npm - @fugood/llama.node - Versions diffs - 0.3.16 → 0.4.0 - Mend

@fugood/llama.node 0.3.16 → 0.4.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (281) hide show

package/CMakeLists.txt +6 -1
package/bin/darwin/arm64/llama-node.node +0 -0
package/bin/darwin/x64/llama-node.node +0 -0
package/bin/linux/arm64/llama-node.node +0 -0
package/bin/linux/x64/llama-node.node +0 -0
package/bin/linux-cuda/arm64/llama-node.node +0 -0
package/bin/linux-cuda/x64/llama-node.node +0 -0
package/bin/linux-vulkan/arm64/llama-node.node +0 -0
package/bin/linux-vulkan/x64/llama-node.node +0 -0
package/bin/win32/arm64/llama-node.node +0 -0
package/bin/win32/arm64/node.lib +0 -0
package/bin/win32/x64/llama-node.node +0 -0
package/bin/win32/x64/node.lib +0 -0
package/bin/win32-vulkan/arm64/llama-node.node +0 -0
package/bin/win32-vulkan/arm64/node.lib +0 -0
package/bin/win32-vulkan/x64/llama-node.node +0 -0
package/bin/win32-vulkan/x64/node.lib +0 -0
package/lib/binding.ts +44 -2
package/lib/index.js +132 -1
package/lib/index.ts +203 -3
package/package.json +2 -1
package/src/EmbeddingWorker.cpp +1 -1
package/src/LlamaCompletionWorker.cpp +374 -19
package/src/LlamaCompletionWorker.h +31 -10
package/src/LlamaContext.cpp +216 -7
package/src/LlamaContext.h +12 -0
package/src/common.hpp +15 -0
package/src/llama.cpp/.github/workflows/build-linux-cross.yml +233 -0
package/src/llama.cpp/.github/workflows/build.yml +89 -767
package/src/llama.cpp/.github/workflows/docker.yml +9 -6
package/src/llama.cpp/.github/workflows/release.yml +716 -0
package/src/llama.cpp/.github/workflows/server.yml +19 -23
package/src/llama.cpp/CMakeLists.txt +11 -1
package/src/llama.cpp/cmake/build-info.cmake +8 -2
package/src/llama.cpp/cmake/x64-windows-llvm.cmake +0 -6
package/src/llama.cpp/common/CMakeLists.txt +35 -4
package/src/llama.cpp/common/arg.cpp +844 -121
package/src/llama.cpp/common/arg.h +9 -0
package/src/llama.cpp/common/chat.cpp +129 -107
package/src/llama.cpp/common/chat.h +2 -0
package/src/llama.cpp/common/common.cpp +64 -518
package/src/llama.cpp/common/common.h +35 -45
package/src/llama.cpp/common/json-schema-to-grammar.cpp +3 -0
package/src/llama.cpp/common/llguidance.cpp +31 -47
package/src/llama.cpp/common/minja/chat-template.hpp +23 -11
package/src/llama.cpp/common/minja/minja.hpp +186 -127
package/src/llama.cpp/common/regex-partial.cpp +204 -0
package/src/llama.cpp/common/regex-partial.h +56 -0
package/src/llama.cpp/common/sampling.cpp +60 -50
package/src/llama.cpp/docs/build.md +122 -7
package/src/llama.cpp/examples/CMakeLists.txt +2 -32
package/src/llama.cpp/examples/batched/batched.cpp +1 -1
package/src/llama.cpp/examples/embedding/embedding.cpp +9 -12
package/src/llama.cpp/examples/gritlm/gritlm.cpp +1 -1
package/src/llama.cpp/examples/llama.android/llama/build.gradle.kts +1 -0
package/src/llama.cpp/examples/parallel/parallel.cpp +89 -15
package/src/llama.cpp/examples/passkey/passkey.cpp +1 -1
package/src/llama.cpp/examples/speculative/speculative.cpp +1 -1
package/src/llama.cpp/examples/speculative-simple/speculative-simple.cpp +1 -1
package/src/llama.cpp/examples/sycl/build.sh +2 -2
package/src/llama.cpp/examples/sycl/win-build-sycl.bat +2 -2
package/src/llama.cpp/examples/training/CMakeLists.txt +5 -0
package/src/llama.cpp/examples/training/finetune.cpp +96 -0
package/src/llama.cpp/ggml/CMakeLists.txt +35 -2
package/src/llama.cpp/ggml/cmake/GitVars.cmake +22 -0
package/src/llama.cpp/ggml/include/ggml-backend.h +4 -4
package/src/llama.cpp/ggml/include/ggml-cpp.h +1 -1
package/src/llama.cpp/ggml/include/ggml-cpu.h +5 -0
package/src/llama.cpp/ggml/include/ggml-opt.h +47 -28
package/src/llama.cpp/ggml/include/ggml-rpc.h +6 -1
package/src/llama.cpp/ggml/include/ggml.h +76 -106
package/src/llama.cpp/ggml/src/CMakeLists.txt +11 -8
package/src/llama.cpp/ggml/src/ggml-alloc.c +4 -1
package/src/llama.cpp/ggml/src/ggml-backend.cpp +9 -5
package/src/llama.cpp/ggml/src/ggml-cann/CMakeLists.txt +0 -2
package/src/llama.cpp/ggml/src/ggml-cann/acl_tensor.cpp +8 -4
package/src/llama.cpp/ggml/src/ggml-cann/acl_tensor.h +5 -5
package/src/llama.cpp/ggml/src/ggml-cann/aclnn_ops.cpp +692 -1534
package/src/llama.cpp/ggml/src/ggml-cann/aclnn_ops.h +613 -122
package/src/llama.cpp/ggml/src/ggml-cann/common.h +135 -1
package/src/llama.cpp/ggml/src/ggml-cann/ggml-cann.cpp +507 -137
package/src/llama.cpp/ggml/src/ggml-common.h +12 -6
package/src/llama.cpp/ggml/src/ggml-cpu/CMakeLists.txt +66 -33
package/src/llama.cpp/ggml/src/ggml-cpu/binary-ops.cpp +158 -0
package/src/llama.cpp/ggml/src/ggml-cpu/binary-ops.h +16 -0
package/src/llama.cpp/ggml/src/ggml-cpu/common.h +72 -0
package/src/llama.cpp/ggml/src/ggml-cpu/cpu-feats-x86.cpp +1 -1
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-aarch64.cpp +896 -194
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-impl.h +2 -21
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-quants.c +1060 -410
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.c +1008 -13533
package/src/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.cpp +31 -16
package/src/llama.cpp/ggml/src/ggml-cpu/kleidiai/kernels.cpp +90 -12
package/src/llama.cpp/ggml/src/ggml-cpu/kleidiai/kernels.h +47 -13
package/src/llama.cpp/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +266 -72
package/src/llama.cpp/ggml/src/ggml-cpu/llamafile/sgemm.cpp +1034 -88
package/src/llama.cpp/ggml/src/ggml-cpu/ops.cpp +8796 -0
package/src/llama.cpp/ggml/src/ggml-cpu/ops.h +110 -0
package/src/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h +892 -0
package/src/llama.cpp/ggml/src/ggml-cpu/unary-ops.cpp +186 -0
package/src/llama.cpp/ggml/src/ggml-cpu/unary-ops.h +28 -0
package/src/llama.cpp/ggml/src/ggml-cpu/vec.cpp +252 -0
package/src/llama.cpp/ggml/src/ggml-cpu/vec.h +802 -0
package/src/llama.cpp/ggml/src/ggml-cuda/CMakeLists.txt +23 -4
package/src/llama.cpp/ggml/src/ggml-cuda/vendors/hip.h +7 -0
package/src/llama.cpp/ggml/src/ggml-cuda/vendors/musa.h +1 -0
package/src/llama.cpp/ggml/src/ggml-hip/CMakeLists.txt +0 -4
package/src/llama.cpp/ggml/src/ggml-impl.h +52 -18
package/src/llama.cpp/ggml/src/ggml-metal/ggml-metal-impl.h +106 -14
package/src/llama.cpp/ggml/src/ggml-opencl/CMakeLists.txt +67 -119
package/src/llama.cpp/ggml/src/ggml-opencl/ggml-opencl.cpp +1023 -262
package/src/llama.cpp/ggml/src/ggml-opt.cpp +368 -190
package/src/llama.cpp/ggml/src/ggml-quants.c +0 -6
package/src/llama.cpp/ggml/src/ggml-rpc/ggml-rpc.cpp +307 -40
package/src/llama.cpp/ggml/src/ggml-sycl/CMakeLists.txt +125 -45
package/src/llama.cpp/ggml/src/ggml-sycl/backend.hpp +10 -8
package/src/llama.cpp/ggml/src/ggml-sycl/binbcast.cpp +239 -0
package/src/llama.cpp/ggml/src/ggml-sycl/binbcast.hpp +39 -0
package/src/llama.cpp/ggml/src/ggml-sycl/common.cpp +0 -35
package/src/llama.cpp/ggml/src/ggml-sycl/common.hpp +9 -307
package/src/llama.cpp/ggml/src/ggml-sycl/convert.cpp +72 -25
package/src/llama.cpp/ggml/src/ggml-sycl/convert.hpp +14 -7
package/src/llama.cpp/ggml/src/ggml-sycl/dequantize.hpp +59 -21
package/src/llama.cpp/ggml/src/ggml-sycl/dmmv.cpp +7 -1
package/src/llama.cpp/ggml/src/ggml-sycl/dpct/helper.hpp +79 -90
package/src/llama.cpp/ggml/src/ggml-sycl/element_wise.cpp +944 -438
package/src/llama.cpp/ggml/src/ggml-sycl/element_wise.hpp +22 -23
package/src/llama.cpp/ggml/src/ggml-sycl/gemm.hpp +37 -8
package/src/llama.cpp/ggml/src/ggml-sycl/getrows.cpp +24 -20
package/src/llama.cpp/ggml/src/ggml-sycl/getrows.hpp +1 -4
package/src/llama.cpp/ggml/src/ggml-sycl/ggml-sycl.cpp +507 -411
package/src/llama.cpp/ggml/src/ggml-sycl/im2col.cpp +84 -74
package/src/llama.cpp/ggml/src/ggml-sycl/im2col.hpp +1 -3
package/src/llama.cpp/ggml/src/ggml-sycl/mmvq.cpp +185 -89
package/src/llama.cpp/ggml/src/ggml-sycl/norm.cpp +37 -49
package/src/llama.cpp/ggml/src/ggml-sycl/norm.hpp +7 -22
package/src/llama.cpp/ggml/src/ggml-sycl/outprod.cpp +4 -14
package/src/llama.cpp/ggml/src/ggml-sycl/quants.hpp +83 -0
package/src/llama.cpp/ggml/src/ggml-sycl/rope.cpp +204 -118
package/src/llama.cpp/ggml/src/ggml-sycl/rope.hpp +1 -3
package/src/llama.cpp/ggml/src/ggml-sycl/vecdotq.hpp +128 -53
package/src/llama.cpp/ggml/src/ggml-vulkan/CMakeLists.txt +83 -49
package/src/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp +1278 -282
package/src/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/CMakeLists.txt +32 -0
package/src/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +133 -30
package/src/llama.cpp/ggml/src/ggml.c +170 -265
package/src/llama.cpp/ggml/src/gguf.cpp +34 -33
