@novastera-oss/llamarn 0.2.5 → 0.2.7

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 (225) hide show
  1. package/RNLlamaCpp.podspec +3 -2
  2. package/android/CMakeLists.txt +6 -3
  3. package/android/src/main/cpp/include/llama.h +140 -38
  4. package/android/src/main/jniLibs/arm64-v8a/libggml-base.so +0 -0
  5. package/android/src/main/jniLibs/arm64-v8a/libggml-cpu.so +0 -0
  6. package/android/src/main/jniLibs/arm64-v8a/libggml.so +0 -0
  7. package/android/src/main/jniLibs/arm64-v8a/libllama.so +0 -0
  8. package/android/src/main/jniLibs/x86_64/libggml-base.so +0 -0
  9. package/android/src/main/jniLibs/x86_64/libggml-cpu.so +0 -0
  10. package/android/src/main/jniLibs/x86_64/libggml.so +0 -0
  11. package/android/src/main/jniLibs/x86_64/libllama.so +0 -0
  12. package/cpp/LlamaCppModel.cpp +48 -67
  13. package/cpp/LlamaCppModel.h +8 -3
  14. package/cpp/PureCppImpl.cpp +1 -1
  15. package/cpp/PureCppImpl.h +2 -2
  16. package/cpp/build-info.cpp +2 -2
  17. package/cpp/llama.cpp/CMakeLists.txt +15 -4
  18. package/cpp/llama.cpp/Makefile +2 -2
  19. package/cpp/llama.cpp/README.md +33 -13
  20. package/cpp/llama.cpp/common/CMakeLists.txt +15 -28
  21. package/cpp/llama.cpp/common/arg.cpp +38 -12
  22. package/cpp/llama.cpp/common/build-info.cpp.in +2 -2
  23. package/cpp/llama.cpp/common/chat-parser.cpp +9 -3
  24. package/cpp/llama.cpp/common/chat-parser.h +4 -1
  25. package/cpp/llama.cpp/common/chat.cpp +16 -13
  26. package/cpp/llama.cpp/common/chat.h +1 -1
  27. package/cpp/llama.cpp/common/common.cpp +52 -40
  28. package/cpp/llama.cpp/common/common.h +5 -2
  29. package/cpp/llama.cpp/common/json-partial.cpp +5 -4
  30. package/cpp/llama.cpp/common/json-partial.h +2 -1
  31. package/cpp/llama.cpp/common/json-schema-to-grammar.cpp +2 -1
  32. package/cpp/llama.cpp/common/json-schema-to-grammar.h +4 -4
  33. package/cpp/llama.cpp/common/speculative.cpp +6 -4
  34. package/cpp/llama.cpp/convert_hf_to_gguf.py +128 -84
  35. package/cpp/llama.cpp/ggml/CMakeLists.txt +47 -2
  36. package/cpp/llama.cpp/ggml/cmake/common.cmake +1 -2
  37. package/cpp/llama.cpp/ggml/include/ggml.h +1 -3
  38. package/cpp/llama.cpp/ggml/src/CMakeLists.txt +49 -13
  39. package/cpp/llama.cpp/ggml/src/ggml-backend-reg.cpp +5 -0
  40. package/cpp/llama.cpp/ggml/src/ggml-backend.cpp +10 -5
  41. package/cpp/llama.cpp/ggml/src/ggml-blas/CMakeLists.txt +3 -3
  42. package/cpp/llama.cpp/ggml/src/ggml-cann/common.h +6 -1
  43. package/cpp/llama.cpp/ggml/src/ggml-cann/ggml-cann.cpp +33 -9
  44. package/cpp/llama.cpp/ggml/src/ggml-common.h +4 -0
  45. package/cpp/llama.cpp/ggml/src/ggml-cpu/CMakeLists.txt +93 -24
  46. package/cpp/llama.cpp/ggml/src/ggml-cpu/amx/amx.cpp +1 -1
  47. package/cpp/llama.cpp/ggml/src/ggml-cpu/amx/mmq.cpp +1 -1
  48. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/arm/cpu-feats.cpp +94 -0
  49. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/arm/quants.c +4113 -0
  50. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/arm/repack.cpp +2174 -0
  51. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/loongarch/quants.c +2638 -0
  52. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/powerpc/quants.c +2731 -0
  53. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/riscv/quants.c +2068 -0
  54. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/riscv/repack.cpp +396 -0
  55. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/s390/quants.c +1299 -0
  56. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/wasm/quants.c +1480 -0
  57. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/x86/quants.c +4310 -0
  58. package/cpp/llama.cpp/ggml/src/ggml-cpu/{ggml-cpu-aarch64.cpp → arch/x86/repack.cpp} +59 -3206
  59. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch-fallback.h +184 -0
  60. package/cpp/llama.cpp/ggml/src/ggml-cpu/common.h +1 -1
  61. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-impl.h +7 -4
  62. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.c +33 -2
  63. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.cpp +8 -8
  64. package/cpp/llama.cpp/ggml/src/ggml-cpu/{ggml-cpu-hbm.cpp → hbm.cpp} +1 -1
  65. package/cpp/llama.cpp/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +1 -1
  66. package/cpp/llama.cpp/ggml/src/ggml-cpu/llamafile/sgemm.cpp +56 -7
  67. package/cpp/llama.cpp/ggml/src/ggml-cpu/llamafile/sgemm.h +5 -0
  68. package/cpp/llama.cpp/ggml/src/ggml-cpu/ops.cpp +2 -2
