@novastera-oss/llamarn 0.2.1 → 0.2.2

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 (266) hide show
  1. package/README.md +80 -14
  2. package/RNLlamaCpp.podspec +10 -3
  3. package/android/CMakeLists.txt +8 -0
  4. package/android/src/main/cpp/include/llama.h +62 -125
  5. package/android/src/main/jniLibs/arm64-v8a/libggml-base.so +0 -0
  6. package/android/src/main/jniLibs/arm64-v8a/libggml-cpu.so +0 -0
  7. package/android/src/main/jniLibs/arm64-v8a/libggml.so +0 -0
  8. package/android/src/main/jniLibs/arm64-v8a/libllama.so +0 -0
  9. package/android/src/main/jniLibs/x86_64/libggml-base.so +0 -0
  10. package/android/src/main/jniLibs/x86_64/libggml-cpu.so +0 -0
  11. package/android/src/main/jniLibs/x86_64/libggml.so +0 -0
  12. package/android/src/main/jniLibs/x86_64/libllama.so +0 -0
  13. package/cpp/build-info.cpp +2 -2
  14. package/cpp/llama.cpp/README.md +11 -3
  15. package/cpp/llama.cpp/build-xcframework.sh +1 -0
  16. package/cpp/llama.cpp/common/CMakeLists.txt +8 -2
  17. package/cpp/llama.cpp/common/arg.cpp +153 -113
  18. package/cpp/llama.cpp/common/chat-parser.cpp +379 -0
  19. package/cpp/llama.cpp/common/chat-parser.h +117 -0
  20. package/cpp/llama.cpp/common/chat.cpp +847 -699
  21. package/cpp/llama.cpp/common/chat.h +73 -6
  22. package/cpp/llama.cpp/common/common.cpp +50 -82
  23. package/cpp/llama.cpp/common/common.h +21 -17
  24. package/cpp/llama.cpp/common/json-partial.cpp +255 -0
  25. package/cpp/llama.cpp/common/json-partial.h +37 -0
  26. package/cpp/llama.cpp/common/minja/chat-template.hpp +9 -5
  27. package/cpp/llama.cpp/common/minja/minja.hpp +69 -36
  28. package/cpp/llama.cpp/common/regex-partial.cpp +204 -0
  29. package/cpp/llama.cpp/common/regex-partial.h +56 -0
  30. package/cpp/llama.cpp/common/sampling.cpp +7 -8
  31. package/cpp/llama.cpp/convert_hf_to_gguf.py +453 -118
  32. package/cpp/llama.cpp/convert_hf_to_gguf_update.py +120 -68
  33. package/cpp/llama.cpp/ggml/CMakeLists.txt +2 -1
  34. package/cpp/llama.cpp/ggml/cmake/common.cmake +25 -0
  35. package/cpp/llama.cpp/ggml/include/ggml-opt.h +49 -28
  36. package/cpp/llama.cpp/ggml/include/ggml.h +26 -7
  37. package/cpp/llama.cpp/ggml/src/CMakeLists.txt +16 -10
  38. package/cpp/llama.cpp/ggml/src/ggml-backend.cpp +4 -1
  39. package/cpp/llama.cpp/ggml/src/ggml-cann/CMakeLists.txt +1 -0
  40. package/cpp/llama.cpp/ggml/src/ggml-cann/acl_tensor.cpp +2 -0
  41. package/cpp/llama.cpp/ggml/src/ggml-cann/aclnn_ops.cpp +604 -0
  42. package/cpp/llama.cpp/ggml/src/ggml-cann/aclnn_ops.h +42 -0
  43. package/cpp/llama.cpp/ggml/src/ggml-cann/ggml-cann.cpp +54 -2
  44. package/cpp/llama.cpp/ggml/src/ggml-cpu/CMakeLists.txt +50 -51
  45. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-aarch64.cpp +2 -2
  46. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-impl.h +5 -9
  47. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-quants.c +779 -19
  48. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.c +22 -0
  49. package/cpp/llama.cpp/ggml/src/ggml-cpu/kleidiai/kernels.cpp +88 -5
  50. package/cpp/llama.cpp/ggml/src/ggml-cpu/kleidiai/kernels.h +47 -12
  51. package/cpp/llama.cpp/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +264 -69
  52. package/cpp/llama.cpp/ggml/src/ggml-cpu/ops.cpp +322 -100
  53. package/cpp/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h +117 -1
  54. package/cpp/llama.cpp/ggml/src/ggml-cpu/vec.cpp +85 -16
  55. package/cpp/llama.cpp/ggml/src/ggml-cpu/vec.h +220 -49
  56. package/cpp/llama.cpp/ggml/src/ggml-cuda/acc.cu +40 -26
  57. package/cpp/llama.cpp/ggml/src/ggml-cuda/common.cuh +1 -1
  58. package/cpp/llama.cpp/ggml/src/ggml-cuda/cpy.cu +11 -1
  59. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-common.cuh +15 -7
  60. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-mma-f16.cuh +266 -64
  61. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-vec-f16.cuh +49 -4
  62. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-vec-f32.cuh +48 -4
  63. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn.cu +2 -1
  64. package/cpp/llama.cpp/ggml/src/ggml-cuda/ggml-cuda.cu +5 -1
  65. package/cpp/llama.cpp/ggml/src/ggml-cuda/mmq.cu +2 -0
  66. package/cpp/llama.cpp/ggml/src/ggml-cuda/quantize.cu +7 -6
  67. package/cpp/llama.cpp/ggml/src/ggml-cuda/sum.cu +1 -1
  68. package/cpp/llama.cpp/ggml/src/ggml-cuda/unary.cu +10 -0
  69. package/cpp/llama.cpp/ggml/src/ggml-cuda/unary.cuh +2 -0
  70. package/cpp/llama.cpp/ggml/src/ggml-impl.h +1 -1
  71. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-impl.h +4 -0