package/src/llama.cpp/include/llama.h +82 -22
package/src/llama.cpp/models/ggml-vocab-llama4.gguf.inp +112 -0
package/src/llama.cpp/models/ggml-vocab-llama4.gguf.out +46 -0
package/src/llama.cpp/models/ggml-vocab-pixtral.gguf.inp +112 -0
package/src/llama.cpp/models/ggml-vocab-pixtral.gguf.out +46 -0
package/src/llama.cpp/requirements/requirements-all.txt +5 -3
package/src/llama.cpp/requirements/requirements-gguf_editor_gui.txt +3 -0
package/src/llama.cpp/scripts/xxd.cmake +1 -1
package/src/llama.cpp/src/CMakeLists.txt +4 -2
package/src/llama.cpp/src/llama-adapter.cpp +43 -1
package/src/llama.cpp/src/llama-arch.cpp +163 -17
package/src/llama.cpp/src/llama-arch.h +16 -0
package/src/llama.cpp/src/llama-batch.cpp +5 -1
package/src/llama.cpp/src/llama-batch.h +2 -1
package/src/llama.cpp/src/llama-chat.cpp +91 -16
package/src/llama.cpp/src/llama-chat.h +7 -2
package/src/llama.cpp/src/llama-context.cpp +479 -575
package/src/llama.cpp/src/llama-context.h +44 -33
package/src/llama.cpp/src/llama-cparams.h +1 -0
package/src/llama.cpp/src/llama-graph.cpp +209 -157
package/src/llama.cpp/src/llama-graph.h +38 -14
package/src/llama.cpp/src/llama-hparams.h +13 -0
package/src/llama.cpp/src/llama-kv-cache.cpp +1604 -543
package/src/llama.cpp/src/llama-kv-cache.h +283 -171
package/src/llama.cpp/src/llama-memory.h +12 -2
package/src/llama.cpp/src/llama-mmap.cpp +1 -1
package/src/llama.cpp/src/llama-model-loader.cpp +34 -20
package/src/llama.cpp/src/llama-model-loader.h +5 -3
package/src/llama.cpp/src/llama-model-saver.cpp +281 -0
package/src/llama.cpp/src/llama-model-saver.h +37 -0
package/src/llama.cpp/src/llama-model.cpp +1803 -330
package/src/llama.cpp/src/llama-model.h +21 -2
package/src/llama.cpp/src/llama-quant.cpp +33 -10
package/src/llama.cpp/src/llama-sampling.cpp +25 -7
package/src/llama.cpp/src/llama-vocab.cpp +86 -10
package/src/llama.cpp/src/llama-vocab.h +6 -0
package/src/llama.cpp/src/llama.cpp +15 -1
package/src/llama.cpp/tests/CMakeLists.txt +52 -31
package/src/llama.cpp/tests/test-arg-parser.cpp +51 -4
package/src/llama.cpp/tests/test-backend-ops.cpp +189 -90
package/src/llama.cpp/tests/test-chat-template.cpp +26 -6
package/src/llama.cpp/tests/test-chat.cpp +15 -3
package/src/llama.cpp/{examples/gbnf-validator/gbnf-validator.cpp → tests/test-gbnf-validator.cpp} +2 -2
package/src/llama.cpp/tests/test-grammar-integration.cpp +3 -2
package/src/llama.cpp/tests/test-grammar-llguidance.cpp +63 -2
package/src/llama.cpp/tests/test-grammar-parser.cpp +3 -1
package/src/llama.cpp/tests/test-json-schema-to-grammar.cpp +17 -1
package/src/llama.cpp/tests/test-llama-grammar.cpp +2 -1
package/src/llama.cpp/tests/test-mtmd-c-api.c +63 -0
package/src/llama.cpp/tests/test-opt.cpp +33 -21
package/src/llama.cpp/{examples/quantize-stats/quantize-stats.cpp → tests/test-quantize-stats.cpp} +3 -1
package/src/llama.cpp/tests/test-regex-partial.cpp +288 -0
package/src/llama.cpp/tests/test-sampling.cpp +1 -1
package/src/llama.cpp/tests/test-tokenizer-1-bpe.cpp +2 -1
package/src/llama.cpp/tests/test-tokenizer-1-spm.cpp +2 -1
package/src/llama.cpp/tools/CMakeLists.txt +39 -0
package/src/llama.cpp/{examples → tools}/batched-bench/batched-bench.cpp +3 -3
package/src/llama.cpp/{examples → tools}/export-lora/export-lora.cpp +1 -1
package/src/llama.cpp/{examples → tools}/gguf-split/gguf-split.cpp +15 -16
package/src/llama.cpp/{examples → tools}/imatrix/imatrix.cpp +11 -9
package/src/llama.cpp/{examples → tools}/llama-bench/llama-bench.cpp +623 -274
package/src/llama.cpp/{examples → tools}/main/main.cpp +22 -14
package/src/llama.cpp/tools/mtmd/CMakeLists.txt +47 -0
package/src/llama.cpp/tools/mtmd/clip-impl.h +365 -0
package/src/llama.cpp/tools/mtmd/clip.cpp +3646 -0
package/src/llama.cpp/tools/mtmd/clip.h +99 -0
package/src/llama.cpp/tools/mtmd/deprecation-warning.cpp +22 -0
package/src/llama.cpp/tools/mtmd/mtmd-cli.cpp +370 -0
package/src/llama.cpp/tools/mtmd/mtmd-helper.cpp +310 -0
package/src/llama.cpp/tools/mtmd/mtmd.cpp +678 -0
package/src/llama.cpp/tools/mtmd/mtmd.h +331 -0
package/src/llama.cpp/{examples → tools}/perplexity/perplexity.cpp +21 -5
package/src/llama.cpp/{examples → tools}/quantize/quantize.cpp +53 -3
package/src/llama.cpp/tools/rpc/CMakeLists.txt +4 -0
package/src/llama.cpp/tools/rpc/rpc-server.cpp +322 -0
package/src/llama.cpp/tools/run/CMakeLists.txt +16 -0
package/src/llama.cpp/{examples → tools}/run/run.cpp +30 -30
package/src/llama.cpp/{examples → tools}/server/CMakeLists.txt +2 -1
package/src/llama.cpp/{examples → tools}/server/httplib.h +313 -247
package/src/llama.cpp/{examples → tools}/server/server.cpp +529 -215
package/src/llama.cpp/{examples → tools}/server/utils.hpp +427 -6
package/src/llama.cpp/{examples → tools}/tts/tts.cpp +6 -9
package/src/llama.cpp/cmake/arm64-windows-msvc.cmake +0 -6
package/src/llama.cpp/examples/gbnf-validator/CMakeLists.txt +0 -5
package/src/llama.cpp/examples/infill/CMakeLists.txt +0 -5
package/src/llama.cpp/examples/infill/infill.cpp +0 -590
package/src/llama.cpp/examples/llava/CMakeLists.txt +0 -66
package/src/llama.cpp/examples/llava/android/build_64.sh +0 -8
package/src/llama.cpp/examples/llava/clip-quantize-cli.cpp +0 -59
package/src/llama.cpp/examples/llava/clip.cpp +0 -3206
package/src/llama.cpp/examples/llava/clip.h +0 -118
package/src/llama.cpp/examples/llava/gemma3-cli.cpp +0 -341
package/src/llama.cpp/examples/llava/llava-cli.cpp +0 -332
package/src/llama.cpp/examples/llava/llava.cpp +0 -574
package/src/llama.cpp/examples/llava/llava.h +0 -49
package/src/llama.cpp/examples/llava/minicpmv-cli.cpp +0 -354
package/src/llama.cpp/examples/llava/qwen2vl-cli.cpp +0 -584
package/src/llama.cpp/examples/quantize-stats/CMakeLists.txt +0 -6
package/src/llama.cpp/examples/rpc/CMakeLists.txt +0 -2
package/src/llama.cpp/examples/rpc/rpc-server.cpp +0 -171
package/src/llama.cpp/examples/run/CMakeLists.txt +0 -5
package/src/llama.cpp/ggml/src/ggml-cann/kernels/CMakeLists.txt +0 -30
package/src/llama.cpp/ggml/src/ggml-cann/kernels/ascendc_kernels.h +0 -19
package/src/llama.cpp/ggml/src/ggml-cann/kernels/dup.cpp +0 -234
package/src/llama.cpp/ggml/src/ggml-cann/kernels/get_row_f16.cpp +0 -197
package/src/llama.cpp/ggml/src/ggml-cann/kernels/get_row_f32.cpp +0 -190
package/src/llama.cpp/ggml/src/ggml-cann/kernels/get_row_q4_0.cpp +0 -204
package/src/llama.cpp/ggml/src/ggml-cann/kernels/get_row_q8_0.cpp +0 -191
package/src/llama.cpp/ggml/src/ggml-cann/kernels/quantize_f16_q8_0.cpp +0 -218
package/src/llama.cpp/ggml/src/ggml-cann/kernels/quantize_f32_q8_0.cpp +0 -216
package/src/llama.cpp/ggml/src/ggml-cann/kernels/quantize_float_to_q4_0.cpp +0 -295
/package/src/llama.cpp/{examples → tools}/batched-bench/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/cvector-generator/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/cvector-generator/completions.txt +0 -0
/package/src/llama.cpp/{examples → tools}/cvector-generator/cvector-generator.cpp +0 -0
/package/src/llama.cpp/{examples → tools}/cvector-generator/mean.hpp +0 -0
/package/src/llama.cpp/{examples → tools}/cvector-generator/negative.txt +0 -0
/package/src/llama.cpp/{examples → tools}/cvector-generator/pca.hpp +0 -0
/package/src/llama.cpp/{examples → tools}/cvector-generator/positive.txt +0 -0
/package/src/llama.cpp/{examples → tools}/export-lora/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/gguf-split/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/imatrix/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/llama-bench/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/main/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples/llava → tools/mtmd}/requirements.txt +0 -0
/package/src/llama.cpp/{examples → tools}/perplexity/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/quantize/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/run/linenoise.cpp/linenoise.cpp +0 -0
/package/src/llama.cpp/{examples → tools}/run/linenoise.cpp/linenoise.h +0 -0
/package/src/llama.cpp/{examples → tools}/server/bench/requirements.txt +0 -0
/package/src/llama.cpp/{examples → tools}/server/tests/requirements.txt +0 -0
/package/src/llama.cpp/{examples → tools}/tokenize/CMakeLists.txt +0 -0
/package/src/llama.cpp/{examples → tools}/tokenize/tokenize.cpp +0 -0
/package/src/llama.cpp/{examples → tools}/tts/CMakeLists.txt +0 -0

package/src/llama.cpp/ggml/src/ggml-sycl/element_wise.cpp CHANGED Viewed

@@ -1,4 +1,5 @@
 #include "common.hpp"
+#include "ggml.h"
 #include "element_wise.hpp"
