- ClikaRT/compute/q_tensor.h consumes blocks (contract §6) consumes scales (contract §6)
make_quantized_nvfp4ClikaRT::make_quantized_nvfp4 <- ClikaRT/compute/q_tensor.h consumes codes (contract §6) consumes sub_scales (contract §6) optional; NULL = absent; consumes global_scale when present (contract §6)

Functions​

make_quantized​

clika_rt_error*(* make_quantized)(clika_rt_tensor *payload, const char *scheme, size_t scheme_len, const int64_t *logical_shape, size_t logical_shape_count, clika_rt_tensor *global_scale, clika_rt_qtensor **out)

ClikaRT::make_quantized <- ClikaRT/compute/q_tensor.h consumes payload (contract §6) optional; NULL = absent; consumes global_scale when present (contract §6) C++ defaults (dropped at this surface): global_scale={}

Declared in clika_rt/clika_rt_core.h, line 5946

make_quantized_affine​

clika_rt_error*(* make_quantized_affine)(clika_rt_tensor *codes, clika_rt_tensor *scales, clika_rt_tensor *biases, int64_t group_size, int64_t bits, const int64_t *logical_shape, size_t logical_shape_count, clika_rt_qtensor **out)

ClikaRT::make_quantized_affine <- ClikaRT/compute/q_tensor.h consumes codes (contract §6) consumes scales (contract §6) consumes biases (contract §6)

Declared in clika_rt/clika_rt_core.h, line 5997

make_quantized_awq​

clika_rt_error*(* make_quantized_awq)(clika_rt_tensor *qweight, clika_rt_tensor *qzeros, clika_rt_tensor *scales, int64_t bits, int64_t group_size, clika_rt_qtensor **out)

ClikaRT::make_quantized_awq <- ClikaRT/compute/q_tensor.h consumes qweight (contract §6) consumes qzeros (contract §6) consumes scales (contract §6)

Declared in clika_rt/clika_rt_core.h, line 6051

make_quantized_bnb4​

clika_rt_error*(* make_quantized_bnb4)(clika_rt_tensor *packed, clika_rt_tensor *absmax, clika_rt_tensor *nested_absmax, clika_rt_tensor *nested_quant_map, clika_rt_tensor *offset, clika_rt_tensor *quant_map, const char *quant_type, size_t quant_type_len, int64_t blocksize, int64_t nested_blocksize, const int64_t *shape, size_t shape_count, clika_rt_qtensor **out)

ClikaRT::make_quantized_bnb4 <- ClikaRT/compute/q_tensor.h consumes packed (contract §6) consumes absmax (contract §6) optional; NULL = absent; consumes nested_absmax when present (contract §6) optional; NULL = absent; consumes nested_quant_map when present (contract §6) optional; NULL = absent; consumes offset when present (contract §6) optional; NULL = absent; consumes quant_map when present (contract §6)

Declared in clika_rt/clika_rt_core.h, line 6033

make_quantized_ct_pack​

clika_rt_error*(* make_quantized_ct_pack)(clika_rt_tensor *packed, clika_rt_tensor *scales, int64_t bits, int64_t group_size, clika_rt_qtensor **out)

ClikaRT::make_quantized_ct_pack <- ClikaRT/compute/q_tensor.h consumes packed (contract §6) consumes scales (contract §6)

Declared in clika_rt/clika_rt_core.h, line 6061

make_quantized_fp8​

clika_rt_error*(* make_quantized_fp8)(clika_rt_tensor *codes, clika_rt_tensor *scale, const int64_t *logical_shape, size_t logical_shape_count, clika_rt_qtensor **out)

ClikaRT::make_quantized_fp8 <- ClikaRT/compute/q_tensor.h consumes codes (contract §6) consumes scale (contract §6)

Declared in clika_rt/clika_rt_core.h, line 5977

make_quantized_fp8_blocked​

clika_rt_error*(* make_quantized_fp8_blocked)(clika_rt_tensor *codes, clika_rt_tensor *scale, int64_t block_size, const int64_t *logical_shape, size_t logical_shape_count, clika_rt_qtensor **out)

ClikaRT::make_quantized_fp8_blocked <- ClikaRT/compute/q_tensor.h consumes codes (contract §6) consumes scale (contract §6)

Declared in clika_rt/clika_rt_core.h, line 5986

make_quantized_gptq​

clika_rt_error*(* make_quantized_gptq)(clika_rt_tensor *qweight, clika_rt_tensor *qzeros, clika_rt_tensor *scales, clika_rt_tensor *g_idx, int64_t bits, int64_t group_size, clika_rt_zeros_convention zeros_convention, clika_rt_qtensor **out)

ClikaRT::make_quantized_gptq <- ClikaRT/compute/q_tensor.h consumes qweight (contract §6) consumes qzeros (contract §6) consumes scales (contract §6) optional; NULL = absent; consumes g_idx when present (contract §6)

Declared in clika_rt/clika_rt_core.h, line 6017

make_quantized_mxfp4​

clika_rt_error*(* make_quantized_mxfp4)(clika_rt_tensor *blocks, clika_rt_tensor *scales, const int64_t *logical_shape, size_t logical_shape_count, clika_rt_qtensor **out)

ClikaRT::make_quantized_mxfp4 <- ClikaRT/compute/q_tensor.h consumes blocks (contract §6) consumes scales (contract §6)

Declared in clika_rt/clika_rt_core.h, line 5957

make_quantized_nvfp4​

clika_rt_error*(* make_quantized_nvfp4)(clika_rt_tensor *codes, clika_rt_tensor *sub_scales, clika_rt_tensor *global_scale, const int64_t *logical_shape, size_t logical_shape_count, clika_rt_qtensor **out)

ClikaRT::make_quantized_nvfp4 <- ClikaRT/compute/q_tensor.h consumes codes (contract §6) consumes sub_scales (contract §6) optional; NULL = absent; consumes global_scale when present (contract §6)

Declared in clika_rt/clika_rt_core.h, line 5967