@novastera-oss/llamarn 0.2.1 → 0.2.3

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 (268) 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/PureCppImpl.cpp +9 -27
  14. package/cpp/SystemUtils.h +2 -2
  15. package/cpp/build-info.cpp +2 -2
  16. package/cpp/llama.cpp/README.md +11 -3
  17. package/cpp/llama.cpp/build-xcframework.sh +1 -0
  18. package/cpp/llama.cpp/common/CMakeLists.txt +8 -2
  19. package/cpp/llama.cpp/common/arg.cpp +153 -113
  20. package/cpp/llama.cpp/common/chat-parser.cpp +379 -0
  21. package/cpp/llama.cpp/common/chat-parser.h +117 -0
  22. package/cpp/llama.cpp/common/chat.cpp +847 -699
  23. package/cpp/llama.cpp/common/chat.h +73 -6
  24. package/cpp/llama.cpp/common/common.cpp +50 -82
  25. package/cpp/llama.cpp/common/common.h +21 -17
  26. package/cpp/llama.cpp/common/json-partial.cpp +255 -0
  27. package/cpp/llama.cpp/common/json-partial.h +37 -0
  28. package/cpp/llama.cpp/common/minja/chat-template.hpp +9 -5
  29. package/cpp/llama.cpp/common/minja/minja.hpp +69 -36
  30. package/cpp/llama.cpp/common/regex-partial.cpp +204 -0
  31. package/cpp/llama.cpp/common/regex-partial.h +56 -0
  32. package/cpp/llama.cpp/common/sampling.cpp +7 -8
  33. package/cpp/llama.cpp/convert_hf_to_gguf.py +453 -118
  34. package/cpp/llama.cpp/convert_hf_to_gguf_update.py +120 -68
  35. package/cpp/llama.cpp/ggml/CMakeLists.txt +2 -1
  36. package/cpp/llama.cpp/ggml/cmake/common.cmake +25 -0
  37. package/cpp/llama.cpp/ggml/include/ggml-opt.h +49 -28
  38. package/cpp/llama.cpp/ggml/include/ggml.h +26 -7
  39. package/cpp/llama.cpp/ggml/src/CMakeLists.txt +16 -10
  40. package/cpp/llama.cpp/ggml/src/ggml-backend.cpp +4 -1
  41. package/cpp/llama.cpp/ggml/src/ggml-cann/CMakeLists.txt +1 -0
  42. package/cpp/llama.cpp/ggml/src/ggml-cann/acl_tensor.cpp +2 -0
  43. package/cpp/llama.cpp/ggml/src/ggml-cann/aclnn_ops.cpp +604 -0
  44. package/cpp/llama.cpp/ggml/src/ggml-cann/aclnn_ops.h +42 -0
  45. package/cpp/llama.cpp/ggml/src/ggml-cann/ggml-cann.cpp +54 -2
  46. package/cpp/llama.cpp/ggml/src/ggml-cpu/CMakeLists.txt +50 -51
  47. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-aarch64.cpp +2 -2
  48. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-impl.h +5 -9
  49. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-quants.c +779 -19
  50. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.c +22 -0
  51. package/cpp/llama.cpp/ggml/src/ggml-cpu/kleidiai/kernels.cpp +88 -5
  52. package/cpp/llama.cpp/ggml/src/ggml-cpu/kleidiai/kernels.h +47 -12
  53. package/cpp/llama.cpp/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +264 -69
  54. package/cpp/llama.cpp/ggml/src/ggml-cpu/ops.cpp +322 -100
  55. package/cpp/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h +117 -1
  56. package/cpp/llama.cpp/ggml/src/ggml-cpu/vec.cpp +85 -16
  57. package/cpp/llama.cpp/ggml/src/ggml-cpu/vec.h +220 -49
  58. package/cpp/llama.cpp/ggml/src/ggml-cuda/acc.cu +40 -26
  59. package/cpp/llama.cpp/ggml/src/ggml-cuda/common.cuh +1 -1
  60. package/cpp/llama.cpp/ggml/src/ggml-cuda/cpy.cu +11 -1
  61. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-common.cuh +15 -7
  62. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-mma-f16.cuh +266 -64
  63. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-vec-f16.cuh +49 -4
  64. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-vec-f32.cuh +48 -4
  65. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn.cu +2 -1
  66. package/cpp/llama.cpp/ggml/src/ggml-cuda/ggml-cuda.cu +5 -1
  67. package/cpp/llama.cpp/ggml/src/ggml-cuda/mmq.cu +2 -0
  68. package/cpp/llama.cpp/ggml/src/ggml-cuda/quantize.cu +7 -6
  69. package/cpp/llama.cpp/ggml/src/ggml-cuda/sum.cu +1 -1
  70. package/cpp/llama.cpp/ggml/src/ggml-cuda/unary.cu +10 -0
  71. package/cpp/llama.cpp/ggml/src/ggml-cuda/unary.cuh +2 -0
  72. package/cpp/llama.cpp/ggml/src/ggml-impl.h +1 -1