  69. package/cpp/llama.cpp/ggml/src/ggml-cpu/quants.c +1157 -0
  70. package/cpp/llama.cpp/ggml/src/ggml-cpu/{ggml-cpu-quants.h → quants.h} +26 -0
  71. package/cpp/llama.cpp/ggml/src/ggml-cpu/repack.cpp +1555 -0
  72. package/cpp/llama.cpp/ggml/src/ggml-cpu/repack.h +98 -0
  73. package/cpp/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h +2 -4
  74. package/cpp/llama.cpp/ggml/src/ggml-cpu/{ggml-cpu-traits.cpp → traits.cpp} +1 -1
  75. package/cpp/llama.cpp/ggml/src/ggml-cuda/common.cuh +6 -8
  76. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-mma-f16.cuh +5 -2
  77. package/cpp/llama.cpp/ggml/src/ggml-cuda/ggml-cuda.cu +25 -16
  78. package/cpp/llama.cpp/ggml/src/ggml-cuda/ssm-scan.cu +6 -4
  79. package/cpp/llama.cpp/ggml/src/ggml-hip/CMakeLists.txt +4 -0
  80. package/cpp/llama.cpp/ggml/src/ggml-impl.h +2 -0
  81. package/cpp/llama.cpp/ggml/src/ggml-metal/CMakeLists.txt +11 -10
  82. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal.m +33 -8
  83. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal.metal +135 -100
  84. package/cpp/llama.cpp/ggml/src/ggml-opencl/CMakeLists.txt +7 -0
  85. package/cpp/llama.cpp/ggml/src/ggml-opencl/ggml-opencl.cpp +908 -3
  86. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/concat.cl +109 -0
  87. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mv_id_q4_0_f32_8x_flat.cl +283 -0
  88. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/pad.cl +30 -0
  89. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/repeat.cl +39 -0
  90. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/tanh.cl +63 -0
  91. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/tsembd.cl +48 -0
  92. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/upscale.cl +121 -0
  93. package/cpp/llama.cpp/ggml/src/ggml-quants.c +0 -2
  94. package/cpp/llama.cpp/ggml/src/ggml-rpc/ggml-rpc.cpp +18 -15
  95. package/cpp/llama.cpp/ggml/src/ggml-sycl/CMakeLists.txt +3 -3
  96. package/cpp/llama.cpp/ggml/src/ggml-sycl/common.hpp +19 -24
  97. package/cpp/llama.cpp/ggml/src/ggml-sycl/convert.cpp +21 -2
  98. package/cpp/llama.cpp/ggml/src/ggml-sycl/cpy.cpp +121 -4
  99. package/cpp/llama.cpp/ggml/src/ggml-sycl/dequantize.hpp +32 -0
  100. package/cpp/llama.cpp/ggml/src/ggml-sycl/gemm.hpp +3 -0
  101. package/cpp/llama.cpp/ggml/src/ggml-sycl/getrows.cpp +2 -96
  102. package/cpp/llama.cpp/ggml/src/ggml-sycl/ggml-sycl.cpp +164 -46
  103. package/cpp/llama.cpp/ggml/src/ggml-sycl/mmvq.cpp +32 -8
  104. package/cpp/llama.cpp/ggml/src/ggml-sycl/quants.hpp +38 -10
  105. package/cpp/llama.cpp/ggml/src/ggml-sycl/rope.cpp +118 -11
  106. package/cpp/llama.cpp/ggml/src/ggml-sycl/vecdotq.hpp +108 -16
  107. package/cpp/llama.cpp/ggml/src/ggml-vulkan/CMakeLists.txt +26 -29
  108. package/cpp/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp +432 -248
  109. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/CMakeLists.txt +0 -12
  110. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/conv_transpose_1d.comp +98 -0
  111. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +2 -0
  112. package/cpp/llama.cpp/ggml/src/ggml.c +9 -8
  113. package/cpp/llama.cpp/ggml/src/ggml.cpp +26 -0
  114. package/cpp/llama.cpp/ggml/src/gguf.cpp +19 -2
  115. package/cpp/llama.cpp/gguf-py/gguf/constants.py +57 -0
  116. package/cpp/llama.cpp/gguf-py/gguf/gguf_writer.py +4 -1
  117. package/cpp/llama.cpp/gguf-py/gguf/tensor_mapping.py +14 -3
  118. package/cpp/llama.cpp/include/llama.h +140 -38
  119. package/cpp/llama.cpp/requirements/requirements-compare-llama-bench.txt +1 -0
  120. package/cpp/llama.cpp/src/CMakeLists.txt +4 -1
  121. package/cpp/llama.cpp/src/llama-arch.cpp +95 -3
  122. package/cpp/llama.cpp/src/llama-arch.h +7 -1
  123. package/cpp/llama.cpp/src/llama-batch.cpp +289 -31
  124. package/cpp/llama.cpp/src/llama-batch.h +47 -17
  125. package/cpp/llama.cpp/src/llama-chat.cpp +19 -2
  126. package/cpp/llama.cpp/src/llama-chat.h +1 -0
  127. package/cpp/llama.cpp/src/llama-context.cpp +488 -313
  128. package/cpp/llama.cpp/src/llama-context.h +38 -17
  129. package/cpp/llama.cpp/src/llama-cparams.cpp +1 -1
  130. package/cpp/llama.cpp/src/llama-cparams.h +1 -1
  131. package/cpp/llama.cpp/src/llama-graph.cpp +275 -152
  132. package/cpp/llama.cpp/src/llama-graph.h +109 -52
  133. package/cpp/llama.cpp/src/llama-hparams.cpp +6 -2
  134. package/cpp/llama.cpp/src/llama-hparams.h +8 -2