 static void acc_f32(const float * x, const float * y, float * dst, const int ne,
@@ -20,10 +21,32 @@ static void acc_f32(const float * x, const float * y, float * dst, const int ne,
     }
 }
-static void gelu_f32(const float * x, float * dst, const int k,
+template<typename T>
+static void sgn(const T * x, T * dst, const int k, const sycl::nd_item<3> &item_ct1) {
+    for(auto i = item_ct1.get_global_id(2); i < (const size_t)k; i += item_ct1.get_global_range(2)) {
+        dst[i] = x[i] > static_cast<T>(0.f) ? static_cast<T>(1.f) : ((x[i] < static_cast<T>(0.f) ? static_cast<T>(-1.f) : static_cast<T>(0.f)));
+    }
+}
+template<typename T>
+static void abs_op(const T * x, T * dst, const int k, const sycl::nd_item<3> &item_ct1) {
+    for(auto i = item_ct1.get_global_id(2); i < (const size_t)k; i += item_ct1.get_global_range(2)) {
+        dst[i] = sycl::fabs(x[i]);
+    }
+}
+template<typename T>
+static void elu_op(const T * x, T * dst, const int k, const sycl::nd_item<3> &item_ct1) {
+    for(auto i = item_ct1.get_global_id(2); i < (const size_t)k; i += item_ct1.get_global_range(2)) {
+        dst[i] = (x[i] > static_cast<T>(0.f)) ? x[i] : sycl::expm1(x[i]);
+    }
+}
+template<typename T>
+static void gelu(const T * x, T * dst, const int k,
                      const sycl::nd_item<3> &item_ct1) {
-    const float GELU_COEF_A    = 0.044715f;
-    const float SQRT_2_OVER_PI = 0.79788456080286535587989211986876f;
+    const T GELU_COEF_A    = static_cast<T>(0.044715f);
+    const T SQRT_2_OVER_PI = static_cast<T>(0.79788456080286535587989211986876f);
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
@@ -32,12 +55,13 @@ static void gelu_f32(const float * x, float * dst, const int k,
     }
     float xi = x[i];
-    dst[i] = 0.5f * xi *
-             (1.0f +
-              sycl::tanh(SQRT_2_OVER_PI * xi * (1.0f + GELU_COEF_A * xi * xi)));
+    dst[i] = static_cast<T>(0.5f) * xi *
+             (static_cast<T>(1.0f) +
+              sycl::tanh(SQRT_2_OVER_PI * xi * (static_cast<T>(1.0f) + GELU_COEF_A * xi * xi)));
 }
-static void silu_f32(const float * x, float * dst, const int k,
+template<typename T>
+static void silu(const T * x, T * dst, const int k,
                      const sycl::nd_item<3> &item_ct1) {
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
@@ -45,10 +69,11 @@ static void silu_f32(const float * x, float * dst, const int k,
     if (i >= k) {
         return;
     }
-    dst[i] = x[i] / (1.0f + sycl::native::exp(-x[i]));
+    dst[i] = x[i] / (static_cast<T>(1.0f) + sycl::native::exp(-x[i]));
 }
-static void gelu_quick_f32(const float *x, float *dst, int k,
+template<typename T>
+static void gelu_quick(const T *x, T *dst, int k,
                            const sycl::nd_item<3> &item_ct1) {
     const float GELU_QUICK_COEF = -1.702f;
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
@@ -56,20 +81,22 @@ static void gelu_quick_f32(const float *x, float *dst, int k,
     if (i >= k) {
         return;
     }
-    dst[i] = x[i] * (1.0f / (1.0f + sycl::native::exp(GELU_QUICK_COEF * x[i])));
+    dst[i] = x[i] * (static_cast<T>(1.0f) / (static_cast<T>(1.0f) + sycl::native::exp(GELU_QUICK_COEF * x[i])));
 }
-static void tanh_f32(const float *x, float *dst, int k,
+template<typename T>
+static void tanh(const T *x, T *dst, int k,
                      const sycl::nd_item<3> &item_ct1) {
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
     if (i >= k) {
         return;
     }
-    dst[i] = sycl::tanh((float)(x[i]));
+    dst[i] = sycl::tanh((x[i]));
 }
-static void relu_f32(const float * x, float * dst, const int k,
+template<typename T>
+static void relu(const T * x, T * dst, const int k,
                      const sycl::nd_item<3> &item_ct1) {
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
@@ -77,10 +104,11 @@ static void relu_f32(const float * x, float * dst, const int k,
     if (i >= k) {
         return;
     }
-    dst[i] = sycl::fmax((float)(x[i]), (float)0);
+    dst[i] = sycl::fmax((x[i]), static_cast<T>(0));
 }
-static void sigmoid_f32(const float * x, float * dst, const int k,
+template<typename T>
+static void sigmoid(const T * x, T * dst, const int k,
                             const sycl::nd_item<3> &item_ct1) {
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
@@ -88,10 +116,11 @@ static void sigmoid_f32(const float * x, float * dst, const int k,
     if (i >= k) {
         return;
     }
-    dst[i] = 1.0f / (1.0f + sycl::native::exp(-x[i]));
+    dst[i] = 1.0f / (static_cast<T>(1.0f) + sycl::native::exp(-x[i]));
 }
-static void sqrt_f32(const float * x, float * dst, const int k,
+template<typename T>
+static void sqrt(const T * x, T * dst, const int k,
                             const sycl::nd_item<3> &item_ct1) {
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
@@ -102,7 +131,8 @@ static void sqrt_f32(const float * x, float * dst, const int k,
     dst[i] = sycl::sqrt(x[i]);
 }
-static void sin_f32(const float * x, float * dst, const int k,
+template<typename T>
+static void sin(const T * x, T * dst, const int k,
                             const sycl::nd_item<3> &item_ct1) {
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
@@ -113,7 +143,8 @@ static void sin_f32(const float * x, float * dst, const int k,
     dst[i] = sycl::sin(x[i]);
 }
-static void cos_f32(const float * x, float * dst, const int k,
+template<typename T>
+static void cos(const T * x, T * dst, const int k,
                             const sycl::nd_item<3> &item_ct1) {
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
@@ -124,7 +155,8 @@ static void cos_f32(const float * x, float * dst, const int k,
     dst[i] = sycl::cos(x[i]);
 }
-static void hardsigmoid_f32(const float * x, float * dst, const int k,
+template<typename T>
+static void hardsigmoid(const T * x, T * dst, const int k,
                             const sycl::nd_item<3> &item_ct1) {
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
@@ -132,10 +164,11 @@ static void hardsigmoid_f32(const float * x, float * dst, const int k,
     if (i >= k) {
         return;
     }
-    dst[i] = sycl::fmin(1.0f, sycl::fmax(0.0f, (x[i] + 3.0f) / 6.0f));
+    dst[i] = sycl::fmin(static_cast<T>(1.0f), sycl::fmax(static_cast<T>(0.0f), (x[i] + static_cast<T>(3.0f)) / static_cast<T>(6.0f)));
 }
-static void hardswish_f32(const float * x, float * dst, const int k,
+template<typename T>
+static void hardswish(const T * x, T * dst, const int k,
                           const sycl::nd_item<3> &item_ct1) {
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
@@ -143,10 +176,11 @@ static void hardswish_f32(const float * x, float * dst, const int k,
     if (i >= k) {
         return;
     }
-    dst[i] = x[i] * sycl::fmin(1.0f, sycl::fmax(0.0f, (x[i] + 3.0f) / 6.0f));
+    dst[i] = x[i] * sycl::fmin(static_cast<T>(1.0f), sycl::fmax(static_cast<T>(0.0f), (x[i] + static_cast<T>(3.0f)) / static_cast<T>(6.0f)));
 }
-static void exp_f32(const float * x, float * dst, const int k,
+template<typename T>
+static void exp(const T * x, T * dst, const int k,
                           const sycl::nd_item<3> &item_ct1) {
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
@@ -157,7 +191,8 @@ static void exp_f32(const float * x, float * dst, const int k,
     dst[i] = sycl::exp(x[i]);
 }
-static void log_f32(const float * x, float * dst, const int k,
+template<typename T>
+static void log(const T * x, T * dst, const int k,
                           const sycl::nd_item<3> &item_ct1) {
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
@@ -165,15 +200,16 @@ static void log_f32(const float * x, float * dst, const int k,
     if (i >= k) {
         return;
     }
-    float xi = x[i];
+    T xi = x[i];
     if (xi <= 0) {
-        dst[i] = -INFINITY;
+        dst[i] = neg_infinity<T>();
     } else {
         dst[i] = sycl::log(xi);
     }
 }
-static void neg_f32(const float * x, float * dst, const int k,
+template<typename T>
+static void neg(const T * x, T * dst, const int k,
                           const sycl::nd_item<3> &item_ct1) {
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
@@ -184,7 +220,8 @@ static void neg_f32(const float * x, float * dst, const int k,
     dst[i] = -x[i];
 }
-static void step_f32(const float * x, float * dst, const int k,
+template<typename T>
+static void step(const T * x, T * dst, const int k,
                           const sycl::nd_item<3> &item_ct1) {