  73. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-impl.h +4 -0
  74. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal.m +99 -17
  75. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal.metal +200 -2
  76. package/cpp/llama.cpp/ggml/src/ggml-musa/CMakeLists.txt +8 -2
  77. package/cpp/llama.cpp/ggml/src/ggml-musa/mudnn.cu +112 -0
  78. package/cpp/llama.cpp/ggml/src/ggml-musa/mudnn.cuh +12 -0
  79. package/cpp/llama.cpp/ggml/src/ggml-opencl/CMakeLists.txt +6 -0
  80. package/cpp/llama.cpp/ggml/src/ggml-opencl/ggml-opencl.cpp +972 -178
  81. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/argsort.cl +86 -0
  82. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/div.cl +72 -0
  83. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/group_norm.cl +72 -0
  84. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/sigmoid.cl +29 -0
  85. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/sub.cl +72 -0
  86. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/sum_rows.cl +39 -0
  87. package/cpp/llama.cpp/ggml/src/ggml-opt.cpp +373 -190
  88. package/cpp/llama.cpp/ggml/src/ggml-sycl/CMakeLists.txt +29 -23
  89. package/cpp/llama.cpp/ggml/src/ggml-sycl/binbcast.cpp +5 -10
  90. package/cpp/llama.cpp/ggml/src/ggml-sycl/common.hpp +101 -5
  91. package/cpp/llama.cpp/ggml/src/ggml-sycl/concat.cpp +31 -33
  92. package/cpp/llama.cpp/ggml/src/ggml-sycl/conv.cpp +1 -0
  93. package/cpp/llama.cpp/ggml/src/ggml-sycl/convert.cpp +29 -2
  94. package/cpp/llama.cpp/ggml/src/ggml-sycl/cpy.cpp +4 -5
  95. package/cpp/llama.cpp/ggml/src/ggml-sycl/dequantize.hpp +59 -21
  96. package/cpp/llama.cpp/ggml/src/ggml-sycl/dmmv.cpp +9 -1
  97. package/cpp/llama.cpp/ggml/src/ggml-sycl/element_wise.cpp +84 -72
  98. package/cpp/llama.cpp/ggml/src/ggml-sycl/element_wise.hpp +2 -0
  99. package/cpp/llama.cpp/ggml/src/ggml-sycl/gemm.hpp +37 -8
  100. package/cpp/llama.cpp/ggml/src/ggml-sycl/getrows.cpp +1 -3
  101. package/cpp/llama.cpp/ggml/src/ggml-sycl/ggml-sycl.cpp +324 -129
  102. package/cpp/llama.cpp/ggml/src/ggml-sycl/gla.cpp +1 -0
  103. package/cpp/llama.cpp/ggml/src/ggml-sycl/mmvq.cpp +31 -2
  104. package/cpp/llama.cpp/ggml/src/ggml-sycl/norm.cpp +95 -68
  105. package/cpp/llama.cpp/ggml/src/ggml-sycl/outprod.cpp +1 -0
  106. package/cpp/llama.cpp/ggml/src/ggml-sycl/quants.hpp +22 -0
  107. package/cpp/llama.cpp/ggml/src/ggml-sycl/rope.cpp +1 -2
  108. package/cpp/llama.cpp/ggml/src/ggml-sycl/softmax.cpp +1 -4
  109. package/cpp/llama.cpp/ggml/src/ggml-sycl/tsembd.cpp +2 -3
  110. package/cpp/llama.cpp/ggml/src/ggml-sycl/vecdotq.hpp +69 -43
  111. package/cpp/llama.cpp/ggml/src/ggml-sycl/wkv.cpp +2 -14
  112. package/cpp/llama.cpp/ggml/src/ggml-vulkan/CMakeLists.txt +81 -91
  113. package/cpp/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp +432 -181
  114. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/CMakeLists.txt +17 -0
  115. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq1_m.comp +1 -1
  116. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +6 -152
  117. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.comp +162 -0
  118. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +360 -0
  119. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +2 -118
  120. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +1 -1
  121. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +12 -1
  122. package/cpp/llama.cpp/ggml/src/ggml.c +107 -36
  123. package/cpp/llama.cpp/ggml/src/gguf.cpp +33 -33
  124. package/cpp/llama.cpp/gguf-py/gguf/constants.py +100 -15
  125. package/cpp/llama.cpp/gguf-py/gguf/gguf_reader.py +1 -1
  126. package/cpp/llama.cpp/gguf-py/gguf/gguf_writer.py +44 -12
  127. package/cpp/llama.cpp/gguf-py/gguf/scripts/gguf_editor_gui.py +21 -10
  128. package/cpp/llama.cpp/gguf-py/gguf/scripts/gguf_new_metadata.py +5 -2
  129. package/cpp/llama.cpp/gguf-py/gguf/tensor_mapping.py +128 -31