  72. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal.m +99 -17
  73. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal.metal +200 -2
  74. package/cpp/llama.cpp/ggml/src/ggml-musa/CMakeLists.txt +8 -2
  75. package/cpp/llama.cpp/ggml/src/ggml-musa/mudnn.cu +112 -0
  76. package/cpp/llama.cpp/ggml/src/ggml-musa/mudnn.cuh +12 -0
  77. package/cpp/llama.cpp/ggml/src/ggml-opencl/CMakeLists.txt +6 -0
  78. package/cpp/llama.cpp/ggml/src/ggml-opencl/ggml-opencl.cpp +972 -178
  79. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/argsort.cl +86 -0
  80. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/div.cl +72 -0
  81. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/group_norm.cl +72 -0
  82. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/sigmoid.cl +29 -0
  83. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/sub.cl +72 -0
  84. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/sum_rows.cl +39 -0
  85. package/cpp/llama.cpp/ggml/src/ggml-opt.cpp +373 -190
  86. package/cpp/llama.cpp/ggml/src/ggml-sycl/CMakeLists.txt +29 -23
  87. package/cpp/llama.cpp/ggml/src/ggml-sycl/binbcast.cpp +5 -10
  88. package/cpp/llama.cpp/ggml/src/ggml-sycl/common.hpp +101 -5
  89. package/cpp/llama.cpp/ggml/src/ggml-sycl/concat.cpp +31 -33
  90. package/cpp/llama.cpp/ggml/src/ggml-sycl/conv.cpp +1 -0
  91. package/cpp/llama.cpp/ggml/src/ggml-sycl/convert.cpp +29 -2
  92. package/cpp/llama.cpp/ggml/src/ggml-sycl/cpy.cpp +4 -5
  93. package/cpp/llama.cpp/ggml/src/ggml-sycl/dequantize.hpp +59 -21
  94. package/cpp/llama.cpp/ggml/src/ggml-sycl/dmmv.cpp +9 -1
  95. package/cpp/llama.cpp/ggml/src/ggml-sycl/element_wise.cpp +84 -72
  96. package/cpp/llama.cpp/ggml/src/ggml-sycl/element_wise.hpp +2 -0
  97. package/cpp/llama.cpp/ggml/src/ggml-sycl/gemm.hpp +37 -8
  98. package/cpp/llama.cpp/ggml/src/ggml-sycl/getrows.cpp +1 -3
  99. package/cpp/llama.cpp/ggml/src/ggml-sycl/ggml-sycl.cpp +324 -129
  100. package/cpp/llama.cpp/ggml/src/ggml-sycl/gla.cpp +1 -0
  101. package/cpp/llama.cpp/ggml/src/ggml-sycl/mmvq.cpp +31 -2
  102. package/cpp/llama.cpp/ggml/src/ggml-sycl/norm.cpp +95 -68
  103. package/cpp/llama.cpp/ggml/src/ggml-sycl/outprod.cpp +1 -0
  104. package/cpp/llama.cpp/ggml/src/ggml-sycl/quants.hpp +22 -0
  105. package/cpp/llama.cpp/ggml/src/ggml-sycl/rope.cpp +1 -2
  106. package/cpp/llama.cpp/ggml/src/ggml-sycl/softmax.cpp +1 -4
  107. package/cpp/llama.cpp/ggml/src/ggml-sycl/tsembd.cpp +2 -3
  108. package/cpp/llama.cpp/ggml/src/ggml-sycl/vecdotq.hpp +69 -43
  109. package/cpp/llama.cpp/ggml/src/ggml-sycl/wkv.cpp +2 -14
  110. package/cpp/llama.cpp/ggml/src/ggml-vulkan/CMakeLists.txt +81 -91
  111. package/cpp/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp +432 -181
  112. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/CMakeLists.txt +17 -0
  113. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq1_m.comp +1 -1
  114. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +6 -152
  115. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.comp +162 -0
  116. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +360 -0
  117. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +2 -118
  118. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +1 -1
  119. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +12 -1
  120. package/cpp/llama.cpp/ggml/src/ggml.c +107 -36
  121. package/cpp/llama.cpp/ggml/src/gguf.cpp +33 -33
  122. package/cpp/llama.cpp/gguf-py/gguf/constants.py +100 -15
  123. package/cpp/llama.cpp/gguf-py/gguf/gguf_reader.py +1 -1
  124. package/cpp/llama.cpp/gguf-py/gguf/gguf_writer.py +44 -12
  125. package/cpp/llama.cpp/gguf-py/gguf/scripts/gguf_editor_gui.py +21 -10
  126. package/cpp/llama.cpp/gguf-py/gguf/scripts/gguf_new_metadata.py +5 -2
  127. package/cpp/llama.cpp/gguf-py/gguf/tensor_mapping.py +128 -31
  128. package/cpp/llama.cpp/gguf-py/gguf/utility.py +1 -1
  129. package/cpp/llama.cpp/gguf-py/pyproject.toml +1 -1
  130. package/cpp/llama.cpp/include/llama.h +62 -125
  131. package/cpp/llama.cpp/models/ggml-vocab-bert-bge.gguf.inp +1 -1
  132. package/cpp/llama.cpp/models/ggml-vocab-bert-bge.gguf.out +1 -1
  133. package/cpp/llama.cpp/models/ggml-vocab-command-r.gguf.inp +1 -1
  134. package/cpp/llama.cpp/models/ggml-vocab-command-r.gguf.out +1 -1
  135. package/cpp/llama.cpp/models/ggml-vocab-deepseek-coder.gguf.inp +1 -1
  136. package/cpp/llama.cpp/models/ggml-vocab-deepseek-coder.gguf.out +1 -1
  137. package/cpp/llama.cpp/models/ggml-vocab-deepseek-llm.gguf.inp +1 -1