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
@@ -192,21 +229,23 @@ static void step_f32(const float * x, float * dst, const int k,
     if (i >= k) {
         return;
     }
-    dst[i] = x[i] > 0.0f;
+    dst[i] = x[i] > static_cast<T>(0.0f);
 }
-static void leaky_relu_f32(const float *x, float *dst, const int k, const float negative_slope,
+template<typename T>
+static void leaky_relu(const T *x, T *dst, const int k, const float negative_slope,
                            const sycl::nd_item<3> &item_ct1) {
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
     if (i >= k) {
         return;
     }
-    dst[i] = sycl::fmax((float)(x[i]), (float)0) +
-             sycl::fmin((float)(x[i]), 0.0f) * negative_slope;
+    dst[i] = sycl::fmax((x[i]), static_cast<T>(0)) +
+             sycl::fmin((x[i]), static_cast<T>(0.0f)) * negative_slope;
 }
-static void sqr_f32(const float * x, float * dst, const int k,
+template<typename T>
+static void sqr(const T * x, T * dst, const int k,
                     const sycl::nd_item<3> &item_ct1) {
     const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
                   item_ct1.get_local_id(2);
@@ -217,7 +256,8 @@ static void sqr_f32(const float * x, float * dst, const int k,
     dst[i] = x[i] * x[i];
 }
-static void upscale_f32(const float  *x, float *dst, const int nb00, const int nb01,
+template<typename  T>
+static void upscale(const T  *x, T *dst, const int nb00, const int nb01,
                         const int nb02, const int nb03, const int ne10, const int ne11,
                         const int ne12, const int ne13, const float sf0, const float sf1,
                         const float sf2, const float sf3, const sycl::nd_item<1> &item_ct1) {
@@ -237,10 +277,11 @@ static void upscale_f32(const float  *x, float *dst, const int nb00, const int n
     int i02 = i12 / sf2;
     int i03 = i13 / sf3;
-    dst[index] = *(const float *)((const char *)x + i03 * nb03 + i02 * nb02 + i01 * nb01 + i00 * nb00);
+    dst[index] = *(const T *)((const char *)x + i03 * nb03 + i02 * nb02 + i01 * nb01 + i00 * nb00);
 }
-static void pad_f32(const float  *x, float *dst, const int ne0, const int ne00, const int ne01, const int ne02,
+template <typename T>
+static void pad(const T  *x, T *dst, const int ne0, const int ne00, const int ne01, const int ne02,
                     const sycl::nd_item<3> &item_ct1) {
     int nidx = item_ct1.get_local_id(2) +
                item_ct1.get_group(2) * item_ct1.get_local_range(2);
@@ -256,11 +297,23 @@ static void pad_f32(const float  *x, float *dst, const int ne0, const int ne00,
                          item_ct1.get_group(0) * ne00 * ne01;
             dst[offset_dst] = x[offset_src];
     } else {
-        dst[offset_dst] = 0.0f;
+        dst[offset_dst] = static_cast<T>(0.0f);
     }
 }
+template<typename T>
+static void clamp(const T * x, T * dst, const float min, const float max, const int k,
+                      const sycl::nd_item<3> &item_ct1) {
+    const int i = item_ct1.get_local_range(2) * item_ct1.get_group(2) +
+                  item_ct1.get_local_id(2);
+    if (i >= k) {
+        return;
+    }
+    dst[i] = x[i] < static_cast<T>(min) ? static_cast<T>(min) : (x[i] > static_cast<T>(max) ? static_cast<T>(max) : x[i]);
+}
 static void acc_f32_sycl(const float *x, const float *y, float *dst,
                          const int n_elements, const int ne10, const int ne11,
@@ -277,7 +330,8 @@ static void acc_f32_sycl(const float *x, const float *y, float *dst,
         });
 }
-static void gelu_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void gelu_sycl(const T *x, T *dst, const int k,
                           queue_ptr stream) {
     const int num_blocks = (k + SYCL_GELU_BLOCK_SIZE - 1) / SYCL_GELU_BLOCK_SIZE;
     stream->parallel_for(
@@ -285,11 +339,12 @@ static void gelu_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_GELU_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_GELU_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            gelu_f32(x, dst, k, item_ct1);
+            gelu(x, dst, k, item_ct1);
         });
 }
-static void silu_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void silu_sycl(const T *x, T *dst, const int k,
                           queue_ptr stream) {
     const int num_blocks = (k + SYCL_SILU_BLOCK_SIZE - 1) / SYCL_SILU_BLOCK_SIZE;
     stream->parallel_for(
@@ -297,11 +352,43 @@ static void silu_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_SILU_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_SILU_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            silu_f32(x, dst, k, item_ct1);
+            silu(x, dst, k, item_ct1);
         });
 }
-static void gelu_quick_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void sgn_sycl(const T * x, T * dst, const int k, queue_ptr stream) {
+    // hard code for now
+    const int num_blocks = ceil_div(k, 256);
+    stream->parallel_for(
+            sycl::nd_range<3>((sycl::range<3>(1, 1, num_blocks) * sycl::range(1, 1, 256)), sycl::range(1, 1, 256)), [=](sycl::nd_item<3> item_ct1) {
+            sgn(x, dst, k, item_ct1);
+            });
+}
+template<typename T>
+static void abs_sycl(const T * x, T * dst, const int k, queue_ptr stream) {
+    // hard code for now
+    const int num_blocks = ceil_div(k, 256);
+    stream->parallel_for(
+            sycl::nd_range<3>((sycl::range<3>(1, 1, num_blocks) * sycl::range<3>(1, 1, 256)), sycl::range<3>(1, 1, 256)), [=](sycl::nd_item<3> item_ct1) {
+            abs_op(x, dst, k, item_ct1);
+            });
+}
+template<typename T>
+static void elu_sycl(const T * x, T * dst, const int k, queue_ptr stream) {
+    // hard code for now
+    const int num_blocks = ceil_div(k, 256);
+    stream->parallel_for(
+            sycl::nd_range<3>((sycl::range<3>(1, 1, num_blocks) * sycl::range<3>(1, 1, 256)), sycl::range<3>(1, 1, 256)), [=](sycl::nd_item<3> item_ct1) {
+            elu_op(x, dst, k, item_ct1);
+            });
+}
+template<typename T>
+static void gelu_quick_sycl(const T *x, T *dst, const int k,
                                 queue_ptr stream) {
     const int num_blocks = (k + SYCL_GELU_BLOCK_SIZE - 1) / SYCL_GELU_BLOCK_SIZE;
     stream->parallel_for(
@@ -309,11 +396,12 @@ static void gelu_quick_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_GELU_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_GELU_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            gelu_quick_f32(x, dst, k, item_ct1);
+            gelu_quick(x, dst, k, item_ct1);
         });
 }
-static void tanh_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void tanh_sycl(const T *x, T *dst, const int k,
                           queue_ptr stream) {
     const int num_blocks = (k + SYCL_TANH_BLOCK_SIZE - 1) / SYCL_TANH_BLOCK_SIZE;
     stream->parallel_for(
@@ -321,11 +409,12 @@ static void tanh_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_TANH_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_TANH_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            tanh_f32(x, dst, k, item_ct1);
+            tanh(x, dst, k, item_ct1);
         });
 }
-static void relu_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void relu_sycl(const T *x, T *dst, const int k,
                           queue_ptr stream) {
     const int num_blocks = (k + SYCL_RELU_BLOCK_SIZE - 1) / SYCL_RELU_BLOCK_SIZE;
     stream->parallel_for(
@@ -333,11 +422,12 @@ static void relu_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_RELU_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_RELU_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            relu_f32(x, dst, k, item_ct1);
+            relu(x, dst, k, item_ct1);
         });
 }
-static void hardsigmoid_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void hardsigmoid_sycl(const T *x, T *dst, const int k,
                                  queue_ptr stream) {
     const int num_blocks = (k + SYCL_HARDSIGMOID_BLOCK_SIZE - 1) / SYCL_HARDSIGMOID_BLOCK_SIZE;
     stream->parallel_for(
@@ -345,11 +435,12 @@ static void hardsigmoid_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_HARDSIGMOID_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_HARDSIGMOID_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            hardsigmoid_f32(x, dst, k, item_ct1);
+            hardsigmoid(x, dst, k, item_ct1);
         });
 }