  135. package/cpp/llama.cpp/src/llama-kv-cache-unified-iswa.cpp +281 -0
  136. package/cpp/llama.cpp/src/llama-kv-cache-unified-iswa.h +133 -0
  137. package/cpp/llama.cpp/src/llama-kv-cache-unified.cpp +1835 -0
  138. package/cpp/llama.cpp/src/llama-kv-cache-unified.h +308 -0
  139. package/cpp/llama.cpp/src/llama-kv-cells.h +53 -17
  140. package/cpp/llama.cpp/src/llama-memory-hybrid.cpp +247 -0
  141. package/cpp/llama.cpp/src/llama-memory-hybrid.h +143 -0
  142. package/cpp/llama.cpp/src/llama-memory-recurrent.cpp +1116 -0
  143. package/cpp/llama.cpp/src/llama-memory-recurrent.h +188 -0
  144. package/cpp/llama.cpp/src/llama-memory.cpp +41 -0
  145. package/cpp/llama.cpp/src/llama-memory.h +89 -4
  146. package/cpp/llama.cpp/src/llama-mmap.cpp +1 -1
  147. package/cpp/llama.cpp/src/llama-model-loader.cpp +42 -17
  148. package/cpp/llama.cpp/src/llama-model.cpp +735 -143
  149. package/cpp/llama.cpp/src/llama-model.h +4 -0
  150. package/cpp/llama.cpp/src/llama-quant.cpp +2 -1
  151. package/cpp/llama.cpp/src/llama-vocab.cpp +39 -25
  152. package/cpp/llama.cpp/src/llama.cpp +11 -7
  153. package/cpp/llama.cpp/src/unicode.cpp +5 -0
  154. package/cpp/llama.cpp/vendor/cpp-httplib/httplib.h +10518 -0
  155. package/cpp/llama.cpp/vendor/miniaudio/miniaudio.h +93468 -0
  156. package/cpp/llama.cpp/{common → vendor}/minja/chat-template.hpp +1 -1
  157. package/cpp/llama.cpp/{common → vendor}/minja/minja.hpp +1 -1
  158. package/cpp/llama.cpp/{common → vendor/nlohmann}/json.hpp +3027 -2267
  159. package/cpp/llama.cpp/vendor/nlohmann/json_fwd.hpp +187 -0
  160. package/cpp/llama.cpp/vendor/stb/stb_image.h +7988 -0
  161. package/cpp/rn-completion.cpp +65 -10
  162. package/cpp/{rn-llama.hpp → rn-llama.h} +1 -1
  163. package/cpp/{rn-utils.hpp → rn-utils.h} +8 -1
  164. package/ios/include/chat.h +1 -1
  165. package/ios/include/common/minja/chat-template.hpp +1 -1
  166. package/ios/include/common/minja/minja.hpp +1 -1
  167. package/ios/include/common.h +5 -2
  168. package/ios/include/json-schema-to-grammar.h +4 -4
  169. package/ios/include/llama.h +140 -38
  170. package/ios/include/{common → nlohmann}/json.hpp +3027 -2267
  171. package/ios/libs/llama.xcframework/Info.plist +20 -20
  172. package/ios/libs/llama.xcframework/ios-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  173. package/ios/libs/llama.xcframework/ios-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4863 -4617
  174. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/ggml.h +1 -3
  175. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/llama.h +140 -38
  176. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/llama +0 -0
  177. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  178. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4834 -4638
  179. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3742 -3557
  180. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +1 -3
  181. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/llama.h +140 -38
  182. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/llama +0 -0
  183. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  184. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4834 -4638
  185. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3744 -3559
  186. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml.h +1 -3
  187. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/llama.h +140 -38
  188. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml.h +1 -3
  189. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/llama.h +140 -38
  190. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/llama +0 -0
  191. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml.h +1 -3
  192. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/llama.h +140 -38
  193. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/llama +0 -0
  194. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/llama +0 -0
  195. package/ios/libs/llama.xcframework/tvos-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  196. package/ios/libs/llama.xcframework/tvos-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4863 -4616
  197. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml.h +1 -3
  198. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/llama.h +140 -38
  199. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/llama +0 -0