  138. package/cpp/llama.cpp/models/ggml-vocab-deepseek-llm.gguf.out +1 -1
  139. package/cpp/llama.cpp/models/ggml-vocab-falcon.gguf.inp +1 -1
  140. package/cpp/llama.cpp/models/ggml-vocab-falcon.gguf.out +1 -1
  141. package/cpp/llama.cpp/models/ggml-vocab-gpt-2.gguf.inp +1 -1
  142. package/cpp/llama.cpp/models/ggml-vocab-gpt-2.gguf.out +1 -1
  143. package/cpp/llama.cpp/models/ggml-vocab-llama-bpe.gguf.inp +1 -1
  144. package/cpp/llama.cpp/models/ggml-vocab-llama-bpe.gguf.out +1 -1
  145. package/cpp/llama.cpp/models/ggml-vocab-llama-spm.gguf.inp +1 -1
  146. package/cpp/llama.cpp/models/ggml-vocab-llama-spm.gguf.out +1 -1
  147. package/cpp/llama.cpp/models/ggml-vocab-mpt.gguf.inp +1 -1
  148. package/cpp/llama.cpp/models/ggml-vocab-mpt.gguf.out +1 -1
  149. package/cpp/llama.cpp/models/ggml-vocab-nomic-bert-moe.gguf +0 -0
  150. package/cpp/llama.cpp/models/ggml-vocab-phi-3.gguf.inp +1 -1
  151. package/cpp/llama.cpp/models/ggml-vocab-phi-3.gguf.out +1 -1
  152. package/cpp/llama.cpp/models/ggml-vocab-qwen2.gguf.inp +1 -1
  153. package/cpp/llama.cpp/models/ggml-vocab-qwen2.gguf.out +1 -1
  154. package/cpp/llama.cpp/models/ggml-vocab-refact.gguf.inp +1 -1
  155. package/cpp/llama.cpp/models/ggml-vocab-refact.gguf.out +1 -1
  156. package/cpp/llama.cpp/models/ggml-vocab-starcoder.gguf.inp +1 -1
  157. package/cpp/llama.cpp/models/ggml-vocab-starcoder.gguf.out +1 -1
  158. package/cpp/llama.cpp/models/templates/Qwen-QwQ-32B.jinja +62 -0
  159. package/cpp/llama.cpp/models/templates/Qwen-Qwen3-0.6B.jinja +85 -0
  160. package/cpp/llama.cpp/models/templates/README.md +2 -0
  161. package/cpp/llama.cpp/requirements/requirements-convert_hf_to_gguf.txt +5 -1
  162. package/cpp/llama.cpp/requirements/requirements-convert_hf_to_gguf_update.txt +5 -1
  163. package/cpp/llama.cpp/requirements/requirements-convert_lora_to_gguf.txt +2 -0
  164. package/cpp/llama.cpp/requirements/requirements-gguf_editor_gui.txt +1 -1
  165. package/cpp/llama.cpp/src/CMakeLists.txt +2 -0
  166. package/cpp/llama.cpp/src/llama-arch.cpp +6 -0
  167. package/cpp/llama.cpp/src/llama-arch.h +2 -0
  168. package/cpp/llama.cpp/src/llama-batch.cpp +3 -1
  169. package/cpp/llama.cpp/src/llama-context.cpp +340 -123
  170. package/cpp/llama.cpp/src/llama-context.h +30 -0
  171. package/cpp/llama.cpp/src/llama-cparams.cpp +4 -0
  172. package/cpp/llama.cpp/src/llama-cparams.h +2 -0
  173. package/cpp/llama.cpp/src/llama-grammar.cpp +12 -2
  174. package/cpp/llama.cpp/src/llama-graph.cpp +157 -247
  175. package/cpp/llama.cpp/src/llama-graph.h +52 -7
  176. package/cpp/llama.cpp/src/llama-hparams.cpp +17 -1
  177. package/cpp/llama.cpp/src/llama-hparams.h +37 -5
  178. package/cpp/llama.cpp/src/llama-kv-cache.cpp +742 -481
  179. package/cpp/llama.cpp/src/llama-kv-cache.h +196 -99
  180. package/cpp/llama.cpp/src/llama-kv-cells.h +379 -0
  181. package/cpp/llama.cpp/src/llama-memory.h +4 -3
  182. package/cpp/llama.cpp/src/llama-model-loader.cpp +22 -17
  183. package/cpp/llama.cpp/src/llama-model-saver.cpp +281 -0
  184. package/cpp/llama.cpp/src/llama-model-saver.h +37 -0
  185. package/cpp/llama.cpp/src/llama-model.cpp +529 -172
  186. package/cpp/llama.cpp/src/llama-model.h +6 -1
  187. package/cpp/llama.cpp/src/llama-quant.cpp +15 -13
  188. package/cpp/llama.cpp/src/llama-sampling.cpp +2 -2
  189. package/cpp/llama.cpp/src/llama-vocab.cpp +35 -8
  190. package/cpp/llama.cpp/src/llama-vocab.h +6 -0
  191. package/cpp/llama.cpp/src/llama.cpp +14 -0
  192. package/cpp/rn-completion.cpp +4 -2
  193. package/ios/include/chat.h +73 -6
  194. package/ios/include/common/minja/chat-template.hpp +9 -5
  195. package/ios/include/common/minja/minja.hpp +69 -36
  196. package/ios/include/common.h +21 -17
  197. package/ios/include/llama.h +62 -125
  198. package/ios/libs/llama.xcframework/Info.plist +19 -19
  199. package/ios/libs/llama.xcframework/ios-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  200. package/ios/libs/llama.xcframework/ios-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4617 -4487
  201. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/ggml-opt.h +237 -0
  202. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/ggml.h +26 -7
  203. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/llama.h +62 -125
  204. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/llama +0 -0
  205. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  206. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4638 -4508
  207. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3557 -3435