-static void hardswish_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void hardswish_sycl(const T *x, T *dst, const int k,
                                queue_ptr stream) {
     const int num_blocks = (k + SYCL_HARDSWISH_BLOCK_SIZE - 1) / SYCL_HARDSWISH_BLOCK_SIZE;
     stream->parallel_for(
@@ -357,11 +448,12 @@ static void hardswish_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_HARDSWISH_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_HARDSWISH_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            hardswish_f32(x, dst, k, item_ct1);
+            hardswish(x, dst, k, item_ct1);
         });
 }
-static void exp_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void exp_sycl(const T *x, T *dst, const int k,
                                queue_ptr stream) {
     const int num_blocks = (k + SYCL_EXP_BLOCK_SIZE - 1) / SYCL_EXP_BLOCK_SIZE;
     stream->parallel_for(
@@ -369,11 +461,12 @@ static void exp_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_EXP_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_EXP_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            exp_f32(x, dst, k, item_ct1);
+            exp(x, dst, k, item_ct1);
         });
 }
-static void log_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void log_sycl(const T *x, T *dst, const int k,
                                queue_ptr stream) {
     const int num_blocks = (k + SYCL_EXP_BLOCK_SIZE - 1) / SYCL_EXP_BLOCK_SIZE;
     stream->parallel_for(
@@ -381,11 +474,12 @@ static void log_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_EXP_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_EXP_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            log_f32(x, dst, k, item_ct1);
+            log(x, dst, k, item_ct1);
         });
 }
-static void neg_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void neg_sycl(const T *x, T *dst, const int k,
                                queue_ptr stream) {
     const int num_blocks = (k + SYCL_NEG_BLOCK_SIZE - 1) / SYCL_NEG_BLOCK_SIZE;
     stream->parallel_for(
@@ -393,11 +487,12 @@ static void neg_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_NEG_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_NEG_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            neg_f32(x, dst, k, item_ct1);
+            neg(x, dst, k, item_ct1);
         });
 }
-static void step_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void step_sycl(const T *x, T *dst, const int k,
                                queue_ptr stream) {
     const int num_blocks = (k + SYCL_NEG_BLOCK_SIZE - 1) / SYCL_NEG_BLOCK_SIZE;
     stream->parallel_for(
@@ -405,11 +500,12 @@ static void step_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_NEG_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_NEG_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            step_f32(x, dst, k, item_ct1);
+            step(x, dst, k, item_ct1);
         });
 }
-static void sigmoid_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void sigmoid_sycl(const T *x, T *dst, const int k,
                                queue_ptr stream) {
     const int num_blocks = (k + SYCL_SIGMOID_BLOCK_SIZE - 1) / SYCL_SIGMOID_BLOCK_SIZE;
     stream->parallel_for(
@@ -417,11 +513,12 @@ static void sigmoid_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_SIGMOID_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_SIGMOID_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            sigmoid_f32(x, dst, k, item_ct1);
+            sigmoid(x, dst, k, item_ct1);
         });
 }
-static void sqrt_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void sqrt_sycl(const T *x, T *dst, const int k,
                                queue_ptr stream) {
     const int num_blocks = (k + SYCL_SQRT_BLOCK_SIZE - 1) / SYCL_SQRT_BLOCK_SIZE;
     stream->parallel_for(
@@ -429,11 +526,12 @@ static void sqrt_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_SQRT_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_SQRT_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            sqrt_f32(x, dst, k, item_ct1);
+            sqrt(x, dst, k, item_ct1);
         });
 }
-static void sin_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void sin_sycl(const T *x, T *dst, const int k,
                                queue_ptr stream) {
     const int num_blocks = (k + SYCL_SIN_BLOCK_SIZE - 1) / SYCL_SIN_BLOCK_SIZE;
     stream->parallel_for(
@@ -441,11 +539,12 @@ static void sin_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_SIN_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_SIN_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            sin_f32(x, dst, k, item_ct1);
+            sin(x, dst, k, item_ct1);
         });
 }
-static void cos_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void cos_sycl(const T *x, T *dst, const int k,
                                queue_ptr stream) {
     const int num_blocks = (k + SYCL_SIN_BLOCK_SIZE - 1) / SYCL_SIN_BLOCK_SIZE;
     stream->parallel_for(
@@ -453,11 +552,12 @@ static void cos_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_SIN_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_SIN_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            cos_f32(x, dst, k, item_ct1);
+            cos(x, dst, k, item_ct1);
         });
 }
-static void leaky_relu_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void leaky_relu_sycl(const T *x, T *dst, const int k,
                                 const float negative_slope,
                                 queue_ptr stream) {
     const int num_blocks = (k + SYCL_RELU_BLOCK_SIZE - 1) / SYCL_RELU_BLOCK_SIZE;
@@ -466,11 +566,12 @@ static void leaky_relu_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_RELU_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_RELU_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            leaky_relu_f32(x, dst, k, negative_slope, item_ct1);
+            leaky_relu(x, dst, k, negative_slope, item_ct1);
         });
 }
-static void sqr_f32_sycl(const float *x, float *dst, const int k,
+template<typename T>
+static void sqr_sycl(const T *x, T *dst, const int k,
                          queue_ptr stream) {
     const int num_blocks = (k + SYCL_SQR_BLOCK_SIZE - 1) / SYCL_SQR_BLOCK_SIZE;
     stream->parallel_for(
@@ -478,11 +579,12 @@ static void sqr_f32_sycl(const float *x, float *dst, const int k,
                               sycl::range<3>(1, 1, SYCL_SQR_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_SQR_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            sqr_f32(x, dst, k, item_ct1);
+            sqr(x, dst, k, item_ct1);
         });
 }
-static void upscale_f32_sycl(const float *x, float *dst, const int nb00, const int nb01,
+template<typename T>
+static void upscale_sycl(const T *x, T *dst, const int nb00, const int nb01,
                              const int nb02, const int nb03, const int ne10, const int ne11,
                              const int ne12, const int ne13, const float sf0, const float sf1,
                              const float sf2, const float sf3, queue_ptr stream) {
@@ -492,11 +594,12 @@ static void upscale_f32_sycl(const float *x, float *dst, const int nb00, const i
     stream->parallel_for(
         sycl::nd_range<1>(gridDim, sycl::range<1>(SYCL_UPSCALE_BLOCK_SIZE)),
         [=](sycl::nd_item<1> item_ct1) {
-            upscale_f32(x, dst, nb00, nb01, nb02, nb03, ne10, ne11, ne12, ne13, sf0, sf1, sf2, sf3, item_ct1);
+            upscale(x, dst, nb00, nb01, nb02, nb03, ne10, ne11, ne12, ne13, sf0, sf1, sf2, sf3, item_ct1);
         });
 }
-static void pad_f32_sycl(const float *x, float *dst, const int ne00,
+template<typename T>
+static void pad_sycl(const T *x, T *dst, const int ne00,
                          const int ne01, const int ne02, const int ne0,
                          const int ne1, const int ne2, queue_ptr stream) {
     int num_blocks = (ne0 + SYCL_PAD_BLOCK_SIZE - 1) / SYCL_PAD_BLOCK_SIZE;
@@ -505,526 +608,929 @@ static void pad_f32_sycl(const float *x, float *dst, const int ne00,
         sycl::nd_range<3>(gridDim * sycl::range<3>(1, 1, SYCL_PAD_BLOCK_SIZE),
                           sycl::range<3>(1, 1, SYCL_PAD_BLOCK_SIZE)),
         [=](sycl::nd_item<3> item_ct1) {
-            pad_f32(x, dst, ne0, ne00, ne01, ne02, item_ct1);
+            pad(x, dst, ne0, ne00, ne01, ne02, item_ct1);
         });
 }
-inline void ggml_sycl_op_silu(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
