@novastera-oss/llamarn 0.2.9 → 0.3.0

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 (247) hide show
  1. package/android/src/main/jniLibs/arm64-v8a/libggml-base.so +0 -0
  2. package/android/src/main/jniLibs/arm64-v8a/libggml-cpu.so +0 -0
  3. package/android/src/main/jniLibs/arm64-v8a/libggml.so +0 -0
  4. package/android/src/main/jniLibs/arm64-v8a/libllama.so +0 -0
  5. package/android/src/main/jniLibs/armeabi-v7a/libggml-base.so +0 -0
  6. package/android/src/main/jniLibs/armeabi-v7a/libggml-cpu.so +0 -0
  7. package/android/src/main/jniLibs/armeabi-v7a/libggml.so +0 -0
  8. package/android/src/main/jniLibs/armeabi-v7a/libllama.so +0 -0
  9. package/android/src/main/jniLibs/x86/libggml-base.so +0 -0
  10. package/android/src/main/jniLibs/x86/libggml-cpu.so +0 -0
  11. package/android/src/main/jniLibs/x86/libggml.so +0 -0
  12. package/android/src/main/jniLibs/x86/libllama.so +0 -0
  13. package/android/src/main/jniLibs/x86_64/libggml-base.so +0 -0
  14. package/android/src/main/jniLibs/x86_64/libggml-cpu.so +0 -0
  15. package/android/src/main/jniLibs/x86_64/libggml.so +0 -0
  16. package/android/src/main/jniLibs/x86_64/libllama.so +0 -0
  17. package/cpp/build-info.cpp +2 -2
  18. package/cpp/llama.cpp/CMakeLists.txt +0 -1
  19. package/cpp/llama.cpp/README.md +4 -5
  20. package/cpp/llama.cpp/build-xcframework.sh +1 -1
  21. package/cpp/llama.cpp/common/CMakeLists.txt +4 -5
  22. package/cpp/llama.cpp/common/arg.cpp +17 -0
  23. package/cpp/llama.cpp/common/chat.cpp +37 -20
  24. package/cpp/llama.cpp/common/chat.h +2 -0
  25. package/cpp/llama.cpp/common/common.h +4 -0
  26. package/cpp/llama.cpp/convert_hf_to_gguf.py +745 -6
  27. package/cpp/llama.cpp/convert_hf_to_gguf_update.py +9 -0
  28. package/cpp/llama.cpp/ggml/CMakeLists.txt +7 -2
  29. package/cpp/llama.cpp/ggml/include/ggml-backend.h +1 -1
  30. package/cpp/llama.cpp/ggml/include/ggml.h +173 -10
  31. package/cpp/llama.cpp/ggml/src/CMakeLists.txt +0 -1
  32. package/cpp/llama.cpp/ggml/src/ggml-backend-reg.cpp +0 -8
  33. package/cpp/llama.cpp/ggml/src/ggml-backend.cpp +36 -18
  34. package/cpp/llama.cpp/ggml/src/ggml-cann/aclnn_ops.cpp +68 -5
  35. package/cpp/llama.cpp/ggml/src/ggml-cann/ggml-cann.cpp +16 -2
  36. package/cpp/llama.cpp/ggml/src/ggml-cpu/CMakeLists.txt +6 -1
  37. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.c +28 -1
  38. package/cpp/llama.cpp/ggml/src/ggml-cpu/ops.cpp +1203 -163
  39. package/cpp/llama.cpp/ggml/src/ggml-cpu/ops.h +6 -0
  40. package/cpp/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h +1 -1
  41. package/cpp/llama.cpp/ggml/src/ggml-cpu/vec.cpp +33 -9
  42. package/cpp/llama.cpp/ggml/src/ggml-cpu/vec.h +142 -9
  43. package/cpp/llama.cpp/ggml/src/ggml-cuda/common.cuh +17 -0
  44. package/cpp/llama.cpp/ggml/src/ggml-cuda/convert.cu +22 -0
  45. package/cpp/llama.cpp/ggml/src/ggml-cuda/convert.cuh +5 -0
  46. package/cpp/llama.cpp/ggml/src/ggml-cuda/cross-entropy-loss.cu +2 -14
  47. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-common.cuh +4 -1
  48. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-mma-f16.cuh +8 -4
  49. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-tile-f16.cu +6 -4
  50. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-tile-f32.cu +14 -12
  51. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-vec-f16.cuh +5 -3
  52. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-vec-f32.cuh +15 -10
  53. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-wmma-f16.cu +8 -6
  54. package/cpp/llama.cpp/ggml/src/ggml-cuda/getrows.cu +8 -0
  55. package/cpp/llama.cpp/ggml/src/ggml-cuda/ggml-cuda.cu +185 -79
  56. package/cpp/llama.cpp/ggml/src/ggml-cuda/mmq.cuh +2 -8
  57. package/cpp/llama.cpp/ggml/src/ggml-cuda/rope.cu +21 -27
  58. package/cpp/llama.cpp/ggml/src/ggml-cuda/scale.cu +8 -6
  59. package/cpp/llama.cpp/ggml/src/ggml-cuda/softmax.cu +119 -58
  60. package/cpp/llama.cpp/ggml/src/ggml-cuda/ssm-conv.cu +10 -2
  61. package/cpp/llama.cpp/ggml/src/ggml-cuda/ssm-scan.cu +192 -52