  208. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml-opt.h +237 -0
  209. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +26 -7
  210. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/llama.h +62 -125
  211. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/llama +0 -0
  212. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  213. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4638 -4508
  214. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3559 -3437
  215. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml-opt.h +237 -0
  216. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml.h +26 -7
  217. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/llama.h +62 -125
  218. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml-opt.h +237 -0
  219. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml.h +26 -7
  220. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/llama.h +62 -125
  221. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/llama +0 -0
  222. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml-opt.h +237 -0
  223. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml.h +26 -7
  224. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/llama.h +62 -125
  225. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/llama +0 -0
  226. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/llama +0 -0
  227. package/ios/libs/llama.xcframework/tvos-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  228. package/ios/libs/llama.xcframework/tvos-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4616 -4487
  229. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml-opt.h +237 -0
  230. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml.h +26 -7
  231. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/llama.h +62 -125
  232. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/llama +0 -0
  233. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  234. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4637 -4508
  235. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3556 -3435
  236. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml-opt.h +237 -0
  237. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +26 -7
  238. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/llama.h +62 -125
  239. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/llama +0 -0
  240. package/ios/libs/llama.xcframework/xros-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  241. package/ios/libs/llama.xcframework/xros-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4653 -4523
  242. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml-opt.h +237 -0
  243. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml.h +26 -7
  244. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/llama.h +62 -125
  245. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/llama +0 -0
  246. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  247. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4674 -4544
  248. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3587 -3465
  249. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml-opt.h +237 -0
  250. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +26 -7
  251. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/llama.h +62 -125
  252. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/llama +0 -0
  253. package/package.json +1 -1
  254. package/cpp/llama.cpp/common/stb_image.h +0 -7988
  255. package/cpp/llama.cpp/models/ggml-vocab-chameleon.gguf.inp +0 -112
  256. package/cpp/llama.cpp/models/ggml-vocab-chameleon.gguf.out +0 -46
  257. package/cpp/llama.cpp/models/ggml-vocab-deepseek-r1-qwen.gguf.inp +0 -112
  258. package/cpp/llama.cpp/models/ggml-vocab-deepseek-r1-qwen.gguf.out +0 -46
  259. package/cpp/llama.cpp/models/ggml-vocab-gpt-4o.gguf.inp +0 -112
  260. package/cpp/llama.cpp/models/ggml-vocab-gpt-4o.gguf.out +0 -46
  261. package/cpp/llama.cpp/models/ggml-vocab-llama4.gguf.inp +0 -112
  262. package/cpp/llama.cpp/models/ggml-vocab-llama4.gguf.out +0 -46
  263. package/cpp/llama.cpp/models/ggml-vocab-pixtral.gguf.inp +0 -112
  264. package/cpp/llama.cpp/models/ggml-vocab-pixtral.gguf.out +0 -46
  265. package/cpp/llama.cpp/models/ggml-vocab-roberta-bpe.gguf.inp +0 -112
  266. package/cpp/llama.cpp/models/ggml-vocab-roberta-bpe.gguf.out +0 -46
package/cpp/llama.cpp/ggml/src/ggml-cann/ggml-cann.cpp CHANGED
@@ -36,6 +36,7 @@
36
36
  #include "ggml-backend-impl.h"
37
37
  #include "ggml-cann/aclnn_ops.h"
38
38
  #include "ggml-cann/common.h"
39
+ #include "ggml.h"
39
40
 