  200. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  201. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4834 -4637
  202. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3742 -3556
  203. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +1 -3
  204. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/llama.h +140 -38
  205. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/llama +0 -0
  206. package/ios/libs/llama.xcframework/xros-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  207. package/ios/libs/llama.xcframework/xros-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4900 -4653
  208. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml.h +1 -3
  209. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/llama.h +140 -38
  210. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/llama +0 -0
  211. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  212. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4871 -4674
  213. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3773 -3587
  214. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +1 -3
  215. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/llama.h +140 -38
  216. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/llama +0 -0
  217. package/package.json +1 -2
  218. package/cpp/llama.cpp/common/cmake/build-info-gen-cpp.cmake +0 -24
  219. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-aarch64.h +0 -8
  220. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-quants.c +0 -13891
  221. package/cpp/llama.cpp/src/llama-kv-cache.cpp +0 -2747
  222. package/cpp/llama.cpp/src/llama-kv-cache.h +0 -502
  223. /package/cpp/llama.cpp/ggml/src/ggml-cpu/{cpu-feats-x86.cpp → arch/x86/cpu-feats.cpp} +0 -0
  224. /package/cpp/llama.cpp/ggml/src/ggml-cpu/{ggml-cpu-hbm.h → hbm.h} +0 -0
  225. /package/cpp/llama.cpp/ggml/src/ggml-cpu/{ggml-cpu-traits.h → traits.h} +0 -0
package/cpp/llama.cpp/ggml/src/ggml-cpu/{ggml-cpu-quants.h → quants.h} RENAMED
@@ -58,6 +58,32 @@ void ggml_vec_dot_iq4_nl_q8_0 (int n, float * GGML_RESTRICT s, size_t bs, const
58
58
  void ggml_vec_dot_iq4_xs_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
59
59
  void ggml_vec_dot_iq3_s_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
60
60
 
61
+ // Generic implementation
62
+ void quantize_row_q8_0_generic(const float * GGML_RESTRICT x, void * GGML_RESTRICT vy, int64_t k);
63
+ void quantize_row_q8_1_generic(const float * GGML_RESTRICT x, void * GGML_RESTRICT vy, int64_t k);
64
+ void quantize_row_q8_K_generic(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
65
+ void ggml_vec_dot_q4_0_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
66
+ void ggml_vec_dot_q4_1_q8_1_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
67
+ void ggml_vec_dot_q5_0_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
68
+ void ggml_vec_dot_q5_1_q8_1_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
69
+ void ggml_vec_dot_q8_0_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
70
+ void ggml_vec_dot_tq1_0_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
71
+ void ggml_vec_dot_tq2_0_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
72
+ void ggml_vec_dot_q2_K_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
73
+ void ggml_vec_dot_q3_K_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
74
+ void ggml_vec_dot_q4_K_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
75
+ void ggml_vec_dot_q5_K_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
76
+ void ggml_vec_dot_q6_K_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
77
+ void ggml_vec_dot_iq2_xxs_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
78
+ void ggml_vec_dot_iq2_xs_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
79
+ void ggml_vec_dot_iq2_s_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
80
+ void ggml_vec_dot_iq3_xxs_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
81
+ void ggml_vec_dot_iq3_s_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
82
+ void ggml_vec_dot_iq1_s_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
83
+ void ggml_vec_dot_iq1_m_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
84
+ void ggml_vec_dot_iq4_nl_q8_0_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
85
+ void ggml_vec_dot_iq4_xs_q8_K_generic(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
86
+
61
87
  #ifdef __cplusplus
62
88
  }
63
89
  #endif