  62. package/cpp/llama.cpp/ggml/src/ggml-cuda/unary.cu +97 -0
  63. package/cpp/llama.cpp/ggml/src/ggml-cuda/unary.cuh +11 -0
  64. package/cpp/llama.cpp/ggml/src/ggml-cuda/upscale.cu +92 -6
  65. package/cpp/llama.cpp/ggml/src/ggml-cuda/vendors/hip.h +14 -5
  66. package/cpp/llama.cpp/ggml/src/ggml-impl.h +64 -0
  67. package/cpp/llama.cpp/ggml/src/ggml-metal/CMakeLists.txt +4 -2
  68. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-impl.h +35 -9
  69. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal.m +167 -39
  70. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal.metal +254 -57
  71. package/cpp/llama.cpp/ggml/src/ggml-opencl/CMakeLists.txt +3 -0
  72. package/cpp/llama.cpp/ggml/src/ggml-opencl/ggml-opencl.cpp +505 -40
  73. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/gelu.cl +27 -0
  74. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/glu.cl +337 -0
  75. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mat_f16_f32.cl +130 -0
  76. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/scale.cl +3 -2
  77. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/set_rows.cl +95 -0
  78. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/softmax_4_f16.cl +24 -11
  79. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/softmax_4_f32.cl +24 -11
  80. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/softmax_f16.cl +24 -11
  81. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/softmax_f32.cl +24 -11
  82. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/upscale.cl +2 -3
  83. package/cpp/llama.cpp/ggml/src/ggml-quants.c +6 -6
  84. package/cpp/llama.cpp/ggml/src/ggml-sycl/backend.hpp +1 -0
  85. package/cpp/llama.cpp/ggml/src/ggml-sycl/element_wise.cpp +693 -1034
  86. package/cpp/llama.cpp/ggml/src/ggml-sycl/element_wise.hpp +18 -9
  87. package/cpp/llama.cpp/ggml/src/ggml-sycl/ggml-sycl.cpp +60 -9
  88. package/cpp/llama.cpp/ggml/src/ggml-sycl/rope.cpp +15 -18
  89. package/cpp/llama.cpp/ggml/src/ggml-sycl/set_rows.cpp +131 -0
  90. package/cpp/llama.cpp/ggml/src/ggml-sycl/set_rows.hpp +8 -0
  91. package/cpp/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp +711 -292
  92. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp +58 -7
  93. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +28 -23
  94. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_base.comp +14 -9
  95. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +38 -32
  96. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +32 -27
  97. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_split_k_reduce.comp +44 -12
  98. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/geglu.comp +13 -0
  99. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/geglu_erf.comp +27 -0
  100. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/geglu_quick.comp +11 -0
  101. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/gelu_erf.comp +39 -0
  102. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/glu_head.comp +15 -0
  103. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/glu_main.comp +29 -0
  104. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +128 -72
  105. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp +38 -9
  106. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/reglu.comp +9 -0
  107. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp +12 -3
  108. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/roll.comp +46 -0
  109. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_multi.comp +7 -9
  110. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_neox.comp +7 -9
  111. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_norm.comp +7 -9
  112. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/scale.comp +1 -1
  113. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/soft_max.comp +20 -4
  114. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/swiglu.comp +9 -0
  115. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/upscale.comp +69 -5