  130. package/cpp/llama.cpp/gguf-py/gguf/utility.py +1 -1
  131. package/cpp/llama.cpp/gguf-py/pyproject.toml +1 -1
  132. package/cpp/llama.cpp/include/llama.h +62 -125
  133. package/cpp/llama.cpp/models/ggml-vocab-bert-bge.gguf.inp +1 -1
  134. package/cpp/llama.cpp/models/ggml-vocab-bert-bge.gguf.out +1 -1
  135. package/cpp/llama.cpp/models/ggml-vocab-command-r.gguf.inp +1 -1
  136. package/cpp/llama.cpp/models/ggml-vocab-command-r.gguf.out +1 -1
  137. package/cpp/llama.cpp/models/ggml-vocab-deepseek-coder.gguf.inp +1 -1
  138. package/cpp/llama.cpp/models/ggml-vocab-deepseek-coder.gguf.out +1 -1
  139. package/cpp/llama.cpp/models/ggml-vocab-deepseek-llm.gguf.inp +1 -1
  140. package/cpp/llama.cpp/models/ggml-vocab-deepseek-llm.gguf.out +1 -1
  141. package/cpp/llama.cpp/models/ggml-vocab-falcon.gguf.inp +1 -1
  142. package/cpp/llama.cpp/models/ggml-vocab-falcon.gguf.out +1 -1
  143. package/cpp/llama.cpp/models/ggml-vocab-gpt-2.gguf.inp +1 -1
  144. package/cpp/llama.cpp/models/ggml-vocab-gpt-2.gguf.out +1 -1
  145. package/cpp/llama.cpp/models/ggml-vocab-llama-bpe.gguf.inp +1 -1
  146. package/cpp/llama.cpp/models/ggml-vocab-llama-bpe.gguf.out +1 -1
  147. package/cpp/llama.cpp/models/ggml-vocab-llama-spm.gguf.inp +1 -1
  148. package/cpp/llama.cpp/models/ggml-vocab-llama-spm.gguf.out +1 -1
  149. package/cpp/llama.cpp/models/ggml-vocab-mpt.gguf.inp +1 -1
  150. package/cpp/llama.cpp/models/ggml-vocab-mpt.gguf.out +1 -1
  151. package/cpp/llama.cpp/models/ggml-vocab-nomic-bert-moe.gguf +0 -0
  152. package/cpp/llama.cpp/models/ggml-vocab-phi-3.gguf.inp +1 -1
  153. package/cpp/llama.cpp/models/ggml-vocab-phi-3.gguf.out +1 -1
  154. package/cpp/llama.cpp/models/ggml-vocab-qwen2.gguf.inp +1 -1
  155. package/cpp/llama.cpp/models/ggml-vocab-qwen2.gguf.out +1 -1
  156. package/cpp/llama.cpp/models/ggml-vocab-refact.gguf.inp +1 -1
  157. package/cpp/llama.cpp/models/ggml-vocab-refact.gguf.out +1 -1
  158. package/cpp/llama.cpp/models/ggml-vocab-starcoder.gguf.inp +1 -1
  159. package/cpp/llama.cpp/models/ggml-vocab-starcoder.gguf.out +1 -1
  160. package/cpp/llama.cpp/models/templates/Qwen-QwQ-32B.jinja +62 -0
  161. package/cpp/llama.cpp/models/templates/Qwen-Qwen3-0.6B.jinja +85 -0
  162. package/cpp/llama.cpp/models/templates/README.md +2 -0
  163. package/cpp/llama.cpp/requirements/requirements-convert_hf_to_gguf.txt +5 -1
  164. package/cpp/llama.cpp/requirements/requirements-convert_hf_to_gguf_update.txt +5 -1
  165. package/cpp/llama.cpp/requirements/requirements-convert_lora_to_gguf.txt +2 -0
  166. package/cpp/llama.cpp/requirements/requirements-gguf_editor_gui.txt +1 -1
  167. package/cpp/llama.cpp/src/CMakeLists.txt +2 -0
  168. package/cpp/llama.cpp/src/llama-arch.cpp +6 -0
  169. package/cpp/llama.cpp/src/llama-arch.h +2 -0
  170. package/cpp/llama.cpp/src/llama-batch.cpp +3 -1
  171. package/cpp/llama.cpp/src/llama-context.cpp +340 -123
  172. package/cpp/llama.cpp/src/llama-context.h +30 -0
  173. package/cpp/llama.cpp/src/llama-cparams.cpp +4 -0
  174. package/cpp/llama.cpp/src/llama-cparams.h +2 -0
  175. package/cpp/llama.cpp/src/llama-grammar.cpp +12 -2
  176. package/cpp/llama.cpp/src/llama-graph.cpp +157 -247
  177. package/cpp/llama.cpp/src/llama-graph.h +52 -7
  178. package/cpp/llama.cpp/src/llama-hparams.cpp +17 -1
  179. package/cpp/llama.cpp/src/llama-hparams.h +37 -5
  180. package/cpp/llama.cpp/src/llama-kv-cache.cpp +742 -481
  181. package/cpp/llama.cpp/src/llama-kv-cache.h +196 -99
  182. package/cpp/llama.cpp/src/llama-kv-cells.h +379 -0
  183. package/cpp/llama.cpp/src/llama-memory.h +4 -3
  184. package/cpp/llama.cpp/src/llama-model-loader.cpp +22 -17
  185. package/cpp/llama.cpp/src/llama-model-saver.cpp +281 -0
  186. package/cpp/llama.cpp/src/llama-model-saver.h +37 -0
  187. package/cpp/llama.cpp/src/llama-model.cpp +529 -172
