@novastera-oss/llamarn 0.4.1 → 0.4.3-beta4

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 (976) hide show
  1. package/RNLlamaCpp.podspec +3 -0
  2. package/android/CMakeLists.txt +2 -0
  3. package/android/src/main/cpp/include/llama.h +44 -21
  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/armeabi-v7a/libggml-base.so +0 -0
  9. package/android/src/main/jniLibs/armeabi-v7a/libggml-cpu.so +0 -0
  10. package/android/src/main/jniLibs/armeabi-v7a/libggml.so +0 -0
  11. package/android/src/main/jniLibs/armeabi-v7a/libllama.so +0 -0
  12. package/android/src/main/jniLibs/x86/libggml-base.so +0 -0
  13. package/android/src/main/jniLibs/x86/libggml-cpu.so +0 -0
  14. package/android/src/main/jniLibs/x86/libggml.so +0 -0
  15. package/android/src/main/jniLibs/x86/libllama.so +0 -0
  16. package/android/src/main/jniLibs/x86_64/libggml-base.so +0 -0
  17. package/android/src/main/jniLibs/x86_64/libggml-cpu.so +0 -0
  18. package/android/src/main/jniLibs/x86_64/libggml.so +0 -0
  19. package/android/src/main/jniLibs/x86_64/libllama.so +0 -0
  20. package/cpp/build-info.cpp +2 -2
  21. package/cpp/llama.cpp/CMakeLists.txt +12 -0
  22. package/cpp/llama.cpp/CODEOWNERS +116 -10
  23. package/cpp/llama.cpp/CONTRIBUTING.md +30 -3
  24. package/cpp/llama.cpp/README.md +13 -5
  25. package/cpp/llama.cpp/build-xcframework.sh +5 -0
  26. package/cpp/llama.cpp/cmake/riscv64-spacemit-linux-gnu-gcc.cmake +29 -0
  27. package/cpp/llama.cpp/common/CMakeLists.txt +12 -2
  28. package/cpp/llama.cpp/common/arg.cpp +303 -795
  29. package/cpp/llama.cpp/common/arg.h +2 -3
  30. package/cpp/llama.cpp/common/chat-parser-xml-toolcall.cpp +861 -0
  31. package/cpp/llama.cpp/common/chat-parser-xml-toolcall.h +45 -0
  32. package/cpp/llama.cpp/common/chat-parser.cpp +156 -15
  33. package/cpp/llama.cpp/common/chat-parser.h +13 -0
  34. package/cpp/llama.cpp/common/chat.cpp +1147 -88
  35. package/cpp/llama.cpp/common/chat.h +16 -3
  36. package/cpp/llama.cpp/common/common.cpp +70 -15
  37. package/cpp/llama.cpp/common/common.h +57 -19
  38. package/cpp/llama.cpp/common/download.cpp +1072 -0
  39. package/cpp/llama.cpp/common/download.h +55 -0
  40. package/cpp/llama.cpp/common/http.h +73 -0
  41. package/cpp/llama.cpp/common/json-partial.cpp +70 -2
  42. package/cpp/llama.cpp/common/json-schema-to-grammar.cpp +61 -22
  43. package/cpp/llama.cpp/common/json-schema-to-grammar.h +2 -0
  44. package/cpp/llama.cpp/common/log.cpp +59 -2
  45. package/cpp/llama.cpp/common/log.h +12 -4
  46. package/cpp/llama.cpp/common/sampling.cpp +84 -8
  47. package/cpp/llama.cpp/common/sampling.h +3 -1
  48. package/cpp/llama.cpp/common/speculative.cpp +1 -1
  49. package/cpp/llama.cpp/convert_hf_to_gguf.py +1608 -233
  50. package/cpp/llama.cpp/convert_hf_to_gguf_update.py +6 -1
  51. package/cpp/llama.cpp/convert_lora_to_gguf.py +37 -5
  52. package/cpp/llama.cpp/ggml/CMakeLists.txt +47 -28
  53. package/cpp/llama.cpp/ggml/include/ggml-backend.h +19 -1
  54. package/cpp/llama.cpp/ggml/include/ggml-cpu.h +1 -1
  55. package/cpp/llama.cpp/ggml/include/ggml-hexagon.h +19 -0
  56. package/cpp/llama.cpp/ggml/include/ggml-metal.h +1 -6
  57. package/cpp/llama.cpp/ggml/include/ggml-rpc.h +7 -9
  58. package/cpp/llama.cpp/ggml/include/ggml-zdnn.h +2 -1
  59. package/cpp/llama.cpp/ggml/include/ggml.h +199 -6
  60. package/cpp/llama.cpp/ggml/src/CMakeLists.txt +38 -0
  61. package/cpp/llama.cpp/ggml/src/ggml-alloc.c +299 -130
  62. package/cpp/llama.cpp/ggml/src/ggml-backend-impl.h +4 -4
  63. package/cpp/llama.cpp/ggml/src/ggml-backend-reg.cpp +21 -5
  64. package/cpp/llama.cpp/ggml/src/ggml-backend.cpp +99 -2
  65. package/cpp/llama.cpp/ggml/src/ggml-blas/CMakeLists.txt +1 -1
  66. package/cpp/llama.cpp/ggml/src/ggml-blas/ggml-blas.cpp +1 -0
  67. package/cpp/llama.cpp/ggml/src/ggml-cann/acl_tensor.cpp +57 -45
  68. package/cpp/llama.cpp/ggml/src/ggml-cann/acl_tensor.h +138 -47
  69. package/cpp/llama.cpp/ggml/src/ggml-cann/aclnn_ops.cpp +1584 -1773
  70. package/cpp/llama.cpp/ggml/src/ggml-cann/aclnn_ops.h +201 -317
  71. package/cpp/llama.cpp/ggml/src/ggml-cann/common.h +146 -187
  72. package/cpp/llama.cpp/ggml/src/ggml-cann/ggml-cann.cpp +771 -713
  73. package/cpp/llama.cpp/ggml/src/ggml-cpu/CMakeLists.txt +135 -77
  74. package/cpp/llama.cpp/ggml/src/ggml-cpu/amx/amx.cpp +5 -2
  75. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/arm/quants.c +428 -26
  76. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/loongarch/quants.c +16 -17
  77. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/riscv/quants.c +318 -145
  78. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/s390/cpu-feats.cpp +50 -0
  79. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/s390/quants.c +155 -60
  80. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch/x86/repack.cpp +8 -8
  81. package/cpp/llama.cpp/ggml/src/ggml-cpu/arch-fallback.h +0 -1
  82. package/cpp/llama.cpp/ggml/src/ggml-cpu/common.h +14 -0
  83. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-impl.h +10 -9
  84. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.c +108 -64
  85. package/cpp/llama.cpp/ggml/src/ggml-cpu/ggml-cpu.cpp +14 -4
  86. package/cpp/llama.cpp/ggml/src/ggml-cpu/kleidiai/kernels.cpp +530 -87
  87. package/cpp/llama.cpp/ggml/src/ggml-cpu/kleidiai/kernels.h +37 -45
  88. package/cpp/llama.cpp/ggml/src/ggml-cpu/kleidiai/kleidiai.cpp +349 -127
  89. package/cpp/llama.cpp/ggml/src/ggml-cpu/ops.cpp +947 -1218
  90. package/cpp/llama.cpp/ggml/src/ggml-cpu/ops.h +5 -4
  91. package/cpp/llama.cpp/ggml/src/ggml-cpu/repack.cpp +143 -29
  92. package/cpp/llama.cpp/ggml/src/ggml-cpu/simd-mappings.h +82 -76
  93. package/cpp/llama.cpp/ggml/src/ggml-cpu/spacemit/ime.cpp +1025 -0
  94. package/cpp/llama.cpp/ggml/src/ggml-cpu/spacemit/ime.h +13 -0
  95. package/cpp/llama.cpp/ggml/src/ggml-cpu/spacemit/ime1_kernels.cpp +3196 -0
  96. package/cpp/llama.cpp/ggml/src/ggml-cpu/spacemit/ime_kernels.h +26 -0
  97. package/cpp/llama.cpp/ggml/src/ggml-cpu/unary-ops.cpp +151 -0
  98. package/cpp/llama.cpp/ggml/src/ggml-cpu/unary-ops.h +7 -0
  99. package/cpp/llama.cpp/ggml/src/ggml-cpu/vec.cpp +233 -28
  100. package/cpp/llama.cpp/ggml/src/ggml-cpu/vec.h +326 -66
  101. package/cpp/llama.cpp/ggml/src/ggml-cuda/CMakeLists.txt +12 -3
  102. package/cpp/llama.cpp/ggml/src/ggml-cuda/argsort.cu +102 -6
  103. package/cpp/llama.cpp/ggml/src/ggml-cuda/binbcast.cu +110 -76
  104. package/cpp/llama.cpp/ggml/src/ggml-cuda/common.cuh +167 -38
  105. package/cpp/llama.cpp/ggml/src/ggml-cuda/conv2d.cu +6 -11
  106. package/cpp/llama.cpp/ggml/src/ggml-cuda/convert.cuh +12 -0
  107. package/cpp/llama.cpp/ggml/src/ggml-cuda/cpy-utils.cuh +1 -1
  108. package/cpp/llama.cpp/ggml/src/ggml-cuda/cpy.cu +245 -151
  109. package/cpp/llama.cpp/ggml/src/ggml-cuda/cpy.cuh +1 -5
  110. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-common.cuh +341 -289
  111. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-tile.cu +49 -0
  112. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-tile.cuh +1233 -0
  113. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-vec.cuh +586 -0
  114. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-wmma-f16.cu +6 -6
  115. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-wmma-f16.cuh +48 -0
  116. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn.cu +123 -220
  117. package/cpp/llama.cpp/ggml/src/ggml-cuda/getrows.cu +41 -39
  118. package/cpp/llama.cpp/ggml/src/ggml-cuda/ggml-cuda.cu +715 -45
  119. package/cpp/llama.cpp/ggml/src/ggml-cuda/im2col.cu +150 -0
  120. package/cpp/llama.cpp/ggml/src/ggml-cuda/im2col.cuh +1 -0
  121. package/cpp/llama.cpp/ggml/src/ggml-cuda/mma.cuh +321 -24
  122. package/cpp/llama.cpp/ggml/src/ggml-cuda/mmf.cu +93 -351
  123. package/cpp/llama.cpp/ggml/src/ggml-cuda/mmf.cuh +828 -1
  124. package/cpp/llama.cpp/ggml/src/ggml-cuda/mmid.cu +164 -0
  125. package/cpp/llama.cpp/ggml/src/ggml-cuda/mmid.cuh +5 -0
  126. package/cpp/llama.cpp/ggml/src/ggml-cuda/mmq.cu +3 -166
  127. package/cpp/llama.cpp/ggml/src/ggml-cuda/mmq.cuh +1 -1
  128. package/cpp/llama.cpp/ggml/src/ggml-cuda/mmvf.cu +371 -78
  129. package/cpp/llama.cpp/ggml/src/ggml-cuda/mmvf.cuh +3 -2
  130. package/cpp/llama.cpp/ggml/src/ggml-cuda/mmvq.cu +279 -147
  131. package/cpp/llama.cpp/ggml/src/ggml-cuda/mmvq.cuh +1 -1
  132. package/cpp/llama.cpp/ggml/src/ggml-cuda/norm.cu +97 -85
  133. package/cpp/llama.cpp/ggml/src/ggml-cuda/pad.cu +46 -23
  134. package/cpp/llama.cpp/ggml/src/ggml-cuda/pad_reflect_1d.cu +63 -54
  135. package/cpp/llama.cpp/ggml/src/ggml-cuda/quantize.cu +12 -10
  136. package/cpp/llama.cpp/ggml/src/ggml-cuda/rope.cu +192 -77
  137. package/cpp/llama.cpp/ggml/src/ggml-cuda/rope.cuh +2 -0
  138. package/cpp/llama.cpp/ggml/src/ggml-cuda/scale.cu +10 -9
  139. package/cpp/llama.cpp/ggml/src/ggml-cuda/set-rows.cu +137 -75
  140. package/cpp/llama.cpp/ggml/src/ggml-cuda/set.cu +39 -0
  141. package/cpp/llama.cpp/ggml/src/ggml-cuda/set.cuh +7 -0
  142. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq112-dv112.cu +5 -0
  143. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq128-dv128.cu +5 -0
  144. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq256-dv256.cu +5 -0
  145. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq40-dv40.cu +5 -0
  146. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq576-dv512.cu +5 -0
  147. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq64-dv64.cu +5 -0
  148. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq72-dv72.cu +5 -0
  149. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq80-dv80.cu +5 -0
  150. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-tile-instance-dkq96-dv96.cu +5 -0
  151. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-f16.cu +7 -0
  152. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_0.cu +7 -0
  153. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q4_1.cu +7 -0
  154. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_0.cu +7 -0
  155. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q5_1.cu +7 -0
  156. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-f16-q8_0.cu +7 -0
  157. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-f16.cu +7 -0
  158. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_0.cu +7 -0
  159. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_1.cu +7 -0
  160. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_0.cu +7 -0
  161. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q5_1.cu +7 -0
  162. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_0-q8_0.cu +7 -0
  163. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-f16.cu +7 -0
  164. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_0.cu +7 -0
  165. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q4_1.cu +7 -0
  166. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_0.cu +7 -0
  167. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q5_1.cu +7 -0