  116. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +23 -3
  117. package/cpp/llama.cpp/ggml/src/ggml.c +382 -61
  118. package/cpp/llama.cpp/ggml/src/gguf.cpp +8 -1
  119. package/cpp/llama.cpp/gguf-py/gguf/constants.py +209 -0
  120. package/cpp/llama.cpp/gguf-py/gguf/gguf_writer.py +8 -2
  121. package/cpp/llama.cpp/gguf-py/gguf/tensor_mapping.py +73 -21
  122. package/cpp/llama.cpp/gguf-py/gguf/vocab.py +12 -3
  123. package/cpp/llama.cpp/include/llama.h +0 -40
  124. package/cpp/llama.cpp/src/llama-arch.cpp +210 -3
  125. package/cpp/llama.cpp/src/llama-arch.h +18 -1
  126. package/cpp/llama.cpp/src/llama-batch.cpp +27 -1
  127. package/cpp/llama.cpp/src/llama-batch.h +8 -1
  128. package/cpp/llama.cpp/src/llama-chat.cpp +15 -0
  129. package/cpp/llama.cpp/src/llama-chat.h +1 -0
  130. package/cpp/llama.cpp/src/llama-graph.cpp +119 -184
  131. package/cpp/llama.cpp/src/llama-graph.h +47 -60
  132. package/cpp/llama.cpp/src/llama-hparams.cpp +7 -1
  133. package/cpp/llama.cpp/src/llama-hparams.h +3 -0
  134. package/cpp/llama.cpp/src/llama-kv-cache-unified-iswa.cpp +28 -18
  135. package/cpp/llama.cpp/src/llama-kv-cache-unified-iswa.h +4 -2
  136. package/cpp/llama.cpp/src/llama-kv-cache-unified.cpp +214 -65
  137. package/cpp/llama.cpp/src/llama-kv-cache-unified.h +62 -24
  138. package/cpp/llama.cpp/src/llama-kv-cells.h +62 -10
  139. package/cpp/llama.cpp/src/llama-memory-hybrid.cpp +9 -4
  140. package/cpp/llama.cpp/src/llama-memory-hybrid.h +3 -1
  141. package/cpp/llama.cpp/src/llama-memory-recurrent.cpp +20 -10
  142. package/cpp/llama.cpp/src/llama-memory.cpp +17 -0
  143. package/cpp/llama.cpp/src/llama-memory.h +3 -0
  144. package/cpp/llama.cpp/src/llama-model.cpp +2530 -685
  145. package/cpp/llama.cpp/src/llama-model.h +18 -0
  146. package/cpp/llama.cpp/src/llama-quant.cpp +1 -0
  147. package/cpp/llama.cpp/src/llama-vocab.cpp +13 -2
  148. package/cpp/llama.cpp/src/llama-vocab.h +41 -0
  149. package/ios/include/chat.h +2 -0
  150. package/ios/include/common.h +4 -0
  151. package/ios/include/llama.h +0 -40
  152. package/ios/libs/llama.xcframework/Info.plist +19 -19
  153. package/ios/libs/llama.xcframework/ios-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  154. package/ios/libs/llama.xcframework/ios-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +5055 -4886
  155. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/ggml-backend.h +1 -1
  156. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/ggml.h +173 -10
  157. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/llama.h +0 -40
  158. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/llama +0 -0
  159. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  160. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +5030 -4861
  161. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3889 -3764
  162. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml-backend.h +1 -1
  163. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +173 -10
  164. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/llama.h +0 -40
  165. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/llama +0 -0
  166. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  167. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +5030 -4861
  168. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3891 -3766
  169. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml-backend.h +1 -1
  170. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml.h +173 -10
  171. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/llama.h +0 -40
  172. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml-backend.h +1 -1
  173. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml.h +173 -10
  174. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/llama.h +0 -40
  175. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/llama +0 -0
  176. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml-backend.h +1 -1
  177. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml.h +173 -10