  188. package/cpp/llama.cpp/src/llama-model.h +6 -1
  189. package/cpp/llama.cpp/src/llama-quant.cpp +15 -13
  190. package/cpp/llama.cpp/src/llama-sampling.cpp +2 -2
  191. package/cpp/llama.cpp/src/llama-vocab.cpp +35 -8
  192. package/cpp/llama.cpp/src/llama-vocab.h +6 -0
  193. package/cpp/llama.cpp/src/llama.cpp +14 -0
  194. package/cpp/rn-completion.cpp +60 -5
  195. package/ios/include/chat.h +73 -6
  196. package/ios/include/common/minja/chat-template.hpp +9 -5
  197. package/ios/include/common/minja/minja.hpp +69 -36
  198. package/ios/include/common.h +21 -17
  199. package/ios/include/llama.h +62 -125
  200. package/ios/libs/llama.xcframework/Info.plist +19 -19
  201. package/ios/libs/llama.xcframework/ios-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  202. package/ios/libs/llama.xcframework/ios-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4617 -4487
  203. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/ggml-opt.h +237 -0
  204. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/ggml.h +26 -7
  205. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/llama.h +62 -125
  206. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/llama +0 -0
  207. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  208. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4638 -4508
  209. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3557 -3435
  210. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml-opt.h +237 -0
  211. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +26 -7
  212. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/llama.h +62 -125
  213. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/llama +0 -0
  214. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  215. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4638 -4508
  216. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3559 -3437
  217. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml-opt.h +237 -0
  218. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml.h +26 -7
  219. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/llama.h +62 -125
  220. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml-opt.h +237 -0
  221. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml.h +26 -7
  222. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/llama.h +62 -125
  223. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/llama +0 -0
  224. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml-opt.h +237 -0
  225. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml.h +26 -7
  226. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/llama.h +62 -125
  227. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/llama +0 -0
  228. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/llama +0 -0
  229. package/ios/libs/llama.xcframework/tvos-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  230. package/ios/libs/llama.xcframework/tvos-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4616 -4487
  231. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml-opt.h +237 -0
  232. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml.h +26 -7
  233. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/llama.h +62 -125
  234. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/llama +0 -0
  235. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  236. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4637 -4508
  237. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3556 -3435
  238. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml-opt.h +237 -0
  239. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +26 -7
  240. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/llama.h +62 -125
  241. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/llama +0 -0
  242. package/ios/libs/llama.xcframework/xros-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  243. package/ios/libs/llama.xcframework/xros-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4653 -4523