  168. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q4_1-q8_0.cu +7 -0
  169. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-f16.cu +7 -0
  170. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_0.cu +7 -0
  171. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q4_1.cu +7 -0
  172. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_0.cu +7 -0
  173. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q5_1.cu +7 -0
  174. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_0-q8_0.cu +7 -0
  175. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-f16.cu +7 -0
  176. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_0.cu +7 -0
  177. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q4_1.cu +7 -0
  178. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_0.cu +7 -0
  179. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q5_1.cu +7 -0
  180. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q5_1-q8_0.cu +7 -0
  181. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-f16.cu +7 -0
  182. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_0.cu +7 -0
  183. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q4_1.cu +7 -0
  184. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_0.cu +7 -0
  185. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q5_1.cu +7 -0
  186. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-instance-q8_0-q8_0.cu +7 -0
  187. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/generate_cu_files.py +40 -19
  188. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_1.cu +5 -0
  189. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_10.cu +5 -0
  190. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_11.cu +5 -0
  191. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_12.cu +5 -0
  192. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_13.cu +5 -0
  193. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_14.cu +5 -0
  194. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_15.cu +5 -0
  195. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_16.cu +5 -0
  196. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_2.cu +5 -0
  197. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_3.cu +5 -0
  198. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_4.cu +5 -0
  199. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_5.cu +5 -0
  200. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_6.cu +5 -0
  201. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_7.cu +5 -0
  202. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_8.cu +5 -0
  203. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/mmf-instance-ncols_9.cu +5 -0
  204. package/cpp/llama.cpp/ggml/src/ggml-cuda/topk-moe.cu +336 -0
  205. package/cpp/llama.cpp/ggml/src/ggml-cuda/topk-moe.cuh +16 -0
  206. package/cpp/llama.cpp/ggml/src/ggml-cuda/tsembd.cu +3 -3
  207. package/cpp/llama.cpp/ggml/src/ggml-cuda/unary.cu +105 -11
  208. package/cpp/llama.cpp/ggml/src/ggml-cuda/unary.cuh +36 -0
  209. package/cpp/llama.cpp/ggml/src/ggml-cuda/upscale.cu +87 -6
  210. package/cpp/llama.cpp/ggml/src/ggml-cuda/vendors/hip.h +28 -12
  211. package/cpp/llama.cpp/ggml/src/ggml-hexagon/CMakeLists.txt +68 -0
  212. package/cpp/llama.cpp/ggml/src/ggml-hexagon/ggml-hexagon.cpp +3807 -0
  213. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/CMakeLists.txt +40 -0
  214. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/act-ops.c +442 -0
  215. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/binary-ops.c +360 -0
  216. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/cmake-toolchain.cmake +157 -0
  217. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/htp-ctx.h +40 -0
  218. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/htp-dma.c +69 -0
  219. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/htp-dma.h +119 -0
  220. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/htp-msg.h +156 -0
  221. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/htp-ops.h +64 -0
  222. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/htp_iface.idl +16 -0
  223. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/hvx-exp.c +93 -0
  224. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/hvx-inverse.c +60 -0
  225. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/hvx-sigmoid.c +49 -0
  226. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/hvx-utils.c +960 -0
  227. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/hvx-utils.h +1032 -0
  228. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/main.c +829 -0
  229. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/matmul-ops.c +2223 -0
  230. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/ops-utils.h +149 -0
  231. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/rope-ops.c +418 -0
  232. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/softmax-ops.c +402 -0
  233. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/unary-ops.c +255 -0
  234. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/worker-pool.c +297 -0
  235. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp/worker-pool.h +57 -0
  236. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp-utils.c +448 -0
  237. package/cpp/llama.cpp/ggml/src/ggml-hexagon/htp-utils.h +220 -0
  238. package/cpp/llama.cpp/ggml/src/ggml-hip/CMakeLists.txt +8 -13
  239. package/cpp/llama.cpp/ggml/src/ggml-impl.h +110 -12
  240. package/cpp/llama.cpp/ggml/src/ggml-metal/CMakeLists.txt +6 -5
  241. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-common.cpp +446 -0
  242. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-common.h +52 -0
  243. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-context.h +33 -0
  244. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-context.m +599 -0
  245. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-device.cpp +1662 -0
  246. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-device.h +251 -0
  247. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-device.m +1527 -0
  248. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-impl.h +244 -39
  249. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-ops.cpp +3844 -0
  250. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal-ops.h +90 -0
  251. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal.cpp +723 -0
  252. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal.metal +3453 -1907
  253. package/cpp/llama.cpp/ggml/src/ggml-musa/CMakeLists.txt +3 -1
  254. package/cpp/llama.cpp/ggml/src/ggml-opencl/CMakeLists.txt +10 -0
  255. package/cpp/llama.cpp/ggml/src/ggml-opencl/ggml-opencl.cpp +1331 -109
  256. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/cvt.cl +126 -0
  257. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/flash_attn_f16.cl +31 -4
  258. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/flash_attn_f32.cl +35 -7
  259. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/flash_attn_f32_f16.cl +31 -4
  260. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/gemm_moe_mxfp4_f32.cl +162 -0
  261. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/gemv_moe_mxfp4_f32.cl +156 -0
  262. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/get_rows.cl +36 -12
  263. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mm_f16_f32_kq_kqv.cl +273 -0
  264. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mm_f16_f32_l4_lm.cl +24 -10
  265. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mm_f32_f32_l4_lm.cl +24 -10
  266. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mm_q8_0_f32_l4_lm.cl +154 -0
  267. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mv_id_mxfp4_f32_flat.cl +176 -0
  268. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32.cl +140 -0
  269. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mv_id_q8_0_f32_flat.cl +222 -0
  270. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mv_mxfp4_f32_flat.cl +167 -0
  271. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32.cl +125 -0
  272. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/mul_mv_q8_0_f32_flat.cl +202 -0
  273. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/pad.cl +29 -20
  274. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/rms_norm.cl +25 -10
  275. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/rope.cl +50 -24
  276. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/set_rows.cl +123 -10
  277. package/cpp/llama.cpp/ggml/src/ggml-opencl/kernels/tsembd.cl +2 -2
  278. package/cpp/llama.cpp/ggml/src/ggml-quants.c +1 -0
  279. package/cpp/llama.cpp/ggml/src/ggml-rpc/ggml-rpc.cpp +341 -161
  280. package/cpp/llama.cpp/ggml/src/ggml-sycl/backend.hpp +6 -0
  281. package/cpp/llama.cpp/ggml/src/ggml-sycl/binbcast.cpp +6 -5
  282. package/cpp/llama.cpp/ggml/src/ggml-sycl/common.hpp +74 -15
  283. package/cpp/llama.cpp/ggml/src/ggml-sycl/concat.cpp +50 -30
  284. package/cpp/llama.cpp/ggml/src/ggml-sycl/conv.cpp +10 -4
  285. package/cpp/llama.cpp/ggml/src/ggml-sycl/convert.cpp +166 -99
  286. package/cpp/llama.cpp/ggml/src/ggml-sycl/count-equal.cpp +79 -0
  287. package/cpp/llama.cpp/ggml/src/ggml-sycl/count-equal.hpp +9 -0
  288. package/cpp/llama.cpp/ggml/src/ggml-sycl/cpy.cpp +72 -94
  289. package/cpp/llama.cpp/ggml/src/ggml-sycl/dmmv.cpp +67 -49
  290. package/cpp/llama.cpp/ggml/src/ggml-sycl/dpct/helper.hpp +21 -31
  291. package/cpp/llama.cpp/ggml/src/ggml-sycl/element_wise.cpp +252 -316
  292. package/cpp/llama.cpp/ggml/src/ggml-sycl/element_wise.hpp +6 -2
  293. package/cpp/llama.cpp/ggml/src/ggml-sycl/getrows.cpp +9 -6
  294. package/cpp/llama.cpp/ggml/src/ggml-sycl/ggml-sycl.cpp +359 -142
  295. package/cpp/llama.cpp/ggml/src/ggml-sycl/gla.cpp +2 -2
  296. package/cpp/llama.cpp/ggml/src/ggml-sycl/im2col.cpp +1 -1
  297. package/cpp/llama.cpp/ggml/src/ggml-sycl/mmq.cpp +80 -60
  298. package/cpp/llama.cpp/ggml/src/ggml-sycl/mmvq.cpp +201 -132
  299. package/cpp/llama.cpp/ggml/src/ggml-sycl/norm.cpp +230 -55
  300. package/cpp/llama.cpp/ggml/src/ggml-sycl/norm.hpp +2 -0
  301. package/cpp/llama.cpp/ggml/src/ggml-sycl/pad.cpp +97 -0
  302. package/cpp/llama.cpp/ggml/src/ggml-sycl/pad.hpp +24 -0
  303. package/cpp/llama.cpp/ggml/src/ggml-sycl/pad_reflect_1d.cpp +72 -0
  304. package/cpp/llama.cpp/ggml/src/ggml-sycl/pad_reflect_1d.hpp +8 -0
  305. package/cpp/llama.cpp/ggml/src/ggml-sycl/presets.hpp +2 -0
  306. package/cpp/llama.cpp/ggml/src/ggml-sycl/repeat_back.cpp +76 -0
  307. package/cpp/llama.cpp/ggml/src/ggml-sycl/repeat_back.hpp +8 -0
  308. package/cpp/llama.cpp/ggml/src/ggml-sycl/roll.cpp +122 -0
  309. package/cpp/llama.cpp/ggml/src/ggml-sycl/roll.hpp +20 -0
  310. package/cpp/llama.cpp/ggml/src/ggml-sycl/rope.cpp +50 -41
  311. package/cpp/llama.cpp/ggml/src/ggml-sycl/set.cpp +73 -0
  312. package/cpp/llama.cpp/ggml/src/ggml-sycl/set.hpp +5 -0
  313. package/cpp/llama.cpp/ggml/src/ggml-sycl/set_rows.cpp +45 -36