40
41
  #define GGML_COMMON_DECL_C
41
42
 
@@ -1672,7 +1673,8 @@ static bool ggml_cann_compute_forward(ggml_backend_cann_context& ctx,
1672
1673
  ggml_cann_mul_mat(ctx, dst);
1673
1674
  break;
1674
1675
  case GGML_OP_MUL_MAT_ID:
1675
- return false;
1676
+ ggml_cann_mul_mat_id(ctx, dst);
1677
+ break;
1676
1678
  case GGML_OP_SCALE:
1677
1679
  ggml_cann_scale(ctx, dst);
1678
1680
  break;
@@ -1747,6 +1749,9 @@ static bool ggml_cann_compute_forward(ggml_backend_cann_context& ctx,
1747
1749
  case GGML_OP_COUNT_EQUAL:
1748
1750
  ggml_cann_count_equal(ctx, dst);
1749
1751
  break;
1752
+ case GGML_OP_FLASH_ATTN_EXT:
1753
+ ggml_cann_flash_attn_ext(ctx, dst);
1754
+ break;
1750
1755
  default:
1751
1756
  return false;
1752
1757
  }
@@ -2030,7 +2035,22 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
2030
2035
  }
2031
2036
  }
2032
2037
  case GGML_OP_MUL_MAT_ID:
2033
- return false;
2038
+ switch (op->src[0]->type) {
2039
+ case GGML_TYPE_F16:
2040
+ case GGML_TYPE_F32:
2041
+ return true;
2042
+ case GGML_TYPE_Q8_0:
2043
+ case GGML_TYPE_Q4_0:
2044
+ #ifdef ASCEND_310P
2045
+ // Q4 && Q8 per group is not suppor on 310p device
2046
+ return false;
2047
+ #endif
2048
+ // only support contiguous for quantized types.
2049
+ return ggml_is_contiguous(op->src[0]) &&
2050
+ ggml_is_contiguous(op->src[1]);
2051
+ default:
2052
+ return false;
2053
+ }
2034
2054
  // embedding
2035
2055
  case GGML_OP_GET_ROWS: {
2036
2056
  switch (op->src[0]->type) {
@@ -2161,6 +2181,38 @@ static bool ggml_backend_cann_supports_op(ggml_backend_dev_t dev,
2161
2181
  case GGML_OP_PAD_REFLECT_1D:
2162
2182
  case GGML_OP_COUNT_EQUAL:
2163
2183
  return true;
2184
+ case GGML_OP_FLASH_ATTN_EXT:{
2185
+ // derived from [ggml-cuda.cu]
2186
+ if(op->src[1]->type != GGML_TYPE_F16 || op->src[2]->type != GGML_TYPE_F16){
2187
+ return false;
2188
+ }
2189
+ if(op->src[1]->type != GGML_TYPE_F16 && op->src[1]->type != GGML_TYPE_F32 && op->src[1]->type != GGML_TYPE_BF16){
2190
+ return false;
2191
+ }
2192
+ if(op->type != GGML_TYPE_F16 && op->type != GGML_TYPE_F32 && op->type != GGML_TYPE_BF16){
2193
+ return false;
2194
+ }
2195
+ if (op->src[1]->ne[0] != op->src[2]->ne[0]) {
2196
+ // different head sizes of K and V are not supported yet
2197
+ return false;
2198
+ }
2199
+ if (op->src[0]->ne[0] == 192) {
2200
+ return false;
2201
+ }
2202
+ if (op->src[0]->ne[0] == 576) {
2203
+ // DeepSeek MLA
2204
+ return false;
2205
+ }
2206
+ if (op->src[0]->ne[3] != 1) {
2207
+ return false;
2208
+ }
2209
+ float logitSoftcap = 0.0f;
2210
+ memcpy(&logitSoftcap, (float*)op->op_params + 2, sizeof(float));
2211
+ if(logitSoftcap != 0.0f) {
2212
+ return false;
2213
+ }
2214
+ return true;
2215
+ }
2164
2216
  default:
2165
2217
  return false;
2166
2218
  }
package/cpp/llama.cpp/ggml/src/ggml-cpu/CMakeLists.txt CHANGED
@@ -82,13 +82,8 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
82
82
  target_link_libraries(${GGML_CPU_NAME} PUBLIC memkind)
83
83
  endif()
84
84
 