  244. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml-opt.h +237 -0
  245. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml.h +26 -7
  246. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/llama.h +62 -125
  247. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/llama +0 -0
  248. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  249. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +4674 -4544
  250. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3587 -3465
  251. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml-opt.h +237 -0
  252. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +26 -7
  253. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/llama.h +62 -125
  254. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/llama +0 -0
  255. package/package.json +1 -1
  256. package/cpp/llama.cpp/common/stb_image.h +0 -7988
  257. package/cpp/llama.cpp/models/ggml-vocab-chameleon.gguf.inp +0 -112
  258. package/cpp/llama.cpp/models/ggml-vocab-chameleon.gguf.out +0 -46
  259. package/cpp/llama.cpp/models/ggml-vocab-deepseek-r1-qwen.gguf.inp +0 -112
  260. package/cpp/llama.cpp/models/ggml-vocab-deepseek-r1-qwen.gguf.out +0 -46
  261. package/cpp/llama.cpp/models/ggml-vocab-gpt-4o.gguf.inp +0 -112
  262. package/cpp/llama.cpp/models/ggml-vocab-gpt-4o.gguf.out +0 -46
  263. package/cpp/llama.cpp/models/ggml-vocab-llama4.gguf.inp +0 -112
  264. package/cpp/llama.cpp/models/ggml-vocab-llama4.gguf.out +0 -46
  265. package/cpp/llama.cpp/models/ggml-vocab-pixtral.gguf.inp +0 -112
  266. package/cpp/llama.cpp/models/ggml-vocab-pixtral.gguf.out +0 -46
  267. package/cpp/llama.cpp/models/ggml-vocab-roberta-bpe.gguf.inp +0 -112
  268. package/cpp/llama.cpp/models/ggml-vocab-roberta-bpe.gguf.out +0 -46
package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/argsort.cl ADDED
@@ -0,0 +1,86 @@
1
+ #pragma OPENCL EXTENSION cl_khr_fp16 : enable
2
+
3
+ #ifdef cl_intel_subgroups
4
+ #pragma OPENCL EXTENSION cl_intel_subgroups : enable
5
+ #else
6
+ #pragma OPENCL EXTENSION cl_khr_subgroups : enable
7
+ #endif
8
+
9
+ #ifdef cl_intel_required_subgroup_size
10
+ #pragma OPENCL EXTENSION cl_intel_required_subgroup_size : enable
11
+ #define INTEL_GPU 1
12
+ #define REQD_SUBGROUP_SIZE_16 __attribute__((intel_reqd_sub_group_size(16)))
13
+ #define REQD_SUBGROUP_SIZE_32 __attribute__((intel_reqd_sub_group_size(32)))
14
+ #elif defined(cl_qcom_reqd_sub_group_size)
15
+ #pragma OPENCL EXTENSION cl_qcom_reqd_sub_group_size : enable
16
+ #define ADRENO_GPU 1
17
+ #define REQD_SUBGROUP_SIZE_64 __attribute__((qcom_reqd_sub_group_size("half")))
18
+ #define REQD_SUBGROUP_SIZE_128 __attribute__((qcom_reqd_sub_group_size("full")))
19
+ #endif
20
+
21
+ #define SWAP(x, y, T) { T tmp = (x); (x) = (y); (y) = tmp; }
22
+
23
+ enum ggml_sort_order {
24
+ GGML_SORT_ORDER_ASC,
25
+ GGML_SORT_ORDER_DESC,
26
+ };
27
+
28
+ kernel void kernel_argsort_f32_i32(
29
+ global float * src0,
30
+ ulong offset0,
31
+ global int * dst,
32
+ ulong offsetd,
33
+ const int ne00,
34
+ const int ne00_pad,
35
+ const int order,
36
+ local int * dst_row
37
+ ) {
38
+ // bitonic sort
39
+ int col = get_local_id(0);
40
+ int row = get_group_id(1);
41
+
42
+ if (col >= ne00_pad) {
43
+ return;
44
+ }
45
+
46
+ src0 = (global char *)((global char *)src0 + offset0);
47
+ dst = (global float *)((global char *)dst + offsetd);
48
+
49
+ global float * x_row = src0 + row * ne00;
50
+
51
+ // initialize indices
52
+ dst_row[col] = col;
53
+
54
+ barrier(CLK_LOCAL_MEM_FENCE);
55
+
56
+ for (int k = 2; k <= ne00_pad; k *= 2) {
57
+ for (int j = k / 2; j > 0; j /= 2) {
58
+ int ixj = col ^ j;
59
+ if (ixj > col) {
60
+ if ((col & k) == 0) {
61
+ if (dst_row[col] >= ne00 ||
62
+ (dst_row[ixj] < ne00 && (order == GGML_SORT_ORDER_ASC ?