  314. package/cpp/llama.cpp/ggml/src/ggml-sycl/softmax.cpp +330 -165
  315. package/cpp/llama.cpp/ggml/src/ggml-sycl/softmax.hpp +4 -0
  316. package/cpp/llama.cpp/ggml/src/ggml-sycl/ssm_conv.cpp +127 -0
  317. package/cpp/llama.cpp/ggml/src/ggml-sycl/ssm_conv.hpp +5 -0
  318. package/cpp/llama.cpp/ggml/src/ggml-sycl/tsembd.cpp +12 -6
  319. package/cpp/llama.cpp/ggml/src/ggml-sycl/wkv.cpp +16 -12
  320. package/cpp/llama.cpp/ggml/src/ggml-vulkan/CMakeLists.txt +38 -18
  321. package/cpp/llama.cpp/ggml/src/ggml-vulkan/ggml-vulkan.cpp +4184 -2159
  322. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/abs.comp +21 -0
  323. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/acc.comp +2 -2
  324. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/add.comp +2 -2
  325. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/add1.comp +28 -0
  326. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/add_id.comp +1 -1
  327. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/arange.comp +20 -0
  328. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/argmax.comp +2 -2
  329. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/argsort.comp +33 -26
  330. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/argsort_large.comp +114 -0
  331. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/ceil.comp +22 -0
  332. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/clamp.comp +2 -2
  333. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/concat.comp +2 -2
  334. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/contig_copy.comp +2 -2
  335. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_dw.comp +1 -1
  336. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/conv2d_mm.comp +53 -30
  337. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/conv_transpose_1d.comp +1 -1
  338. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/copy.comp +2 -2
  339. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/copy_from_quant.comp +3 -3
  340. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/copy_to_quant.comp +13 -6
  341. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/copy_transpose.comp +67 -0
  342. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/cos.comp +2 -2
  343. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/count_equal.comp +2 -2
  344. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_f32.comp +1 -1
  345. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{dequant_funcs.comp → dequant_funcs.glsl} +138 -2
  346. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{dequant_funcs_cm2.comp → dequant_funcs_cm2.glsl} +18 -4
  347. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{dequant_head.comp → dequant_head.glsl} +1 -1
  348. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq1_m.comp +1 -1
  349. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq1_s.comp +1 -1
  350. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_s.comp +2 -2
  351. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xs.comp +1 -1
  352. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq2_xxs.comp +3 -2
  353. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_s.comp +7 -6
  354. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq3_xxs.comp +5 -3
  355. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq4_nl.comp +1 -1
  356. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_iq4_xs.comp +1 -1
  357. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_mxfp4.comp +3 -3
  358. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q2_k.comp +3 -3
  359. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q3_k.comp +1 -1
  360. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_0.comp +1 -1
  361. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_1.comp +1 -1
  362. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q4_k.comp +3 -3
  363. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_0.comp +1 -1
  364. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_1.comp +1 -1
  365. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q5_k.comp +3 -3
  366. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q6_k.comp +1 -1
  367. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/dequant_q8_0.comp +1 -1
  368. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/diag_mask_inf.comp +1 -1
  369. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/div.comp +2 -2
  370. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/exp.comp +3 -2
  371. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/fill.comp +19 -0
  372. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn.comp +52 -14
  373. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{flash_attn_base.comp → flash_attn_base.glsl} +50 -12
  374. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm1.comp +61 -12
  375. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_cm2.comp +54 -12
  376. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/flash_attn_split_k_reduce.comp +5 -1
  377. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/floor.comp +22 -0
  378. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/geglu.comp +2 -2
  379. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/geglu_erf.comp +2 -2
  380. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/geglu_quick.comp +2 -2
  381. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/gelu.comp +2 -2
  382. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/gelu_erf.comp +2 -2
  383. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/gelu_quick.comp +2 -2
  384. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{generic_binary_head.comp → generic_binary_head.glsl} +10 -2
  385. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/get_rows.comp +21 -12
  386. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/get_rows_quant.comp +28 -18
  387. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{glu_head.comp → glu_head.glsl} +1 -1
  388. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/group_norm.comp +2 -2
  389. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/hardsigmoid.comp +22 -0
  390. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/hardswish.comp +22 -0
  391. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/im2col.comp +15 -7
  392. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/im2col_3d.comp +125 -0
  393. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/l2_norm.comp +2 -2
  394. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/leaky_relu.comp +2 -2
  395. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/log.comp +18 -0
  396. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul.comp +2 -2
  397. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec.comp +1 -1
  398. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_base.glsl +229 -0
  399. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iface.glsl +33 -0
  400. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq1_m.comp +1 -1
  401. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq1_s.comp +1 -1
  402. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_s.comp +1 -1
  403. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xs.comp +1 -1
  404. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq2_xxs.comp +1 -1
  405. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_s.comp +1 -1
  406. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_iq3_xxs.comp +1 -1
  407. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_nc.comp +9 -7
  408. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_p021.comp +9 -7
  409. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q2_k.comp +3 -5
  410. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q3_k.comp +1 -1
  411. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q4_k.comp +3 -5
  412. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q5_k.comp +3 -5
  413. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_q6_k.comp +1 -1
  414. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vecq.comp +140 -0
  415. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm.comp +106 -634
  416. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_cm2.comp +118 -9
  417. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_funcs.glsl +556 -0
  418. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mm_id_funcs.glsl +70 -0
  419. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq.comp +77 -214
  420. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.glsl +589 -0
  421. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_shmem_types.glsl +78 -0
  422. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/multi_add.comp +97 -13
  423. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/neg.comp +20 -0
  424. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/norm.comp +2 -2
  425. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/opt_step_adamw.comp +2 -2
  426. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/opt_step_sgd.comp +1 -1
  427. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/pad.comp +25 -4
  428. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/pool2d.comp +1 -1
  429. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/quantize_q8_1.comp +55 -5
  430. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/reglu.comp +2 -2
  431. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/relu.comp +2 -2
  432. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/repeat.comp +2 -2
  433. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/repeat_back.comp +2 -2
  434. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm.comp +45 -3
  435. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm_back.comp +2 -2
  436. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rms_norm_partials.comp +2 -2
  437. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/roll.comp +2 -2
  438. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_funcs.glsl +227 -0
  439. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.glsl +20 -0
  440. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_multi.comp +5 -52
  441. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_neox.comp +5 -35
  442. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_norm.comp +5 -35
  443. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_params.glsl +27 -0
  444. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_vision.comp +5 -41
  445. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/round.comp +29 -0
  446. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/scale.comp +2 -2
  447. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/sigmoid.comp +2 -2
  448. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/silu.comp +2 -2
  449. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/silu_back.comp +2 -2
  450. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/sin.comp +2 -2
  451. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/soft_max.comp +1 -1