-                              ggml_tensor *dst, const float *src0_dd,
-                              const float *src1_dd, float *dst_dd,
-                              const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
-    silu_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+template<typename T>
+static void clamp_sycl(const T *x, T *dst, const float min,
+                           const float max, const int k,
+                           queue_ptr stream) {
+    const int num_blocks = (k + SYCL_CLAMP_BLOCK_SIZE - 1) / SYCL_CLAMP_BLOCK_SIZE;
+    stream->parallel_for(
+        sycl::nd_range<3>(sycl::range<3>(1, 1, num_blocks) *
+                              sycl::range<3>(1, 1, SYCL_CLAMP_BLOCK_SIZE),
+                          sycl::range<3>(1, 1, SYCL_CLAMP_BLOCK_SIZE)),
+        [=](sycl::nd_item<3> item_ct1) {
+            clamp(x, dst, min, max, k, item_ct1);
+        });
 }
-inline void ggml_sycl_op_gelu(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
-                              ggml_tensor *dst, const float *src0_dd,
-                              const float *src1_dd, float *dst_dd,
-                              const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
+inline void ggml_sycl_op_sgn(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
-    gelu_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                sgn_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                sgn_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_gelu_quick(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
-                                    const ggml_tensor *src1, ggml_tensor *dst,
-                                    const float *src0_dd, const float *src1_dd,
-                                    float *dst_dd,
-                                    const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
-    gelu_quick_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
+inline void ggml_sycl_op_abs(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                abs_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                abs_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_tanh(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
-                              ggml_tensor *dst, const float *src0_dd,
-                              const float *src1_dd, float *dst_dd,
-                              const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
-    tanh_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
+inline void ggml_sycl_op_elu(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                elu_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                elu_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_relu(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
-                              ggml_tensor *dst, const float *src0_dd,
-                              const float *src1_dd, float *dst_dd,
-                              const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
-    relu_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+inline void ggml_sycl_op_silu(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                silu_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                silu_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_hardsigmoid(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
-                                     const ggml_tensor *src1, ggml_tensor *dst,
-                                     const float *src0_dd, const float *src1_dd,
-                                     float *dst_dd,
-                                     const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
-    hardsigmoid_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+inline void ggml_sycl_op_gelu(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                gelu_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                gelu_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_hardswish(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
-                                   const ggml_tensor *src1, ggml_tensor *dst,
-                                   const float *src0_dd, const float *src1_dd,
-                                   float *dst_dd, const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
-    hardswish_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+inline void ggml_sycl_op_gelu_quick(ggml_backend_sycl_context & ctx, ggml_tensor *dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                gelu_quick_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                gelu_quick_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_exp(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
-                                   const ggml_tensor *src1, ggml_tensor *dst,
-                                   const float *src0_dd, const float *src1_dd,
-                                   float *dst_dd, const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
-    exp_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+inline void ggml_sycl_op_tanh(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                tanh_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                tanh_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_log(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
-                                   const ggml_tensor *src1, ggml_tensor *dst,
-                                   const float *src0_dd, const float *src1_dd,
-                                   float *dst_dd, const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
-    log_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+inline void ggml_sycl_op_relu(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                relu_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                relu_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_sigmoid(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
-                                   const ggml_tensor *src1, ggml_tensor *dst,
-                                   const float *src0_dd, const float *src1_dd,
-                                   float *dst_dd, const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
-    sigmoid_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+inline void ggml_sycl_op_hardsigmoid(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                hardsigmoid_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                hardsigmoid_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_sqrt(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
-                                   const ggml_tensor *src1, ggml_tensor *dst,
-                                   const float *src0_dd, const float *src1_dd,
-                                   float *dst_dd, const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
-    sqrt_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+inline void ggml_sycl_op_hardswish(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                hardswish_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                hardswish_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_sin(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
-                                   const ggml_tensor *src1, ggml_tensor *dst,
-                                   const float *src0_dd, const float *src1_dd,
-                                   float *dst_dd, const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
-    sin_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+inline void ggml_sycl_op_exp(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                exp_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                exp_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_cos(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
-                                   const ggml_tensor *src1, ggml_tensor *dst,
-                                   const float *src0_dd, const float *src1_dd,
-                                   float *dst_dd, const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
-    cos_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+inline void ggml_sycl_op_log(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                log_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                log_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_step(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