  178. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/llama.h +0 -40
  179. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/llama +0 -0
  180. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/llama +0 -0
  181. package/ios/libs/llama.xcframework/tvos-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  182. package/ios/libs/llama.xcframework/tvos-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +5059 -4890
  183. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml-backend.h +1 -1
  184. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml.h +173 -10
  185. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/llama.h +0 -40
  186. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/llama +0 -0
  187. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  188. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +5030 -4861
  189. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3889 -3764
  190. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml-backend.h +1 -1
  191. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +173 -10
  192. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/llama.h +0 -40
  193. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/llama +0 -0
  194. package/ios/libs/llama.xcframework/xros-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  195. package/ios/libs/llama.xcframework/xros-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +5091 -4922
  196. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml-backend.h +1 -1
  197. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml.h +173 -10
  198. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/llama.h +0 -40
  199. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/llama +0 -0
  200. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  201. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +5066 -4897
  202. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +3919 -3794
  203. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml-backend.h +1 -1
  204. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +173 -10
  205. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/llama.h +0 -40
  206. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/llama +0 -0
  207. package/package.json +1 -1
  208. package/cpp/llama.cpp/ggml/include/ggml-kompute.h +0 -50
  209. package/cpp/llama.cpp/ggml/src/ggml-kompute/CMakeLists.txt +0 -166
  210. package/cpp/llama.cpp/ggml/src/ggml-kompute/ggml-kompute.cpp +0 -2251
  211. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/common.comp +0 -112
  212. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_add.comp +0 -58
  213. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_addrow.comp +0 -25
  214. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f16_f16.comp +0 -52
  215. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f16_f32.comp +0 -52
  216. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f32_f16.comp +0 -52
  217. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_cpy_f32_f32.comp +0 -52
  218. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_diagmask.comp +0 -30
  219. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_gelu.comp +0 -22
  220. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_getrows.comp +0 -17
  221. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_getrows_f16.comp +0 -31
  222. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_getrows_f32.comp +0 -31
  223. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q4_0.comp +0 -38
  224. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q4_1.comp +0 -39
  225. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_getrows_q6_k.comp +0 -44
  226. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_mul.comp +0 -52
  227. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_f16.comp +0 -69
  228. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_mat_f32.comp +0 -51
  229. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_0.comp +0 -33