63
+ x_row[dst_row[col]] > x_row[dst_row[ixj]] :
64
+ x_row[dst_row[col]] < x_row[dst_row[ixj]]))
65
+ ) {
66
+ SWAP(dst_row[col], dst_row[ixj], int);
67
+ }
68
+ } else {
69
+ if (dst_row[ixj] >= ne00 ||
70
+ (dst_row[col] < ne00 && (order == GGML_SORT_ORDER_ASC ?
71
+ x_row[dst_row[col]] < x_row[dst_row[ixj]] :
72
+ x_row[dst_row[col]] > x_row[dst_row[ixj]]))
73
+ ) {
74
+ SWAP(dst_row[col], dst_row[ixj], int);
75
+ }
76
+ }
77
+ }
78
+ barrier(CLK_LOCAL_MEM_FENCE);
79
+ }
80
+ }
81
+
82
+ // copy the result to dst without the padding
83
+ if (col < ne00) {
84
+ dst[row * ne00 + col] = dst_row[col];
85
+ }
86
+ }
package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/div.cl ADDED
@@ -0,0 +1,72 @@
1
+ #pragma OPENCL EXTENSION cl_khr_fp16 : enable
2
+
3
+ //------------------------------------------------------------------------------
4
+ // div
5
+ //------------------------------------------------------------------------------
6
+ kernel void kernel_div(
7
+ global char * src0,
8
+ ulong offset0,
9
+ global char * src1,
10
+ ulong offset1,
11
+ global char * dst,
12
+ ulong offsetd,
13
+ ulong nb00,
14
+ ulong nb01,
15
+ ulong nb02,
16
+ ulong nb03,
17
+ int ne10,
18
+ int ne11,
19
+ int ne12,
20
+ int ne13,
21
+ ulong nb10,
22
+ ulong nb11,
23
+ ulong nb12,
24
+ ulong nb13,
25
+ int ne0,
26
+ ulong nb0,
27
+ ulong nb1,
28
+ ulong nb2,
29
+ ulong nb3
30
+ ) {
31
+ src0 = src0 + offset0;
32
+ src1 = src1 + offset1;
33
+ dst = dst + offsetd;
34
+
35
+ int i03 = get_group_id(2);
36
+ int i02 = get_group_id(1);
37
+ int i01 = get_group_id(0);
38
+
39
+ int i13 = i03 % ne13;
40
+ int i12 = i02 % ne12;
41
+ int i11 = i01 % ne11;
42
+
43
+ global char * src0_ptr = src0 + i03*nb03 + i02*nb02 + i01*nb01;
44
+ global char * src1_ptr = src1 + i13*nb13 + i12*nb12 + i11*nb11;
45
+ global char * dst_ptr = dst + i03*nb3 + i02*nb2 + i01*nb1;
46
+
47
+ for (int i0 = get_local_id(0); i0 < ne0; i0 += get_local_size(0)) {
48
+ const int i10 = i0 % ne10;
49
+ *((global float *)(dst_ptr + i0*nb0)) = *((global float *)(src0_ptr + i0*nb00)) / *((global float *)(src1_ptr + i10*nb10));
50
+ }
51
+ }
52
+
53
+ // assumption: src1 is a row
54
+ // broadcast src1 into src0
55
+ kernel void kernel_div_row(
56
+ global float4 * src0,
57
+ ulong offset0,
58
+ global float4 * src1,
59
+ ulong offset1,
60
+ global float4 * dst,
61
+ ulong offsetd,
62
+ int ne
63
+ ) {
64
+ src0 = (global float4*)((global char*)src0 + offset0);
65
+ src1 = (global float4*)((global char*)src1 + offset1);
66
+ dst = (global float4*)((global char*)dst + offsetd);
67
+
68
+ // This performs better than using %.