  452. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/soft_max_back.comp +6 -2
  453. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/softplus.comp +23 -0
  454. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/sqrt.comp +2 -2
  455. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/square.comp +2 -2
  456. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/ssm_conv.comp +44 -0
  457. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/ssm_scan.comp +140 -0
  458. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/step.comp +22 -0
  459. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/sub.comp +2 -2
  460. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/sum_rows.comp +1 -1
  461. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/swiglu.comp +2 -2
  462. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/swiglu_oai.comp +2 -2
  463. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/tanh.comp +2 -2
  464. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/timestep_embedding.comp +5 -4
  465. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/topk_moe.comp +171 -0
  466. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/trunc.comp +22 -0
  467. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{types.comp → types.glsl} +79 -29
  468. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/upscale.comp +36 -12
  469. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +471 -196
  470. package/cpp/llama.cpp/ggml/src/ggml-webgpu/CMakeLists.txt +8 -0
  471. package/cpp/llama.cpp/ggml/src/ggml-webgpu/ggml-webgpu.cpp +1690 -383
  472. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/bin_op.tmpl.wgsl +188 -0
  473. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/binary_head.tmpl +45 -0
  474. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/common_decls.tmpl +930 -0
  475. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/cpy.tmpl.wgsl +101 -0
  476. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py +57 -10
  477. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/get_rows.tmpl.wgsl +874 -0
  478. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/glu.tmpl.wgsl +323 -0
  479. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat.tmpl.wgsl +25 -912
  480. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_decls.tmpl +97 -0
  481. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_reg_tile.tmpl.wgsl +247 -0
  482. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_subgroup_matrix.tmpl.wgsl +302 -0
  483. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec.tmpl.wgsl +267 -0
  484. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/rms_norm.wgsl +123 -0
  485. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/rope.tmpl.wgsl +295 -0
  486. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/scale.tmpl.wgsl +90 -0
  487. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/{set_rows.wgsl → set_rows.tmpl.wgsl} +38 -8
  488. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/soft_max.tmpl.wgsl +345 -0
  489. package/cpp/llama.cpp/ggml/src/ggml-zdnn/common.hpp +59 -0
  490. package/cpp/llama.cpp/ggml/src/ggml-zdnn/ggml-zdnn.cpp +96 -314
  491. package/cpp/llama.cpp/ggml/src/ggml-zdnn/mmf.cpp +80 -0
  492. package/cpp/llama.cpp/ggml/src/ggml-zdnn/mmf.hpp +12 -0
  493. package/cpp/llama.cpp/ggml/src/ggml-zdnn/utils.cpp +79 -0
  494. package/cpp/llama.cpp/ggml/src/ggml-zdnn/utils.hpp +19 -0
  495. package/cpp/llama.cpp/ggml/src/ggml.c +440 -17
  496. package/cpp/llama.cpp/ggml/src/gguf.cpp +104 -29
  497. package/cpp/llama.cpp/gguf-py/gguf/constants.py +363 -13
  498. package/cpp/llama.cpp/gguf-py/gguf/gguf_writer.py +64 -0
  499. package/cpp/llama.cpp/gguf-py/gguf/lazy.py +8 -3
  500. package/cpp/llama.cpp/gguf-py/gguf/scripts/gguf_convert_endian.py +6 -0
  501. package/cpp/llama.cpp/gguf-py/gguf/tensor_mapping.py +156 -18
  502. package/cpp/llama.cpp/gguf-py/gguf/utility.py +80 -0
  503. package/cpp/llama.cpp/gguf-py/gguf/vocab.py +4 -4
  504. package/cpp/llama.cpp/include/llama.h +44 -21
  505. package/cpp/llama.cpp/media/llama1-icon-transparent.png +0 -0
  506. package/cpp/llama.cpp/media/llama1-icon-transparent.svg +77 -0
  507. package/cpp/llama.cpp/media/llama1-icon.png +0 -0
  508. package/cpp/llama.cpp/media/llama1-icon.svg +87 -0
  509. package/cpp/llama.cpp/requirements/requirements-all.txt +2 -0
  510. package/cpp/llama.cpp/requirements/requirements-convert_hf_to_gguf.txt +3 -3
  511. package/cpp/llama.cpp/requirements/requirements-convert_legacy_llama.txt +3 -1
  512. package/cpp/llama.cpp/requirements/requirements-tool_bench.txt +1 -1
  513. package/cpp/llama.cpp/src/CMakeLists.txt +101 -0
  514. package/cpp/llama.cpp/src/llama-adapter.cpp +33 -0
  515. package/cpp/llama.cpp/src/llama-adapter.h +3 -0
  516. package/cpp/llama.cpp/src/llama-arch.cpp +344 -14
  517. package/cpp/llama.cpp/src/llama-arch.h +50 -0
  518. package/cpp/llama.cpp/src/llama-batch.cpp +63 -31
  519. package/cpp/llama.cpp/src/llama-batch.h +13 -2
  520. package/cpp/llama.cpp/src/llama-chat.cpp +85 -3
  521. package/cpp/llama.cpp/src/llama-chat.h +4 -0
  522. package/cpp/llama.cpp/src/llama-context.cpp +300 -45
  523. package/cpp/llama.cpp/src/llama-context.h +16 -6
  524. package/cpp/llama.cpp/src/llama-cparams.h +2 -1
  525. package/cpp/llama.cpp/src/llama-grammar.cpp +17 -9
  526. package/cpp/llama.cpp/src/llama-graph.cpp +226 -64
  527. package/cpp/llama.cpp/src/llama-graph.h +27 -5
  528. package/cpp/llama.cpp/src/llama-hparams.cpp +53 -2
  529. package/cpp/llama.cpp/src/llama-hparams.h +48 -8
  530. package/cpp/llama.cpp/src/llama-impl.cpp +3 -3
  531. package/cpp/llama.cpp/src/llama-impl.h +2 -0
  532. package/cpp/llama.cpp/src/llama-kv-cache-iswa.cpp +13 -3
  533. package/cpp/llama.cpp/src/llama-kv-cache-iswa.h +2 -0
  534. package/cpp/llama.cpp/src/llama-kv-cache.cpp +120 -62
  535. package/cpp/llama.cpp/src/llama-kv-cache.h +13 -4
  536. package/cpp/llama.cpp/src/llama-kv-cells.h +44 -2
  537. package/cpp/llama.cpp/src/llama-memory-hybrid.cpp +19 -9
  538. package/cpp/llama.cpp/src/llama-memory-hybrid.h +2 -0
  539. package/cpp/llama.cpp/src/llama-memory-recurrent.cpp +38 -17
  540. package/cpp/llama.cpp/src/llama-memory-recurrent.h +5 -2
  541. package/cpp/llama.cpp/src/llama-memory.h +3 -0
  542. package/cpp/llama.cpp/src/llama-model-loader.cpp +2 -0
  543. package/cpp/llama.cpp/src/llama-model.cpp +1070 -12614
  544. package/cpp/llama.cpp/src/llama-model.h +40 -4
  545. package/cpp/llama.cpp/src/llama-quant.cpp +14 -6
  546. package/cpp/llama.cpp/src/llama-sampling.cpp +243 -136
  547. package/cpp/llama.cpp/src/llama-vocab.cpp +43 -3
  548. package/cpp/llama.cpp/src/llama-vocab.h +43 -39
  549. package/cpp/llama.cpp/src/llama.cpp +69 -10
  550. package/cpp/llama.cpp/src/models/afmoe.cpp +187 -0
  551. package/cpp/llama.cpp/src/models/apertus.cpp +125 -0
  552. package/cpp/llama.cpp/src/models/arcee.cpp +135 -0
  553. package/cpp/llama.cpp/src/models/arctic.cpp +138 -0
  554. package/cpp/llama.cpp/src/models/arwkv7.cpp +86 -0
  555. package/cpp/llama.cpp/src/models/baichuan.cpp +122 -0
  556. package/cpp/llama.cpp/src/models/bailingmoe.cpp +144 -0
  557. package/cpp/llama.cpp/src/models/bailingmoe2.cpp +135 -0
  558. package/cpp/llama.cpp/src/models/bert.cpp +176 -0
  559. package/cpp/llama.cpp/src/models/bitnet.cpp +160 -0
  560. package/cpp/llama.cpp/src/models/bloom.cpp +101 -0
  561. package/cpp/llama.cpp/src/models/chameleon.cpp +178 -0
  562. package/cpp/llama.cpp/src/models/chatglm.cpp +132 -0
  563. package/cpp/llama.cpp/src/models/codeshell.cpp +111 -0
  564. package/cpp/llama.cpp/src/models/cogvlm.cpp +100 -0
  565. package/cpp/llama.cpp/src/models/cohere2-iswa.cpp +131 -0
  566. package/cpp/llama.cpp/src/models/command-r.cpp +122 -0
  567. package/cpp/llama.cpp/src/models/dbrx.cpp +123 -0
  568. package/cpp/llama.cpp/src/models/deci.cpp +135 -0
  569. package/cpp/llama.cpp/src/models/deepseek.cpp +144 -0
  570. package/cpp/llama.cpp/src/models/deepseek2.cpp +237 -0
  571. package/cpp/llama.cpp/src/models/dots1.cpp +134 -0
  572. package/cpp/llama.cpp/src/models/dream.cpp +105 -0
  573. package/cpp/llama.cpp/src/models/ernie4-5-moe.cpp +150 -0
  574. package/cpp/llama.cpp/src/models/ernie4-5.cpp +110 -0
  575. package/cpp/llama.cpp/src/models/exaone.cpp +114 -0
  576. package/cpp/llama.cpp/src/models/exaone4.cpp +123 -0
  577. package/cpp/llama.cpp/src/models/falcon-h1.cpp +113 -0
  578. package/cpp/llama.cpp/src/models/falcon.cpp +120 -0
  579. package/cpp/llama.cpp/src/models/gemma-embedding.cpp +120 -0
  580. package/cpp/llama.cpp/src/models/gemma.cpp +112 -0
  581. package/cpp/llama.cpp/src/models/gemma2-iswa.cpp +125 -0
  582. package/cpp/llama.cpp/src/models/gemma3-iswa.cpp +131 -0
  583. package/cpp/llama.cpp/src/models/gemma3n-iswa.cpp +377 -0
  584. package/cpp/llama.cpp/src/models/glm4-moe.cpp +153 -0
  585. package/cpp/llama.cpp/src/models/glm4.cpp +127 -0
  586. package/cpp/llama.cpp/src/models/gpt2.cpp +105 -0
  587. package/cpp/llama.cpp/src/models/gptneox.cpp +144 -0
  588. package/cpp/llama.cpp/src/models/granite-hybrid.cpp +196 -0
  589. package/cpp/llama.cpp/src/models/granite.cpp +211 -0
  590. package/cpp/llama.cpp/src/models/graph-context-mamba.cpp +283 -0
  591. package/cpp/llama.cpp/src/models/grok.cpp +159 -0
  592. package/cpp/llama.cpp/src/models/grovemoe.cpp +141 -0
  593. package/cpp/llama.cpp/src/models/hunyuan-dense.cpp +132 -0
  594. package/cpp/llama.cpp/src/models/hunyuan-moe.cpp +154 -0
  595. package/cpp/llama.cpp/src/models/internlm2.cpp +120 -0
  596. package/cpp/llama.cpp/src/models/jais.cpp +86 -0
  597. package/cpp/llama.cpp/src/models/jamba.cpp +106 -0
  598. package/cpp/llama.cpp/src/models/lfm2.cpp +173 -0
  599. package/cpp/llama.cpp/src/models/llada-moe.cpp +122 -0
  600. package/cpp/llama.cpp/src/models/llada.cpp +99 -0
  601. package/cpp/llama.cpp/src/models/llama-iswa.cpp +174 -0
  602. package/cpp/llama.cpp/src/models/llama.cpp +155 -0
  603. package/cpp/llama.cpp/src/models/mamba.cpp +55 -0
  604. package/cpp/llama.cpp/src/models/minicpm3.cpp +199 -0
  605. package/cpp/llama.cpp/src/models/minimax-m2.cpp +124 -0
  606. package/cpp/llama.cpp/src/models/models.h +485 -0
  607. package/cpp/llama.cpp/src/models/mpt.cpp +126 -0
  608. package/cpp/llama.cpp/src/models/nemotron-h.cpp +121 -0
  609. package/cpp/llama.cpp/src/models/nemotron.cpp +122 -0
  610. package/cpp/llama.cpp/src/models/neo-bert.cpp +104 -0
  611. package/cpp/llama.cpp/src/models/olmo.cpp +121 -0
  612. package/cpp/llama.cpp/src/models/olmo2.cpp +150 -0
  613. package/cpp/llama.cpp/src/models/olmoe.cpp +124 -0
  614. package/cpp/llama.cpp/src/models/openai-moe-iswa.cpp +124 -0
  615. package/cpp/llama.cpp/src/models/openelm.cpp +124 -0
  616. package/cpp/llama.cpp/src/models/orion.cpp +123 -0
  617. package/cpp/llama.cpp/src/models/pangu-embedded.cpp +121 -0
  618. package/cpp/llama.cpp/src/models/phi2.cpp +121 -0