-                                   const ggml_tensor *src1, ggml_tensor *dst,
-                                   const float *src0_dd, const float *src1_dd,
-                                   float *dst_dd, const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
-    step_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+inline void ggml_sycl_op_sigmoid(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                sigmoid_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                sigmoid_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_neg(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
-                                   const ggml_tensor *src1, ggml_tensor *dst,
-                                   const float *src0_dd, const float *src1_dd,
-                                   float *dst_dd, const queue_ptr &main_stream) {
+inline void ggml_sycl_op_sqrt(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                sqrt_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                sqrt_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
+}
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
+inline void ggml_sycl_op_sin(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                sin_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                sin_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
+}
-    neg_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
+inline void ggml_sycl_op_cos(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                cos_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                cos_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
+}
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+inline void ggml_sycl_op_step(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                step_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                step_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_leaky_relu(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
-                                    const ggml_tensor *src1, ggml_tensor *dst,
-                                    const float *src0_dd, const float *src1_dd,
-                                    float *dst_dd,
-                                    const queue_ptr &main_stream) {
+inline void ggml_sycl_op_neg(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                neg_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                neg_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
+}
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
+inline void ggml_sycl_op_leaky_relu(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
     float negative_slope;
     memcpy(&negative_slope, dst->op_params, sizeof(float));
-    leaky_relu_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), negative_slope, main_stream);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                leaky_relu_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), negative_slope, main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                leaky_relu_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), negative_slope, main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_sqr(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
-                             ggml_tensor *dst, const float *src0_dd,
-                             const float *src1_dd, float *dst_dd,
-                             const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT( dst->type == GGML_TYPE_F32);
-    sqr_f32_sycl(src0_dd, dst_dd, ggml_nelements(src0), main_stream);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+inline void ggml_sycl_op_sqr(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+ #if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                sqr_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                sqr_sycl(data_pts.src, data_pts.dst, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_upscale(ggml_backend_sycl_context & ctx, const ggml_tensor *src0,
-                                 const ggml_tensor *src1, ggml_tensor *dst,
-                                 const float *src0_dd, const float *src1_dd,
-                                 float *dst_dd,
-                                 const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
+inline void ggml_sycl_op_upscale(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
     GGML_ASSERT(dst->type == GGML_TYPE_F32);
-    const float sf0 = (float)dst->ne[0]/src0->ne[0];
-    const float sf1 = (float)dst->ne[1]/src0->ne[1];
-    const float sf2 = (float)dst->ne[2]/src0->ne[2];
-    const float sf3 = (float)dst->ne[3]/src0->ne[3];
-    upscale_f32_sycl(src0_dd, dst_dd, src0->nb[0], src0->nb[1], src0->nb[2], src0->nb[3],
-                     dst->ne[0], dst->ne[1], dst->ne[2], dst->ne[3], sf0, sf1, sf2, sf3,
-                     main_stream);
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    const float sf0 = (float) dst->ne[0] / dst->src[0]->ne[0];
+    const float sf1 = (float) dst->ne[1] / dst->src[0]->ne[1];
+    const float sf2 = (float) dst->ne[2] / dst->src[0]->ne[2];
+    const float sf3 = (float) dst->ne[3] / dst->src[0]->ne[3];
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                upscale_sycl(data_pts.src, data_pts.dst, dst->src[0]->nb[0], dst->src[0]->nb[1], dst->src[0]->nb[2],
+                        dst->src[0]->nb[3], dst->ne[0], dst->ne[1], dst->ne[2], dst->ne[3], sf0, sf1, sf2, sf3,
+                        main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                upscale_sycl(data_pts.src, data_pts.dst, dst->src[0]->nb[0], dst->src[0]->nb[1], dst->src[0]->nb[2],
+                        dst->src[0]->nb[3], dst->ne[0], dst->ne[1], dst->ne[2], dst->ne[3], sf0, sf1, sf2, sf3,
+                        main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_pad(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
-                             ggml_tensor *dst, const float *src0_dd,
-                             const float *src1_dd, float *dst_dd,
-                             const queue_ptr &main_stream) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
+inline void ggml_sycl_op_pad(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined (GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
     GGML_ASSERT(dst->type == GGML_TYPE_F32);
-    GGML_ASSERT(src0->ne[3] == 1 && dst->ne[3] == 1); // just 3D tensors
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    GGML_ASSERT(dst->src[0]->ne[3] == 1 && dst->ne[3] == 1);  // just 3D tensors
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    switch (dst->type) {
+#if defined (GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                pad_sycl(data_pts.src, data_pts.dst, dst->src[0]->ne[0], dst->src[0]->ne[1], dst->src[0]->ne[2], dst->ne[0],
+                        dst->ne[1], dst->ne[2], main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                pad_sycl(data_pts.src, data_pts.dst, dst->src[0]->ne[0], dst->src[0]->ne[1], dst->src[0]->ne[2], dst->ne[0],
+                        dst->ne[1], dst->ne[2], main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
+}
-    pad_f32_sycl(src0_dd, dst_dd,
-        src0->ne[0], src0->ne[1], src0->ne[2],
-        dst->ne[0], dst->ne[1], dst->ne[2], main_stream);
+inline void ggml_sycl_op_clamp(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+#if defined(GGML_SYCL_F16)
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32 || dst->src[0]->type == GGML_TYPE_F16);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32 || dst->type == GGML_TYPE_F16);
+#else
-    GGML_UNUSED(src1);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(src1_dd);
-    GGML_UNUSED(ctx);
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->type == GGML_TYPE_F32);
+#endif
+    GGML_ASSERT(dst->src[0]->type == dst->type);
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    float min;
+    float max;
+    memcpy(&min, dst->op_params, sizeof(float));
+    memcpy(&max, (float *) dst->op_params + 1, sizeof(float));
+    switch (dst->type) {
+#if defined(GGML_SYCL_F16)
+        case GGML_TYPE_F16:
+            {
+                auto data_pts = cast_data<sycl::half>(dst);