85
- if (CMAKE_OSX_ARCHITECTURES STREQUAL "arm64" OR
86
- CMAKE_GENERATOR_PLATFORM_LWR STREQUAL "arm64" OR
87
- (NOT CMAKE_OSX_ARCHITECTURES AND NOT CMAKE_GENERATOR_PLATFORM_LWR AND
88
- CMAKE_SYSTEM_PROCESSOR MATCHES "^(aarch64|arm.*|ARM64)$"))
89
-
85
+ if (GGML_SYSTEM_ARCH STREQUAL "ARM")
90
86
  message(STATUS "ARM detected")
91
-
92
87
  if (MSVC AND NOT CMAKE_C_COMPILER_ID STREQUAL "Clang")
93
88
  message(FATAL_ERROR "MSVC is not supported for ARM, use clang")
94
89
  else()
@@ -170,12 +165,8 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
170
165
  endforeach()
171
166
  endif()
172
167
  endif()
173
- elseif (CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64" OR CMAKE_GENERATOR_PLATFORM_LWR MATCHES "^(x86_64|i686|amd64|x64|win32)$" OR
174
- (NOT CMAKE_OSX_ARCHITECTURES AND NOT CMAKE_GENERATOR_PLATFORM_LWR AND
175
- CMAKE_SYSTEM_PROCESSOR MATCHES "^(x86_64|i686|AMD64|amd64)$"))
176
-
168
+ elseif (GGML_SYSTEM_ARCH STREQUAL "x86")
177
169
  message(STATUS "x86 detected")
178
-
179
170
  if (MSVC)
180
171
  # instruction set detection for MSVC only
181
172
  if (GGML_NATIVE)
@@ -299,7 +290,26 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
299
290
  endif()
300
291
  endif()
301
292
  endif()
302
- elseif ("${CMAKE_SYSTEM_PROCESSOR} " STREQUAL "ppc64le " OR "${CMAKE_SYSTEM_PROCESSOR} " STREQUAL "powerpc ")
293
+
294
+ if (GGML_BACKEND_DL)
295
+ if (GGML_NATIVE)
296
+ # the feature check relies on ARCH_DEFINITIONS, but it is not set with GGML_NATIVE
297
+ message(FATAL_ERROR "GGML_NATIVE is not compatible with GGML_BACKEND_DL, consider using GGML_CPU_ALL_VARIANTS")
298
+ endif()
299
+
300
+ # The feature detection code is compiled as a separate target so that
301
+ # it can be built without the architecture flags
302
+ # Since multiple variants of the CPU backend may be included in the same
303
+ # build, using set_source_files_properties() to set the arch flags is not possible
304
+ set(GGML_CPU_FEATS_NAME ${GGML_CPU_NAME}-feats)
305
+ add_library(${GGML_CPU_FEATS_NAME} OBJECT ggml-cpu/cpu-feats-x86.cpp)
306
+ target_include_directories(${GGML_CPU_FEATS_NAME} PRIVATE . .. ../include)
307
+ target_compile_definitions(${GGML_CPU_FEATS_NAME} PRIVATE ${ARCH_DEFINITIONS})
308
+ target_compile_definitions(${GGML_CPU_FEATS_NAME} PRIVATE GGML_BACKEND_DL GGML_BACKEND_BUILD GGML_BACKEND_SHARED)
309
+ set_target_properties(${GGML_CPU_FEATS_NAME} PROPERTIES POSITION_INDEPENDENT_CODE ON)
310
+ target_link_libraries(${GGML_CPU_NAME} PRIVATE ${GGML_CPU_FEATS_NAME})
311
+ endif()
312
+ elseif (GGML_SYSTEM_ARCH STREQUAL "PowerPC")
303
313
  message(STATUS "PowerPC detected")
304
314
  if (GGML_NATIVE)
305
315
  if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "ppc64")
@@ -325,9 +335,8 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
325
335
  list(APPEND ARCH_FLAGS -mcpu=${GGML_CPU_POWERPC_CPUTYPE})
326
336
  endif()
327
337
  endif()
328
- elseif (${CMAKE_SYSTEM_PROCESSOR} MATCHES "loongarch64")
338
+ elseif (GGML_SYSTEM_ARCH STREQUAL "loongarch64")
329
339
  message(STATUS "loongarch64 detected")
330
-
331
340
  list(APPEND ARCH_FLAGS -march=loongarch64)
332
341
  if (GGML_LASX)
333
342
  list(APPEND ARCH_FLAGS -mlasx)
@@ -335,16 +344,18 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
335
344
  if (GGML_LSX)
336
345
  list(APPEND ARCH_FLAGS -mlsx)
337
346
  endif()
338
- elseif (${CMAKE_SYSTEM_PROCESSOR} MATCHES "riscv64")
339
- message(STATUS "RISC-V detected")
347
+ elseif (GGML_SYSTEM_ARCH STREQUAL "riscv64")
348
+ message(STATUS "riscv64 detected")
340
349
  if (GGML_RVV)
341
- if (GGML_RV_ZFH)
342
- list(APPEND ARCH_FLAGS -march=rv64gcv_zfhmin -DGGML_RV_ZFH -mabi=lp64d)
350
+ if (GGML_XTHEADVECTOR)
351
+ list(APPEND ARCH_FLAGS -march=rv64gc_xtheadvector -mabi=lp64d)
352
+ elseif (GGML_RV_ZFH)
353
+ list(APPEND ARCH_FLAGS -march=rv64gcv_zfhmin -mabi=lp64d)
343
354
  else()
344
355
  list(APPEND ARCH_FLAGS -march=rv64gcv -mabi=lp64d)
345
356
  endif()
346
357
  endif()
347
- elseif (${CMAKE_SYSTEM_PROCESSOR} MATCHES "s390x")
358
+ elseif (GGML_SYSTEM_ARCH STREQUAL "s390x")
348
359
  message(STATUS "s390x detected")
349
360
  file(READ "/proc/cpuinfo" CPUINFO_CONTENTS)
350
361
  string(REGEX REPLACE "machine[ \t\r\n]*=[ \t\r\n]*([0-9]+)" "\\1" S390X_M ${CPUINFO_CONTENTS})
@@ -385,9 +396,9 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
385
396
 