  619. package/cpp/llama.cpp/src/models/phi3.cpp +152 -0
  620. package/cpp/llama.cpp/src/models/plamo.cpp +110 -0
  621. package/cpp/llama.cpp/src/models/plamo2.cpp +316 -0
  622. package/cpp/llama.cpp/src/models/plm.cpp +168 -0
  623. package/cpp/llama.cpp/src/models/qwen.cpp +108 -0
  624. package/cpp/llama.cpp/src/models/qwen2.cpp +117 -0
  625. package/cpp/llama.cpp/src/models/qwen2moe.cpp +151 -0
  626. package/cpp/llama.cpp/src/models/qwen2vl.cpp +117 -0
  627. package/cpp/llama.cpp/src/models/qwen3.cpp +117 -0
  628. package/cpp/llama.cpp/src/models/qwen3moe.cpp +124 -0
  629. package/cpp/llama.cpp/src/models/qwen3vl-moe.cpp +149 -0
  630. package/cpp/llama.cpp/src/models/qwen3vl.cpp +141 -0
  631. package/cpp/llama.cpp/src/models/refact.cpp +94 -0
  632. package/cpp/llama.cpp/src/models/rwkv6-base.cpp +162 -0
  633. package/cpp/llama.cpp/src/models/rwkv6.cpp +94 -0
  634. package/cpp/llama.cpp/src/models/rwkv6qwen2.cpp +86 -0
  635. package/cpp/llama.cpp/src/models/rwkv7-base.cpp +135 -0
  636. package/cpp/llama.cpp/src/models/rwkv7.cpp +90 -0
  637. package/cpp/llama.cpp/src/models/seed-oss.cpp +124 -0
  638. package/cpp/llama.cpp/src/models/smallthinker.cpp +120 -0
  639. package/cpp/llama.cpp/src/models/smollm3.cpp +128 -0
  640. package/cpp/llama.cpp/src/models/stablelm.cpp +146 -0
  641. package/cpp/llama.cpp/src/models/starcoder.cpp +100 -0
  642. package/cpp/llama.cpp/src/models/starcoder2.cpp +121 -0
  643. package/cpp/llama.cpp/src/models/t5-dec.cpp +166 -0
  644. package/cpp/llama.cpp/src/models/t5-enc.cpp +96 -0
  645. package/cpp/llama.cpp/src/models/wavtokenizer-dec.cpp +149 -0
  646. package/cpp/llama.cpp/src/models/xverse.cpp +108 -0
  647. package/cpp/llama.cpp/src/unicode.cpp +77 -0
  648. package/cpp/llama.cpp/src/unicode.h +43 -0
  649. package/cpp/llama.cpp/vendor/cpp-httplib/CMakeLists.txt +94 -0
  650. package/cpp/llama.cpp/vendor/cpp-httplib/httplib.cpp +9339 -0
  651. package/cpp/llama.cpp/vendor/cpp-httplib/httplib.h +433 -8222
  652. package/cpp/llama.cpp/vendor/cpp-httplib/patch-boringssl.cmake +6 -0
  653. package/cpp/llama.cpp/vendor/miniaudio/miniaudio.h +4179 -1900
  654. package/cpp/llama.cpp/vendor/minja/chat-template.hpp +9 -2
  655. package/cpp/llama.cpp/vendor/minja/minja.hpp +101 -22
  656. package/ios/include/chat.h +16 -3
  657. package/ios/include/common/minja/chat-template.hpp +9 -2
  658. package/ios/include/common/minja/minja.hpp +101 -22
  659. package/ios/include/common.h +57 -19
  660. package/ios/include/json-schema-to-grammar.h +2 -0
  661. package/ios/include/llama.h +44 -21
  662. package/ios/include/log.h +12 -4
  663. package/ios/include/sampling.h +3 -1
  664. package/ios/libs/llama.xcframework/Info.plist +20 -20
  665. package/ios/libs/llama.xcframework/ios-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  666. package/ios/libs/llama.xcframework/ios-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +6399 -5557
  667. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/ggml-backend.h +19 -1
  668. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/ggml-cpu.h +1 -1
  669. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/ggml-metal.h +1 -6
  670. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/ggml.h +199 -6
  671. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/Headers/llama.h +44 -21
  672. package/ios/libs/llama.xcframework/ios-arm64/llama.framework/llama +0 -0
  673. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  674. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +6362 -5520
  675. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +4813 -4241
  676. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml-backend.h +19 -1
  677. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml-cpu.h +1 -1
  678. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml-metal.h +1 -6
  679. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +199 -6
  680. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/Headers/llama.h +44 -21
  681. package/ios/libs/llama.xcframework/ios-arm64_x86_64-simulator/llama.framework/llama +0 -0
  682. package/package.json +10 -4
  683. package/cpp/llama.cpp/ggml/src/ggml-cann/Doxyfile +0 -2579
  684. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-tile-f16.cu +0 -371
  685. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-tile-f16.cuh +0 -3
  686. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-tile-f32.cu +0 -379
  687. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-tile-f32.cuh +0 -3
  688. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-vec-f16.cuh +0 -495
  689. package/cpp/llama.cpp/ggml/src/ggml-cuda/fattn-vec-f32.cuh +0 -486
  690. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-f16.cu +0 -5
  691. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_0.cu +0 -5
  692. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q4_1.cu +0 -5
  693. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_0.cu +0 -5
  694. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q5_1.cu +0 -5
  695. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-f16-q8_0.cu +0 -5
  696. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-f16.cu +0 -5
  697. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_0.cu +0 -5
  698. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q4_1.cu +0 -5
  699. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_0.cu +0 -5
  700. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q5_1.cu +0 -5
  701. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_0-q8_0.cu +0 -5
  702. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-f16.cu +0 -5
  703. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_0.cu +0 -5
  704. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q4_1.cu +0 -5
  705. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_0.cu +0 -5
  706. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q5_1.cu +0 -5
  707. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q4_1-q8_0.cu +0 -5
  708. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-f16.cu +0 -5
  709. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_0.cu +0 -5
  710. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q4_1.cu +0 -5
  711. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_0.cu +0 -5
  712. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q5_1.cu +0 -5
  713. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_0-q8_0.cu +0 -5
  714. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-f16.cu +0 -5
  715. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_0.cu +0 -5
  716. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q4_1.cu +0 -5
  717. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_0.cu +0 -5
  718. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q5_1.cu +0 -5
  719. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q5_1-q8_0.cu +0 -5
  720. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-f16.cu +0 -5
  721. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_0.cu +0 -5
  722. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q4_1.cu +0 -5
  723. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_0.cu +0 -5
  724. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q5_1.cu +0 -5
  725. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q8_0.cu +0 -5
  726. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs256-f16-f16.cu +0 -5
  727. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-f16.cu +0 -5
  728. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_0.cu +0 -5
  729. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q4_1.cu +0 -5
  730. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_0.cu +0 -5
  731. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q5_1.cu +0 -5
  732. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs64-f16-q8_0.cu +0 -5
  733. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-f16.cu +0 -5
  734. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_0.cu +0 -5
  735. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q4_1.cu +0 -5
  736. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_0.cu +0 -5
  737. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q5_1.cu +0 -5
  738. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-f16-q8_0.cu +0 -5
  739. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-f16.cu +0 -5
  740. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_0.cu +0 -5
  741. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q4_1.cu +0 -5
  742. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_0.cu +0 -5
  743. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q5_1.cu +0 -5
  744. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_0-q8_0.cu +0 -5
  745. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-f16.cu +0 -5
  746. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_0.cu +0 -5
  747. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q4_1.cu +0 -5
  748. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_0.cu +0 -5
  749. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q5_1.cu +0 -5
  750. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q4_1-q8_0.cu +0 -5
  751. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-f16.cu +0 -5
  752. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_0.cu +0 -5
  753. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q4_1.cu +0 -5
  754. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_0.cu +0 -5
  755. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q5_1.cu +0 -5
  756. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_0-q8_0.cu +0 -5
  757. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-f16.cu +0 -5
  758. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_0.cu +0 -5
  759. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q4_1.cu +0 -5
  760. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_0.cu +0 -5
  761. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q5_1.cu +0 -5
  762. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q5_1-q8_0.cu +0 -5
  763. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-f16.cu +0 -5
  764. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_0.cu +0 -5
  765. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q4_1.cu +0 -5
  766. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_0.cu +0 -5
  767. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q5_1.cu +0 -5
  768. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q8_0.cu +0 -5
  769. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs256-f16-f16.cu +0 -5
  770. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-f16.cu +0 -5
  771. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_0.cu +0 -5
  772. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q4_1.cu +0 -5
  773. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_0.cu +0 -5
  774. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q5_1.cu +0 -5