+                clamp_sycl(data_pts.src, data_pts.dst, min, max, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+#endif
+        case GGML_TYPE_F32:
+            {
+                auto data_pts = cast_data<float>(dst);
+                clamp_sycl(data_pts.src, data_pts.dst, min, max, ggml_nelements(dst->src[0]), main_stream);
+                break;
+            }
+        default:
+            GGML_ABORT("GGML tensor type not supported!\n");
+    }
 }
-inline void ggml_sycl_op_acc(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
-                             ggml_tensor *dst, const float *src0_dd,
-                             const float *src1_dd, float *dst_dd,
-                             const queue_ptr &main_stream) {
+inline void ggml_sycl_op_acc(ggml_backend_sycl_context & ctx, ggml_tensor *dst) {
-    GGML_ASSERT(src0->type == GGML_TYPE_F32);
-    GGML_ASSERT(src1->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->src[0]->type == GGML_TYPE_F32);
+    GGML_ASSERT(dst->src[1]->type == GGML_TYPE_F32);
     GGML_ASSERT( dst->type == GGML_TYPE_F32);
     GGML_ASSERT(dst->ne[3] == 1); // just 3D tensors supported
+    dpct::queue_ptr main_stream = ctx.stream();
+    SYCL_CHECK(ggml_sycl_set_device(ctx.device));
+    const float * src0_dd = static_cast<const float *>(dst->src[0]->data);
+    const float * src1_dd = static_cast<const float*>(dst->src[1]->data);
+    float *       dst_dd  = static_cast<float *>(dst->data);
     int nb1 = dst->op_params[0] / 4; // 4 bytes of float32
     int nb2 = dst->op_params[1] / 4; // 4 bytes of float32
     // int nb3 = dst->op_params[2] / 4; // 4 bytes of float32 - unused
     int offset = dst->op_params[3] / 4; // offset in bytes
-    acc_f32_sycl(src0_dd, src1_dd, dst_dd, ggml_nelements(dst), src1->ne[0], src1->ne[1], src1->ne[2], nb1, nb2, offset, main_stream);
-    GGML_UNUSED(dst);
-    GGML_UNUSED(ctx);
-}
-inline void ggml_sycl_op_add(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
-                             ggml_tensor *dst, const float *src0_dd,
-                             const float *src1_dd, float *dst_dd,
-                             const queue_ptr &main_stream) {
-    ggml_sycl_op_bin_bcast<bin_bcast_sycl<op_add>>(ctx, src0, src1, dst, src0_dd, src1_dd, dst_dd, main_stream);
-}
-inline void ggml_sycl_op_sub(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
-                             ggml_tensor *dst, const float *src0_dd,
-                             const float *src1_dd, float *dst_dd,
-                             const queue_ptr &main_stream) {
-    ggml_sycl_op_bin_bcast<bin_bcast_sycl<op_sub>>(ctx, src0, src1, dst, src0_dd, src1_dd, dst_dd, main_stream);
-}
-inline void ggml_sycl_op_mul(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
-                             ggml_tensor *dst, const float *src0_dd,
-                             const float *src1_dd, float *dst_dd,
-                             const queue_ptr &main_stream) {
-    ggml_sycl_op_bin_bcast<bin_bcast_sycl<op_mul>>(ctx, src0, src1, dst, src0_dd, src1_dd, dst_dd, main_stream);
-}
-inline void ggml_sycl_op_div(ggml_backend_sycl_context & ctx, const ggml_tensor *src0, const ggml_tensor *src1,
-                             ggml_tensor *dst, const float *src0_dd,
-                             const float *src1_dd, float *dst_dd,
-                             const queue_ptr &main_stream) {
-    ggml_sycl_op_bin_bcast<bin_bcast_sycl<op_div>>(ctx, src0, src1, dst, src0_dd, src1_dd, dst_dd, main_stream);
+    acc_f32_sycl(src0_dd, src1_dd, dst_dd, ggml_nelements(dst), dst->src[1]->ne[0], dst->src[1]->ne[1], dst->src[1]->ne[2], nb1, nb2, offset, main_stream);
 }
 void ggml_sycl_sqrt(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_sqrt);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_sqrt(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_sin(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_sin);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_sin(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_cos(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_cos);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_cos(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_acc(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_acc);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_acc(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_gelu(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_gelu);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_gelu(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_silu(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_silu);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_silu(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_gelu_quick(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_gelu_quick);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_gelu_quick(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_tanh(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_tanh);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_tanh(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_relu(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_relu);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_relu(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_sigmoid(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_sigmoid);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_sigmoid(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_hardsigmoid(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_hardsigmoid);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_hardsigmoid(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_hardswish(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_hardswish);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_hardswish(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_exp(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_exp);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_exp(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_log(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_log);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_log(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_neg(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_neg);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_neg(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_step(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_step);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_step(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_leaky_relu(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_leaky_relu);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_leaky_relu(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_sqr(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_sqr);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_sqr(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_upscale(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_upscale);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_upscale(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
 void ggml_sycl_pad(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_pad);
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_pad(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
-void ggml_sycl_add(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_add);
+void ggml_sycl_clamp(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_clamp(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
-void ggml_sycl_sub(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_sub);
+void ggml_sycl_sgn(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_sgn(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
-void ggml_sycl_mul(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_mul);
+void ggml_sycl_abs(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_abs(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }
-void ggml_sycl_div(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
-    GGML_SYCL_DEBUG("call %s\n", __func__);
-    ggml_sycl_op_flatten(ctx, dst->src[0], dst->src[1], dst, ggml_sycl_op_div);
+void ggml_sycl_elu(ggml_backend_sycl_context & ctx, ggml_tensor * dst) {
+    GGML_SYCL_DEBUG("call %s: DST Tensor type: %s\n", __func__, ggml_type_name(dst->type));
+    ggml_sycl_op_elu(ctx, dst);
     GGML_SYCL_DEBUG("call %s done\n", __func__);
 }