69
+ uint gid = get_global_id(0);
70
+ uint idx1 = gid - (gid/ne)*ne; // get_global_id(0) % ne
71
+ dst[gid] = src0[gid] / src1[idx1];
72
+ }
package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/group_norm.cl ADDED
@@ -0,0 +1,72 @@
1
+ #pragma OPENCL EXTENSION cl_khr_fp16 : enable
2
+
3
+ #ifdef cl_intel_subgroups
4
+ #pragma OPENCL EXTENSION cl_intel_subgroups : enable
5
+ #else
6
+ #pragma OPENCL EXTENSION cl_khr_subgroups : enable
7
+ #endif
8
+
9
+ #ifdef cl_intel_required_subgroup_size
10
+ #pragma OPENCL EXTENSION cl_intel_required_subgroup_size : enable
11
+ #define INTEL_GPU 1
12
+ #define REQD_SUBGROUP_SIZE_16 __attribute__((intel_reqd_sub_group_size(16)))
13
+ #define REQD_SUBGROUP_SIZE_32 __attribute__((intel_reqd_sub_group_size(32)))
14
+ #elif defined(cl_qcom_reqd_sub_group_size)
15
+ #pragma OPENCL EXTENSION cl_qcom_reqd_sub_group_size : enable
16
+ #define ADRENO_GPU 1
17
+ #define REQD_SUBGROUP_SIZE_64 __attribute__((qcom_reqd_sub_group_size("half")))
18
+ #define REQD_SUBGROUP_SIZE_128 __attribute__((qcom_reqd_sub_group_size("full")))
19
+ #endif
20
+
21
+ // Workgroup must be a subgroup
22
+ #ifdef INTEL_GPU
23
+ REQD_SUBGROUP_SIZE_32
24
+ #elif defined (ADRENO_GPU)
25
+ REQD_SUBGROUP_SIZE_64
26
+ #endif
27
+ kernel void kernel_group_norm(
28
+ global float * src0,
29
+ ulong offset0,
30
+ global float * dst,
31
+ ulong offsetd,
32
+ int ne,
33
+ int group_size,
34
+ float eps
35
+ ) {
36
+ src0 = (global float *)((global char *)src0 + offset0);
37
+ dst = (global float *)((global char *)dst + offsetd);
38
+
39
+ int start = get_group_id(0) * group_size;
40
+ int end = start + group_size;
41
+
42
+ start += get_local_id(0);
43
+
44
+ if (end >= ne) {
45
+ end = ne;
46
+ }
47
+
48
+ float tmp = 0.0f;
49
+
50
+ for (int j = start; j < end; j += get_local_size(0)) {
51
+ tmp += src0[j];
52
+ }
53
+
54
+ tmp = sub_group_reduce_add(tmp);
55
+
56
+ const float mean = tmp / group_size;
57
+ tmp = 0.0f;
58
+
59
+ for (int j = start; j < end; j += get_local_size(0)) {
60
+ float xi = src0[j] - mean;
61
+ dst[j] = xi;
62
+ tmp += xi * xi;
63
+ }
64
+
65
+ tmp = sub_group_reduce_add(tmp);
66
+
67
+ const float variance = tmp / group_size;
68
+ const float scale = 1.0f/sqrt(variance + eps);
69
+ for (int j = start; j < end; j += get_local_size(0)) {
70
+ dst[j] *= scale;
71
+ }
72
+ }
package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/sigmoid.cl ADDED
@@ -0,0 +1,29 @@
1
+ #pragma OPENCL EXTENSION cl_khr_fp16 : enable
2
+
3
+ //------------------------------------------------------------------------------
4
+ // sigmoid
5
+ //------------------------------------------------------------------------------
6
+
7
+ kernel void kernel_sigmoid_f32(
8
+ global float * src0,
9
+ ulong offset0,
10
+ global float * dst,
11
+ ulong offsetd
12
+ ) {
13
+ src0 = (global float*)((global char*)src0 + offset0);
14
+ dst = (global float*)((global char*)dst + offsetd);
15
+
16
+ dst[get_global_id(0)] = 1.0f / (1.0f + exp(-src0[get_global_id(0)]));
17
+ }
18
+
19
+ kernel void kernel_sigmoid_f16(
20
+ global half * src0,
21
+ ulong offset0,
22
+ global half * dst,
23
+ ulong offsetd
24
+ ) {
25
+ src0 = (global half*)((global char*)src0 + offset0);
26
+ dst = (global half*)((global char*)dst + offsetd);