  230. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_1.comp +0 -35
  231. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q4_k.comp +0 -140
  232. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q6_k.comp +0 -106
  233. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_mul_mat_q8_0.comp +0 -73
  234. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_mul_mv_q_n.comp +0 -52
  235. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_mul_mv_q_n_pre.comp +0 -28
  236. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_norm.comp +0 -84
  237. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_relu.comp +0 -21
  238. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_rmsnorm.comp +0 -53
  239. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_rope_neox_f16.comp +0 -52
  240. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_rope_neox_f32.comp +0 -52
  241. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_rope_norm_f16.comp +0 -52
  242. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_rope_norm_f32.comp +0 -52
  243. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_scale.comp +0 -19
  244. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_scale_8.comp +0 -23
  245. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_silu.comp +0 -22
  246. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/op_softmax.comp +0 -72
  247. package/cpp/llama.cpp/ggml/src/ggml-kompute/kompute-shaders/rope_common.comp +0 -71
package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.c CHANGED
@@ -1193,7 +1193,7 @@ static void ggml_compute_forward_mul_mat_one_chunk(
1193
1193
  }
1194
1194
  }
1195
1195
 
1196
- static void ggml_compute_forward_mul_mat(
1196
+ void ggml_compute_forward_mul_mat(
1197
1197
  const struct ggml_compute_params * params,
1198
1198
  struct ggml_tensor * dst) {
1199
1199
 
@@ -1866,6 +1866,10 @@ static void ggml_compute_forward(struct ggml_compute_params * params, struct ggm
1866
1866
  {
1867
1867
  ggml_compute_forward_im2col_back_f32(params, tensor);
1868
1868
  } break;
1869
+ case GGML_OP_CONV_2D:
1870
+ {
1871
+ ggml_compute_forward_conv_2d(params, tensor);
1872
+ } break;
1869
1873
  case GGML_OP_CONV_2D_DW:
1870
1874
  {
1871
1875
  ggml_compute_forward_conv_2d_dw(params, tensor);
@@ -1949,6 +1953,10 @@ static void ggml_compute_forward(struct ggml_compute_params * params, struct ggm
1949
1953
  {
1950
1954
  ggml_compute_forward_unary(params, tensor);
1951
1955
  } break;
1956
+ case GGML_OP_GLU:
1957
+ {
1958
+ ggml_compute_forward_glu(params, tensor);
1959
+ } break;
1952
1960
  case GGML_OP_GET_REL_POS:
1953
1961
  {
1954
1962
  ggml_compute_forward_get_rel_pos(params, tensor);
@@ -2159,6 +2167,20 @@ static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads) {
2159
2167
  GGML_ABORT("fatal error");
2160
2168
  }
2161
2169
  break;
2170
+ case GGML_OP_GLU:
2171
+ switch (ggml_get_glu_op(node)) {
2172
+ case GGML_GLU_OP_REGLU:
2173
+ case GGML_GLU_OP_GEGLU:
2174
+ case GGML_GLU_OP_SWIGLU:
2175
+ case GGML_GLU_OP_GEGLU_ERF:
2176
+ case GGML_GLU_OP_GEGLU_QUICK:
2177
+ {
2178
+ n_tasks = n_threads;
2179
+ } break;
2180
+ default:
2181
+ GGML_ABORT("fatal error");
2182
+ }
2183
+ break;
2162
2184
  case GGML_OP_SILU_BACK:
2163
2185
  case GGML_OP_MUL:
2164
2186
  case GGML_OP_DIV:
@@ -2212,6 +2234,7 @@ static int ggml_get_n_tasks(struct ggml_tensor * node, int n_threads) {
2212
2234
  } break;
2213
2235
  case GGML_OP_IM2COL:
2214
2236
  case GGML_OP_IM2COL_BACK:
2237
+ case GGML_OP_CONV_2D:
2215
2238
  case GGML_OP_CONV_2D_DW:
2216
2239
  case GGML_OP_CONV_TRANSPOSE_1D:
2217
2240
  case GGML_OP_CONV_TRANSPOSE_2D:
@@ -2730,6 +2753,10 @@ struct ggml_cplan ggml_graph_plan(
2730
2753
  GGML_ABORT("fatal error");
2731
2754
  }
2732
2755
  } break;
2756
+ case GGML_OP_CONV_2D:
2757
+ {
2758
+ cur = GGML_IM2COL_WORK_SIZE;
2759
+ } break;
2733
2760
  case GGML_OP_CONV_TRANSPOSE_2D:
2734
2761
  {
2735
2762
  const int64_t ne00 = node->src[0]->ne[0]; // W