  775. package/cpp/llama.cpp/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs64-f16-q8_0.cu +0 -5
  776. package/cpp/llama.cpp/ggml/src/ggml-metal/ggml-metal.m +0 -6886
  777. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mat_vec_base.comp +0 -154
  778. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/mul_mmq_funcs.comp +0 -105
  779. package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/rope_head.comp +0 -55
  780. package/cpp/llama.cpp/ggml/src/ggml-webgpu/wgsl-shaders/cpy.wgsl +0 -60
  781. package/cpp/llama.cpp/ggml/src/ggml-zdnn/ggml-zdnn-impl.h +0 -97
  782. package/cpp/llama.cpp/models/ggml-vocab-aquila.gguf +0 -0
  783. package/cpp/llama.cpp/models/ggml-vocab-baichuan.gguf +0 -0
  784. package/cpp/llama.cpp/models/ggml-vocab-bert-bge.gguf +0 -0
  785. package/cpp/llama.cpp/models/ggml-vocab-bert-bge.gguf.inp +0 -112
  786. package/cpp/llama.cpp/models/ggml-vocab-bert-bge.gguf.out +0 -46
  787. package/cpp/llama.cpp/models/ggml-vocab-command-r.gguf +0 -0
  788. package/cpp/llama.cpp/models/ggml-vocab-command-r.gguf.inp +0 -112
  789. package/cpp/llama.cpp/models/ggml-vocab-command-r.gguf.out +0 -46
  790. package/cpp/llama.cpp/models/ggml-vocab-deepseek-coder.gguf +0 -0
  791. package/cpp/llama.cpp/models/ggml-vocab-deepseek-coder.gguf.inp +0 -112
  792. package/cpp/llama.cpp/models/ggml-vocab-deepseek-coder.gguf.out +0 -46
  793. package/cpp/llama.cpp/models/ggml-vocab-deepseek-llm.gguf +0 -0
  794. package/cpp/llama.cpp/models/ggml-vocab-deepseek-llm.gguf.inp +0 -112
  795. package/cpp/llama.cpp/models/ggml-vocab-deepseek-llm.gguf.out +0 -46
  796. package/cpp/llama.cpp/models/ggml-vocab-falcon.gguf +0 -0
  797. package/cpp/llama.cpp/models/ggml-vocab-falcon.gguf.inp +0 -112
  798. package/cpp/llama.cpp/models/ggml-vocab-falcon.gguf.out +0 -46
  799. package/cpp/llama.cpp/models/ggml-vocab-gpt-2.gguf +0 -0
  800. package/cpp/llama.cpp/models/ggml-vocab-gpt-2.gguf.inp +0 -112
  801. package/cpp/llama.cpp/models/ggml-vocab-gpt-2.gguf.out +0 -46
  802. package/cpp/llama.cpp/models/ggml-vocab-gpt-neox.gguf +0 -0
  803. package/cpp/llama.cpp/models/ggml-vocab-llama-bpe.gguf +0 -0
  804. package/cpp/llama.cpp/models/ggml-vocab-llama-bpe.gguf.inp +0 -112
  805. package/cpp/llama.cpp/models/ggml-vocab-llama-bpe.gguf.out +0 -46
  806. package/cpp/llama.cpp/models/ggml-vocab-llama-spm.gguf +0 -0
  807. package/cpp/llama.cpp/models/ggml-vocab-llama-spm.gguf.inp +0 -112
  808. package/cpp/llama.cpp/models/ggml-vocab-llama-spm.gguf.out +0 -46
  809. package/cpp/llama.cpp/models/ggml-vocab-mpt.gguf +0 -0
  810. package/cpp/llama.cpp/models/ggml-vocab-mpt.gguf.inp +0 -112
  811. package/cpp/llama.cpp/models/ggml-vocab-mpt.gguf.out +0 -46
  812. package/cpp/llama.cpp/models/ggml-vocab-nomic-bert-moe.gguf +0 -0
  813. package/cpp/llama.cpp/models/ggml-vocab-phi-3.gguf +0 -0
  814. package/cpp/llama.cpp/models/ggml-vocab-phi-3.gguf.inp +0 -112
  815. package/cpp/llama.cpp/models/ggml-vocab-phi-3.gguf.out +0 -46
  816. package/cpp/llama.cpp/models/ggml-vocab-qwen2.gguf +0 -0
  817. package/cpp/llama.cpp/models/ggml-vocab-qwen2.gguf.inp +0 -112
  818. package/cpp/llama.cpp/models/ggml-vocab-qwen2.gguf.out +0 -46
  819. package/cpp/llama.cpp/models/ggml-vocab-refact.gguf +0 -0
  820. package/cpp/llama.cpp/models/ggml-vocab-refact.gguf.inp +0 -112
  821. package/cpp/llama.cpp/models/ggml-vocab-refact.gguf.out +0 -46
  822. package/cpp/llama.cpp/models/ggml-vocab-starcoder.gguf +0 -0
  823. package/cpp/llama.cpp/models/ggml-vocab-starcoder.gguf.inp +0 -112
  824. package/cpp/llama.cpp/models/ggml-vocab-starcoder.gguf.out +0 -46
  825. package/cpp/llama.cpp/models/templates/ByteDance-Seed-OSS.jinja +0 -171
  826. package/cpp/llama.cpp/models/templates/CohereForAI-c4ai-command-r-plus-tool_use.jinja +0 -202
  827. package/cpp/llama.cpp/models/templates/CohereForAI-c4ai-command-r7b-12-2024-tool_use.jinja +0 -156
  828. package/cpp/llama.cpp/models/templates/Mistral-Small-3.2-24B-Instruct-2506.jinja +0 -124
  829. package/cpp/llama.cpp/models/templates/NousResearch-Hermes-2-Pro-Llama-3-8B-tool_use.jinja +0 -152
  830. package/cpp/llama.cpp/models/templates/NousResearch-Hermes-3-Llama-3.1-8B-tool_use.jinja +0 -152
  831. package/cpp/llama.cpp/models/templates/Qwen-QwQ-32B.jinja +0 -62
  832. package/cpp/llama.cpp/models/templates/Qwen-Qwen2.5-7B-Instruct.jinja +0 -54
  833. package/cpp/llama.cpp/models/templates/Qwen-Qwen3-0.6B.jinja +0 -85
  834. package/cpp/llama.cpp/models/templates/README.md +0 -25
  835. package/cpp/llama.cpp/models/templates/deepseek-ai-DeepSeek-R1-Distill-Llama-8B.jinja +0 -1
  836. package/cpp/llama.cpp/models/templates/deepseek-ai-DeepSeek-R1-Distill-Qwen-32B.jinja +0 -1
  837. package/cpp/llama.cpp/models/templates/fireworks-ai-llama-3-firefunction-v2.jinja +0 -57
  838. package/cpp/llama.cpp/models/templates/google-gemma-2-2b-it.jinja +0 -4
  839. package/cpp/llama.cpp/models/templates/ibm-granite-granite-3.3-2B-Instruct.jinja +0 -59
  840. package/cpp/llama.cpp/models/templates/llama-cpp-deepseek-r1.jinja +0 -76
  841. package/cpp/llama.cpp/models/templates/llama-cpp-rwkv-world.jinja +0 -34
  842. package/cpp/llama.cpp/models/templates/meetkai-functionary-medium-v3.1.jinja +0 -58
  843. package/cpp/llama.cpp/models/templates/meetkai-functionary-medium-v3.2.jinja +0 -287
  844. package/cpp/llama.cpp/models/templates/meta-llama-Llama-3.1-8B-Instruct.jinja +0 -109
  845. package/cpp/llama.cpp/models/templates/meta-llama-Llama-3.2-3B-Instruct.jinja +0 -93
  846. package/cpp/llama.cpp/models/templates/meta-llama-Llama-3.3-70B-Instruct.jinja +0 -109
  847. package/cpp/llama.cpp/models/templates/microsoft-Phi-3.5-mini-instruct.jinja +0 -8
  848. package/cpp/llama.cpp/models/templates/mistralai-Mistral-Nemo-Instruct-2407.jinja +0 -87
  849. package/cpp/llama.cpp/models/templates/moonshotai-Kimi-K2.jinja +0 -43
  850. package/cpp/llama.cpp/models/templates/openai-gpt-oss-120b.jinja +0 -331
  851. package/cpp/llama.cpp/models/templates/unsloth-mistral-Devstral-Small-2507.jinja +0 -105
  852. package/cpp/llama.cpp/prompts/LLM-questions.txt +0 -49
  853. package/cpp/llama.cpp/prompts/alpaca.txt +0 -1
  854. package/cpp/llama.cpp/prompts/assistant.txt +0 -31
  855. package/cpp/llama.cpp/prompts/chat-with-baichuan.txt +0 -4
  856. package/cpp/llama.cpp/prompts/chat-with-bob.txt +0 -7
  857. package/cpp/llama.cpp/prompts/chat-with-qwen.txt +0 -1
  858. package/cpp/llama.cpp/prompts/chat-with-vicuna-v0.txt +0 -7
  859. package/cpp/llama.cpp/prompts/chat-with-vicuna-v1.txt +0 -7
  860. package/cpp/llama.cpp/prompts/chat.txt +0 -28
  861. package/cpp/llama.cpp/prompts/dan-modified.txt +0 -1
  862. package/cpp/llama.cpp/prompts/dan.txt +0 -1
  863. package/cpp/llama.cpp/prompts/mnemonics.txt +0 -93
  864. package/cpp/llama.cpp/prompts/parallel-questions.txt +0 -43
  865. package/cpp/llama.cpp/prompts/reason-act.txt +0 -18
  866. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Info.plist +0 -20
  867. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  868. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +0 -5524
  869. package/ios/libs/llama.xcframework/macos-arm64_x86_64/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +0 -4247
  870. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml-alloc.h +0 -76
  871. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml-backend.h +0 -354
  872. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml-blas.h +0 -25
  873. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml-cpu.h +0 -145
  874. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml-metal.h +0 -66
  875. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml-opt.h +0 -256
  876. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/ggml.h +0 -2492
  877. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/gguf.h +0 -202
  878. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Headers/llama.h +0 -1391
  879. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Modules/module.modulemap +0 -17
  880. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Resources/Info.plist +0 -32
  881. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml-alloc.h +0 -76
  882. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml-backend.h +0 -354
  883. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml-blas.h +0 -25
  884. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml-cpu.h +0 -145
  885. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml-metal.h +0 -66
  886. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml-opt.h +0 -256
  887. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/ggml.h +0 -2492
  888. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/gguf.h +0 -202
  889. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Headers/llama.h +0 -1391
  890. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Modules/module.modulemap +0 -17
  891. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/Resources/Info.plist +0 -32
  892. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/A/llama +0 -0
  893. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml-alloc.h +0 -76
  894. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml-backend.h +0 -354
  895. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml-blas.h +0 -25
  896. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml-cpu.h +0 -145
  897. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml-metal.h +0 -66
  898. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml-opt.h +0 -256
  899. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/ggml.h +0 -2492
  900. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/gguf.h +0 -202
  901. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Headers/llama.h +0 -1391
  902. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Modules/module.modulemap +0 -17
  903. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/Resources/Info.plist +0 -32
  904. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/Versions/Current/llama +0 -0
  905. package/ios/libs/llama.xcframework/macos-arm64_x86_64/llama.framework/llama +0 -0
  906. package/ios/libs/llama.xcframework/tvos-arm64/dSYMs/llama.dSYM/Contents/Info.plist +0 -20
  907. package/ios/libs/llama.xcframework/tvos-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  908. package/ios/libs/llama.xcframework/tvos-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +0 -5561
  909. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml-alloc.h +0 -76
  910. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml-backend.h +0 -354
  911. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml-blas.h +0 -25
  912. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml-cpu.h +0 -145
  913. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml-metal.h +0 -66
  914. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml-opt.h +0 -256
  915. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/ggml.h +0 -2492