27
+
28
+ dst[get_global_id(0)] = 1.0f / (1.0f + exp(-src0[get_global_id(0)]));
29
+ }
package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/sub.cl ADDED
@@ -0,0 +1,72 @@
1
+ #pragma OPENCL EXTENSION cl_khr_fp16 : enable
2
+
3
+ //------------------------------------------------------------------------------
4
+ // div
5
+ //------------------------------------------------------------------------------
6
+ kernel void kernel_sub(
7
+ global char * src0,
8
+ ulong offset0,
9
+ global char * src1,
10
+ ulong offset1,
11
+ global char * dst,
12
+ ulong offsetd,
13
+ ulong nb00,
14
+ ulong nb01,
15
+ ulong nb02,
16
+ ulong nb03,
17
+ int ne10,
18
+ int ne11,
19
+ int ne12,
20
+ int ne13,
21
+ ulong nb10,
22
+ ulong nb11,
23
+ ulong nb12,
24
+ ulong nb13,
25
+ int ne0,
26
+ ulong nb0,
27
+ ulong nb1,
28
+ ulong nb2,
29
+ ulong nb3
30
+ ) {
31
+ src0 = src0 + offset0;
32
+ src1 = src1 + offset1;
33
+ dst = dst + offsetd;
34
+
35
+ int i03 = get_group_id(2);
36
+ int i02 = get_group_id(1);
37
+ int i01 = get_group_id(0);
38
+
39
+ int i13 = i03 % ne13;
40
+ int i12 = i02 % ne12;
41
+ int i11 = i01 % ne11;
42
+
43
+ global char * src0_ptr = src0 + i03*nb03 + i02*nb02 + i01*nb01;
44
+ global char * src1_ptr = src1 + i13*nb13 + i12*nb12 + i11*nb11;
45
+ global char * dst_ptr = dst + i03*nb3 + i02*nb2 + i01*nb1;
46
+
47
+ for (int i0 = get_local_id(0); i0 < ne0; i0 += get_local_size(0)) {
48
+ const int i10 = i0 % ne10;
49
+ *((global float *)(dst_ptr + i0*nb0)) = *((global float *)(src0_ptr + i0*nb00)) - *((global float *)(src1_ptr + i10*nb10));
50
+ }
51
+ }
52
+
53
+ // assumption: src1 is a row
54
+ // broadcast src1 into src0
55
+ kernel void kernel_sub_row(
56
+ global float4 * src0,
57
+ ulong offset0,
58
+ global float4 * src1,
59
+ ulong offset1,
60
+ global float4 * dst,
61
+ ulong offsetd,
62
+ int ne
63
+ ) {
64
+ src0 = (global float4*)((global char*)src0 + offset0);
65
+ src1 = (global float4*)((global char*)src1 + offset1);
66
+ dst = (global float4*)((global char*)dst + offsetd);
67
+
68
+ // This performs better than using %.
69
+ uint gid = get_global_id(0);
70
+ uint idx1 = gid - (gid/ne)*ne; // get_global_id(0) % ne
71
+ dst[gid] = src0[gid] - src1[idx1];
72
+ }
package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/sum_rows.cl ADDED
@@ -0,0 +1,39 @@
1
+
2
+ kernel void kernel_sum_rows_f32(
3
+ global float * src0,
4
+ ulong offset0,
5
+ global float * dst,
6
+ ulong offsetd,
7
+ int ne00,
8
+ int ne01,
9
+ int ne02,
10
+ int ne03,
11
+ ulong nb01,
12
+ ulong nb02,
13
+ ulong nb03,
14
+ ulong nb1,
15
+ ulong nb2,
16
+ ulong nb3
17
+ ) {
18
+ src0 = (global float *)((global char *)src0 + offset0);
19
+ dst = (global float *)((global char *)dst + offsetd);
20
+
21
+ int i3 = get_global_id(2);
22
+ int i2 = get_global_id(1);
23
+ int i1 = get_global_id(0);
24
+
25
+ if (i3 >= ne03 || i2 >= ne02 || i1 >= ne01) {
26
+ return;
27
+ }
28
+
29
+ global float * src_row = (global float *) ((global char *) src0 + i1*nb01 + i2*nb02 + i3*nb03);
30
+ global float * dst_row = (global float *) ((global char *) dst + i1*nb1 + i2*nb2 + i3*nb3);
31
+
32
+ float row_sum = 0;
33
+
34
+ for (int i0 = 0; i0 < ne00; i0++) {
35
+ row_sum += src_row[i0];
36
+ }
37
+
38
+ dst_row[0] = row_sum;
39
+ }