386
397
  # Fetch KleidiAI sources:
387
398
  include(FetchContent)
388
- set(KLEIDIAI_COMMIT_TAG "v1.5.0")
399
+ set(KLEIDIAI_COMMIT_TAG "v1.6.0")
389
400
  set(KLEIDIAI_DOWNLOAD_URL "https://github.com/ARM-software/kleidiai/archive/refs/tags/${KLEIDIAI_COMMIT_TAG}.tar.gz")
390
- set(KLEIDIAI_ARCHIVE_MD5 "ea22e1aefb800e9bc8c74d91633cc58e")
401
+ set(KLEIDIAI_ARCHIVE_MD5 "75b4ad68f25ab673dcc01065e5a0b05f")
391
402
 
392
403
  if (POLICY CMP0135)
393
404
  cmake_policy(SET CMP0135 NEW)
@@ -428,6 +439,7 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
428
439
  ${KLEIDIAI_SRC}/kai/ukernels/
429
440
  ${KLEIDIAI_SRC}/kai/ukernels/matmul/
430
441
  ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/
442
+ ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_fp32_bf16p_bf16p/
431
443
  ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/)
432
444
 
433
445
  set(ARCH_FLAGS_TEMP "${ARCH_FLAGS}")
@@ -438,17 +450,19 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
438
450
  string(FIND "${ARCH_FLAGS_TEMP}" "+i8mm" I8MM_ENABLED)
439
451
  string(FIND "${ARCH_FLAGS_TEMP}" "+sme" SME_ENABLED)
440
452
 
441
- set(PRIVATE_ARCH_FLAGS ${ARCH_FLAGS})
453
+ set(PRIVATE_ARCH_FLAGS ${ARCH_FLAGS_TEMP})
442
454
 
443
- list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_lhs_quant_pack_qsi8d32p_f32.c)
444
- list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_rhs_pack_nxk_qsi4c32ps1s0scalef16_qsu4c32s16s0_neon.c)
445
- list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_lhs_quant_pack_qsi8d32p_f32_neon.c)
446
- list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_rhs_pack_nxk_qsi4c32pscalef16_qsu4c32s16s0.c)
455
+ list(APPEND GGML_KLEIDIAI_SOURCES
456
+ ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_lhs_quant_pack_qsi8d32p_f32.c
457
+ ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_rhs_pack_nxk_qsi4c32ps1s0scalef16_qsu4c32s16s0_neon.c
458
+ ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_lhs_quant_pack_qsi8d32p_f32_neon.c
459
+ ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_rhs_pack_nxk_qsi4c32pscalef16_qsu4c32s16s0.c)
447
460
 
448
461
  if (NOT DOTPROD_ENABLED MATCHES -1)
449
- list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p1x8_qsi4c32p4x8_1x4x32_neon_dotprod.c)
450
- list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p1x4_qsi4c32p4x4_1x4_neon_dotprod.c)
451
- list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p4x4_qsi4c32p4x4_16x4_neon_dotprod.c)
462
+ list(APPEND GGML_KLEIDIAI_SOURCES
463
+ ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p1x8_qsi4c32p4x8_1x4x32_neon_dotprod.c
464
+ ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p1x4_qsi4c32p4x4_1x4_neon_dotprod.c
465
+ ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p4x4_qsi4c32p4x4_16x4_neon_dotprod.c)
452
466
  endif()
453
467
 
454
468
  if (NOT I8MM_ENABLED MATCHES -1)
@@ -456,9 +470,13 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
456
470
  endif()
457
471
 
458
472
  if (NOT SME_ENABLED MATCHES -1)
459
- list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p1vlx4_qsi4c32p4vlx4_1vlx4vl_sme2_mopa.c)
460
- list(APPEND GGML_KLEIDIAI_SOURCES ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p1x4_qsi4c32p4vlx4_1x4vl_sme2_sdot.c)
461
- set(PRIVATE_ARCH_FLAGS "${PRIVATE_ARCH_FLAGS}+sve+sve2")
473
+ list(APPEND GGML_KLEIDIAI_SOURCES
474
+ ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p1vlx4_qsi4c32p4vlx4_1vlx4vl_sme2_mopa.c
475
+ ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_f32_qsi8d32p_qsi4c32p/kai_matmul_clamp_f32_qsi8d32p1x4_qsi4c32p4vlx4_1x4vl_sme2_sdot.c
476
+ ${KLEIDIAI_SRC}/kai/ukernels/matmul/matmul_clamp_fp32_bf16p_bf16p/kai_matmul_clamp_f32_bf16p2vlx2_bf16p2vlx2_2vlx2vl_sme2_mopa.c
477
+ ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_lhs_pack_bf16p2vlx2_f32_sme.c
478
+ ${KLEIDIAI_SRC}/kai/ukernels/matmul/pack/kai_rhs_pack_kxn_bf16p2vlx2b_f32_x32_sme.c)
479
+ set(PRIVATE_ARCH_FLAGS "-fno-tree-vectorize;${PRIVATE_ARCH_FLAGS}+sve+sve2")
462
480
  endif()
463
481
 