  916. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/gguf.h +0 -202
  917. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Headers/llama.h +0 -1391
  918. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Info.plist +0 -35
  919. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/Modules/module.modulemap +0 -17
  920. package/ios/libs/llama.xcframework/tvos-arm64/llama.framework/llama +0 -0
  921. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Info.plist +0 -20
  922. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  923. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +0 -5524
  924. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +0 -4246
  925. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml-alloc.h +0 -76
  926. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml-backend.h +0 -354
  927. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml-blas.h +0 -25
  928. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml-cpu.h +0 -145
  929. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml-metal.h +0 -66
  930. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml-opt.h +0 -256
  931. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +0 -2492
  932. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/gguf.h +0 -202
  933. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Headers/llama.h +0 -1391
  934. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Info.plist +0 -35
  935. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/Modules/module.modulemap +0 -17
  936. package/ios/libs/llama.xcframework/tvos-arm64_x86_64-simulator/llama.framework/llama +0 -0
  937. package/ios/libs/llama.xcframework/xros-arm64/dSYMs/llama.dSYM/Contents/Info.plist +0 -20
  938. package/ios/libs/llama.xcframework/xros-arm64/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  939. package/ios/libs/llama.xcframework/xros-arm64/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +0 -5558
  940. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml-alloc.h +0 -76
  941. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml-backend.h +0 -354
  942. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml-blas.h +0 -25
  943. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml-cpu.h +0 -145
  944. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml-metal.h +0 -66
  945. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml-opt.h +0 -256
  946. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/ggml.h +0 -2492
  947. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/gguf.h +0 -202
  948. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Headers/llama.h +0 -1391
  949. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Info.plist +0 -32
  950. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/Modules/module.modulemap +0 -17
  951. package/ios/libs/llama.xcframework/xros-arm64/llama.framework/llama +0 -0
  952. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Info.plist +0 -20
  953. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/DWARF/llama +0 -0
  954. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/aarch64/llama.yml +0 -5520
  955. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/dSYMs/llama.dSYM/Contents/Resources/Relocations/x86_64/llama.yml +0 -4243
  956. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml-alloc.h +0 -76
  957. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml-backend.h +0 -354
  958. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml-blas.h +0 -25
  959. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml-cpu.h +0 -145
  960. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml-metal.h +0 -66
  961. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml-opt.h +0 -256
  962. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/ggml.h +0 -2492
  963. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/gguf.h +0 -202
  964. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Headers/llama.h +0 -1391
  965. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Info.plist +0 -32
  966. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/Modules/module.modulemap +0 -17
  967. package/ios/libs/llama.xcframework/xros-arm64_x86_64-simulator/llama.framework/llama +0 -0
  968. /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{test_bfloat16_support.comp → feature-tests/bfloat16.comp} +0 -0
  969. /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{test_coopmat_support.comp → feature-tests/coopmat.comp} +0 -0
  970. /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{test_coopmat2_support.comp → feature-tests/coopmat2.comp} +0 -0
  971. /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{test_integer_dot_support.comp → feature-tests/integer_dot.comp} +0 -0
  972. /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{generic_head.comp → generic_head.glsl} +0 -0
  973. /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{generic_unary_head.comp → generic_unary_head.glsl} +0 -0
  974. /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{glu_main.comp → glu_main.glsl} +0 -0
  975. /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{rte.comp → rte.glsl} +0 -0
  976. /package/cpp/llama.cpp/ggml/src/ggml-vulkan/vulkan-shaders/{utils.comp → utils.glsl} +0 -0
@@ -0,0 +1,861 @@
1
+ #include "chat.h"
2
+ #include "chat-parser.h"
3
+ #include "common.h"
4
+ #include "json-partial.h"
5
+ #include "json-schema-to-grammar.h"
6
+ #include "log.h"
7
+ #include "regex-partial.h"
8
+
9
+ using json = nlohmann::ordered_json;
10
+
11
+ class xml_toolcall_syntax_exception : public std::runtime_error {
12
+ public:
13
+ xml_toolcall_syntax_exception(const std::string & message) : std::runtime_error(message) {}
14
+ };
15
+
16
+ template<typename T>
17
+ inline void sort_uniq(std::vector<T> &vec) {
18
+ std::sort(vec.begin(), vec.end());
19
+ vec.erase(std::unique(vec.begin(), vec.end()), vec.end());
20
+ }
21
+
22
+ template<typename T>
23
+ inline bool all_space(const T &str) {
24
+ return std::all_of(str.begin(), str.end(), [](unsigned char ch) { return std::isspace(ch); });
25
+ }
26
+
27
+ static size_t utf8_truncate_safe(const std::string_view s) {
28
+ size_t len = s.size();
29
+ if (len == 0) return 0;
30
+ size_t i = len;
31
+ for (size_t back = 0; back < 4 && i > 0; ++back) {
32
+ --i;
33
+ unsigned char c = s[i];
34
+ if ((c & 0x80) == 0) {
35
+ return len;
36
+ } else if ((c & 0xC0) == 0xC0) {
37
+ size_t expected_len = 0;
38
+ if ((c & 0xE0) == 0xC0) expected_len = 2;
39
+ else if ((c & 0xF0) == 0xE0) expected_len = 3;
40
+ else if ((c & 0xF8) == 0xF0) expected_len = 4;
41
+ else return i;
42
+ if (len - i >= expected_len) {
43
+ return len;
44
+ } else {
45
+ return i;
46
+ }
47
+ }
48
+ }
49
+ return len - std::min(len, size_t(3));
50
+ }
51
+
52
+ inline void utf8_truncate_safe_resize(std::string &s) {
53
+ s.resize(utf8_truncate_safe(s));
54
+ }
55
+
56
+ inline std::string_view utf8_truncate_safe_view(const std::string_view s) {
57
+ return s.substr(0, utf8_truncate_safe(s));
58
+ }
59
+
60
+ static std::optional<common_chat_msg_parser::find_regex_result> try_find_2_literal_splited_by_spaces(common_chat_msg_parser & builder, const std::string & literal1, const std::string & literal2) {
61
+ if (literal1.size() == 0) return builder.try_find_literal(literal2);
62
+ const auto saved_pos = builder.pos();
63
+ while (auto res = builder.try_find_literal(literal1)) {
64
+ builder.consume_spaces();
65
+ const auto match_len = std::min(literal2.size(), builder.input().size() - builder.pos());
66
+ if (builder.input().compare(builder.pos(), match_len, literal2, 0, match_len) == 0) {
67
+ if (res->prelude.size() != res->groups[0].begin - saved_pos) {
68
+ res->prelude = builder.str({saved_pos, res->groups[0].begin});
69
+ }
70
+ builder.move_to(builder.pos() + match_len);
71
+ res->groups[0].end = builder.pos();
72
+ GGML_ASSERT(res->groups[0].begin != res->groups[0].end);
73
+ return res;
74
+ }
75
+ builder.move_to(res->groups[0].begin + 1);
76
+ }
77
+ builder.move_to(saved_pos);
78
+ return std::nullopt;
79
+ }
80
+
81
+ /**
82
+ * make a GBNF that accept any strings except those containing any of the forbidden strings.
83
+ */
84
+ std::string make_gbnf_excluding(std::vector<std::string> forbids) {
85
+ constexpr auto charclass_escape = [](unsigned char c) -> std::string {
86
+ if (c == '\\' || c == ']' || c == '^' || c == '-') {
87
+ std::string s = "\\";
88
+ s.push_back((char)c);
89
+ return s;
90
+ }
91
+ if (isprint(c)) {
92
+ return std::string(1, (char)c);
93
+ }
94
+ char buf[16];
95
+ snprintf(buf, 15, "\\x%02X", c);
96
+ return std::string(buf);
97
+ };
98
+ constexpr auto build_expr = [charclass_escape](auto self, const std::vector<std::string>& forbids, int l, int r, int depth) -> std::string {
99
+ std::vector<std::pair<unsigned char, std::pair<int,int>>> children;
100
+ int i = l;
101
+ while (i < r) {
102
+ const std::string &s = forbids[i];
103
+ if ((int)s.size() == depth) {
104
+ ++i;
105
+ continue;
106
+ }
107
+ unsigned char c = (unsigned char)s[depth];
108
+ int j = i;
109
+ while (j < r && (int)forbids[j].size() > depth &&
110
+ (unsigned char)forbids[j][depth] == c) {
111
+ ++j;
112
+ }
113
+ children.push_back({c, {i, j}});
114
+ i = j;
115
+ }
116
+ std::vector<std::string> alts;
117
+ if (!children.empty()) {
118
+ std::string cls;
119
+ for (auto &ch : children) cls += charclass_escape(ch.first);
120
+ alts.push_back(std::string("[^") + cls + "]");
121
+ }
122
+ for (auto &ch : children) {
123
+ std::string childExpr = self(self, forbids, ch.second.first, ch.second.second, depth+1);
124
+ if (!childExpr.empty()) {
125
+ std::string quoted_ch = "\"";
126
+ if (ch.first == '\\') quoted_ch += "\\\\";
127
+ else if (ch.first == '"') quoted_ch += "\\\"";
128
+ else if (isprint(ch.first)) quoted_ch.push_back(ch.first);
129
+ else {
130
+ char buf[16];
131
+ snprintf(buf, 15, "\\x%02X", ch.first);
132
+ quoted_ch += buf;
133
+ }
134
+ quoted_ch += "\"";
135
+ std::string branch = quoted_ch + std::string(" ") + childExpr;
136
+ alts.push_back(branch);
137
+ }
138
+ }
139
+ if (alts.empty()) return "";
140
+ std::ostringstream oss;
141
+ oss << "( ";
142
+ for (size_t k = 0; k < alts.size(); ++k) {
143
+ if (k) oss << " | ";
144
+ oss << alts[k];
145
+ }
146
+ oss << " )";
147
+ return oss.str();
148
+ };
149
+ if (forbids.empty()) return "( . )*";
150
+ sort(forbids.begin(), forbids.end());
151
+ std::string expr = build_expr(build_expr, forbids, 0, forbids.size(), 0);
152
+ if (expr.empty()) {
153
+ std::string cls;
154
+ for (auto &s : forbids) if (!s.empty()) cls += charclass_escape((unsigned char)s[0]);
155
+ expr = std::string("( [^") + cls + "] )";
156
+ }
157
+ if (forbids.size() == 1)
158
+ return expr + "*";
159
+ else
160
+ return std::string("( ") + expr + " )*";
161
+ }
162
+
163
+ /**
164
+ * Build grammar for xml-style tool call
165
+ * form.scope_start and form.scope_end can be empty.