464
482
  set_source_files_properties(${GGML_KLEIDIAI_SOURCES} PROPERTIES COMPILE_OPTIONS "${PRIVATE_ARCH_FLAGS}")
@@ -470,25 +488,6 @@ function(ggml_add_cpu_backend_variant_impl tag_name)
470
488
  target_compile_options(${GGML_CPU_NAME} PRIVATE ${ARCH_FLAGS})
471
489
  target_compile_definitions(${GGML_CPU_NAME} PRIVATE ${ARCH_DEFINITIONS})
472
490
 
473
- if (GGML_BACKEND_DL)
474
- if (GGML_NATIVE)
475
- # the feature check relies on ARCH_DEFINITIONS, but it is not set with GGML_NATIVE
476
- message(FATAL_ERROR "GGML_NATIVE is not compatible with GGML_BACKEND_DL, consider using GGML_CPU_ALL_VARIANTS")
477
- endif()
478
-
479
- # The feature detection code is compiled as a separate target so that
480
- # it can be built without the architecture flags
481
- # Since multiple variants of the CPU backend may be included in the same
482
- # build, using set_source_files_properties() to set the arch flags is not possible
483
- set(GGML_CPU_FEATS_NAME ${GGML_CPU_NAME}-feats)
484
- add_library(${GGML_CPU_FEATS_NAME} OBJECT ggml-cpu/cpu-feats-x86.cpp)
485
- target_include_directories(${GGML_CPU_FEATS_NAME} PRIVATE . .. ../include)
486
- target_compile_definitions(${GGML_CPU_FEATS_NAME} PRIVATE ${ARCH_DEFINITIONS})
487
- target_compile_definitions(${GGML_CPU_FEATS_NAME} PRIVATE GGML_BACKEND_DL GGML_BACKEND_BUILD GGML_BACKEND_SHARED)
488
- set_target_properties(${GGML_CPU_FEATS_NAME} PROPERTIES POSITION_INDEPENDENT_CODE ON)
489
- target_link_libraries(${GGML_CPU_NAME} PRIVATE ${GGML_CPU_FEATS_NAME})
490
- endif()
491
-
492
491
  if (EMSCRIPTEN)
493
492
  set_target_properties(${GGML_CPU_NAME} PROPERTIES COMPILE_FLAGS "-msimd128")
494
493
  endif()
package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-aarch64.cpp CHANGED
@@ -1191,7 +1191,7 @@ static void ggml_gemv_q4_0_8x8_q8_0(int n, float * GGML_RESTRICT s, size_t bs, c
1191
1191
  }
1192
1192
  }
1193
1193
  return;
1194
- #elif defined(__riscv_v_intrinsic)
1194
+ #elif defined __riscv_v
1195
1195
  if (__riscv_vlenb() >= QK4_0) {
1196
1196
  const size_t vl = QK4_0;
1197
1197
 
@@ -3783,7 +3783,7 @@ static void ggml_gemm_q4_0_8x8_q8_0(int n, float * GGML_RESTRICT s, size_t bs, c
3783
3783
  }
3784
3784
  return;
3785
3785
  }
3786
- #elif defined(__riscv_v_intrinsic)
3786
+ #elif defined __riscv_v
3787
3787
  if (__riscv_vlenb() >= QK4_0) {
3788
3788
  const size_t vl = QK4_0;
3789
3789
 
package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-impl.h CHANGED
@@ -320,21 +320,17 @@ inline static int32x4_t ggml_vdotq_s32(int32x4_t acc, int8x16_t a, int8x16_t b)
320
320
 
321
321
  #ifdef __wasm_simd128__
322
322
  #include <wasm_simd128.h>
323
- #else
323
+ #endif
324
+
324
325
  #ifdef __POWER9_VECTOR__
325
326
  #include <altivec.h>
326
- #else
327
+ #endif
328
+
327
329
  #if defined(_MSC_VER) || defined(__MINGW32__)
328
330
  #include <intrin.h>
329
- #else
330
- #if defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__) || defined(__SSSE3__) || defined(__SSE3__) || defined(__SSE__)
331
- #if !defined(__riscv)
331
+ #elif defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__) || defined(__SSSE3__) || defined(__SSE3__) || defined(__SSE__)
332
332
  #include <immintrin.h>
333
333
  #endif
334
- #endif
335
- #endif
336
- #endif
337
- #endif
338
334
 
339
335
  #ifdef __riscv_v_intrinsic
340
336
  #include <riscv_vector.h>