166
+ * Requires data.format for model-specific hacks.
167
+ */
168
+ void build_grammar_xml_tool_call(common_chat_params & data, const json & tools, const struct xml_tool_call_format & form) {
169
+ GGML_ASSERT(!form.tool_start.empty());
170
+ GGML_ASSERT(!form.tool_sep.empty());
171
+ GGML_ASSERT(!form.key_start.empty());
172
+ GGML_ASSERT(!form.val_end.empty());
173
+ GGML_ASSERT(!form.tool_end.empty());
174
+
175
+ std::string key_val_sep = form.key_val_sep;
176
+ if (form.key_val_sep2) {
177
+ key_val_sep += "\n";
178
+ key_val_sep += *form.key_val_sep2;
179
+ }
180
+ GGML_ASSERT(!key_val_sep.empty());
181
+
182
+ if (tools.is_array() && !tools.empty()) {
183
+ data.grammar = build_grammar([&](const common_grammar_builder &builder) {
184
+ auto string_arg_val = form.last_val_end ?
185
+ builder.add_rule("string-arg-val", make_gbnf_excluding({form.val_end, *form.last_val_end})) :
186
+ builder.add_rule("string-arg-val", make_gbnf_excluding({form.val_end}));
187
+
188
+ std::vector<std::string> tool_rules;
189
+ for (const auto & tool : tools) {
190
+ if (!tool.contains("type") || tool.at("type") != "function" || !tool.contains("function")) {
191
+ LOG_WRN("Skipping tool without function: %s", tool.dump(2).c_str());
192
+ continue;
193
+ }
194
+ const auto & function = tool.at("function");
195
+ if (!function.contains("name") || !function.at("name").is_string()) {
196
+ LOG_WRN("Skipping invalid function (invalid name): %s", function.dump(2).c_str());
197
+ continue;
198
+ }
199
+ if (!function.contains("parameters") || !function.at("parameters").is_object()) {
200
+ LOG_WRN("Skipping invalid function (invalid parameters): %s", function.dump(2).c_str());
201
+ continue;
202
+ }
203
+ std::string name = function.at("name");
204
+ auto parameters = function.at("parameters");
205
+ builder.resolve_refs(parameters);
206
+
207
+ struct parameter_rule {
208
+ std::string symbol_name;
209
+ bool is_required;
210
+ };
211
+ std::vector<parameter_rule> arg_rules;
212
+ if (!parameters.contains("properties") || !parameters.at("properties").is_object()) {
213
+ LOG_WRN("Skipping invalid function (invalid properties): %s", function.dump(2).c_str());
214
+ continue;
215
+ } else {
216
+ std::vector<std::string> requiredParameters;
217
+ if (parameters.contains("required")) {
218
+ try { parameters.at("required").get_to(requiredParameters); }
219
+ catch (const std::runtime_error&) {
220
+ LOG_WRN("Invalid function required parameters, ignoring: %s", function.at("required").dump(2).c_str());
221
+ }
222
+ }
223
+ sort_uniq(requiredParameters);
224
+ for (const auto & [key, value] : parameters.at("properties").items()) {
225
+ std::string quoted_key = key;
226
+ bool required = std::binary_search(requiredParameters.begin(), requiredParameters.end(), key);
227
+ if (form.key_start.back() == '"' && key_val_sep[0] == '"') {
228
+ quoted_key = gbnf_format_literal(key);
229
+ quoted_key = quoted_key.substr(1, quoted_key.size() - 2);
230
+ }
231
+ arg_rules.push_back(parameter_rule {builder.add_rule("func-" + name + "-kv-" + key,
232
+ gbnf_format_literal(form.key_start) + " " +
233
+ gbnf_format_literal(quoted_key) + " " +
234
+ gbnf_format_literal(key_val_sep) + " " +
235
+ ((value.contains("type") && value["type"].is_string() && value["type"] == "string" && (!form.raw_argval || *form.raw_argval)) ?
236
+ (form.raw_argval ?
237
+ string_arg_val :
238
+ "( " + string_arg_val + " | " + builder.add_schema(name + "-arg-" + key, value) + " )"
239
+ ) :
240
+ builder.add_schema(name + "-arg-" + key, value)
241
+ )
242
+ ), required});
243
+ }
244
+ }
245
+
246
+ auto next_arg_with_sep = builder.add_rule(name + "-last-arg-end", form.last_val_end ? gbnf_format_literal(*form.last_val_end) : gbnf_format_literal(form.val_end));
247
+ decltype(next_arg_with_sep) next_arg = "\"\"";
248
+ for (auto i = arg_rules.size() - 1; /* i >= 0 && */ i < arg_rules.size(); --i) {
249
+ std::string include_this_arg = arg_rules[i].symbol_name + " " + next_arg_with_sep;
250
+ next_arg = builder.add_rule(name + "-arg-after-" + std::to_string(i), arg_rules[i].is_required ?
251
+ include_this_arg : "( " + include_this_arg + " ) | " + next_arg
252
+ );
253
+ include_this_arg = gbnf_format_literal(form.val_end) + " " + include_this_arg;
254
+ next_arg_with_sep = builder.add_rule(name + "-arg-after-" + std::to_string(i) + "-with-sep", arg_rules[i].is_required ?
255
+ include_this_arg : "( " + include_this_arg + " ) | " + next_arg_with_sep
256
+ );
257
+ }
258
+
259
+ std::string quoted_name = name;
260
+ if (form.tool_start.back() == '"' && form.tool_sep[0] == '"') {
261
+ quoted_name = gbnf_format_literal(name);
262
+ quoted_name = quoted_name.substr(1, quoted_name.size() - 2);
263
+ }
264
+ quoted_name = gbnf_format_literal(quoted_name);
265
+ // Kimi-K2 uses functions.{{ tool_call['function']['name'] }}:{{ loop.index }} as function name
266
+ if (data.format == COMMON_CHAT_FORMAT_KIMI_K2) {
267
+ quoted_name = "\"functions.\" " + quoted_name + " \":\" [0-9]+";
268
+ }
269
+ tool_rules.push_back(builder.add_rule(name + "-call",
270
+ gbnf_format_literal(form.tool_start) + " " +
271
+ quoted_name + " " +
272
+ gbnf_format_literal(form.tool_sep) + " " +
273
+ next_arg
274
+ ));
275
+ }
276
+
277
+ auto tool_call_once = builder.add_rule("root-tool-call-once", string_join(tool_rules, " | "));
278
+ auto tool_call_more = builder.add_rule("root-tool-call-more", gbnf_format_literal(form.tool_end) + " " + tool_call_once);
279
+ auto call_end = builder.add_rule("root-call-end", form.last_tool_end ? gbnf_format_literal(*form.last_tool_end) : gbnf_format_literal(form.tool_end));
280
+ auto tool_call_multiple_with_end = builder.add_rule("root-tool-call-multiple-with-end", tool_call_once + " " + tool_call_more + "* " + call_end);
281
+ builder.add_rule("root",
282
+ (form.scope_start.empty() ? "" : gbnf_format_literal(form.scope_start) + " ") +
283
+ tool_call_multiple_with_end + "?" +
284
+ (form.scope_end.empty() ? "" : " " + gbnf_format_literal(form.scope_end))
285
+ );
286
+ });
287
+
288
+ // grammar trigger for tool call
289
+ data.grammar_triggers.push_back({ COMMON_GRAMMAR_TRIGGER_TYPE_WORD, form.scope_start + form.tool_start });
290
+ }
291
+ }
292
+
293
+ /**
294
+ * Parse XML-Style tool call for given xml_tool_call_format. Return false for invalid syntax and get the position untouched.
295
+ * Throws xml_toolcall_syntax_exception if there is invalid syntax and cannot recover the original status for common_chat_msg_parser.
296
+ * form.scope_start, form.tool_sep and form.scope_end can be empty.
297
+ */
298
+ inline bool parse_xml_tool_calls(common_chat_msg_parser & builder, const struct xml_tool_call_format & form) {
299
+ GGML_ASSERT(!form.tool_start.empty());
300
+ GGML_ASSERT(!form.key_start.empty());
301
+ GGML_ASSERT(!form.key_val_sep.empty());
302
+ GGML_ASSERT(!form.val_end.empty());
303
+ GGML_ASSERT(!form.tool_end.empty());
304
+
305
+ // Helper to choose return false or throw error
306
+ constexpr auto return_error = [](common_chat_msg_parser & builder, auto &start_pos, const bool &recovery) {
307
+ LOG_DBG("Failed to parse XML-Style tool call at position: %s\n", gbnf_format_literal(builder.consume_rest().substr(0, 20)).c_str());
308
+ if (recovery) {
309
+ builder.move_to(start_pos);
310
+ return false;
311
+ } else throw xml_toolcall_syntax_exception("Tool call parsing failed with unrecoverable errors. Try using a grammar to constrain the model’s output.");
312
+ };
313
+ // Drop substring from needle to end from a JSON
314
+ constexpr auto partial_json = [](std::string &json_str, std::string_view needle = "XML_TOOL_CALL_PARTIAL_FLAG") {
315
+ auto pos = json_str.rfind(needle);
316
+ if (pos == std::string::npos) {
317
+ return false;
318
+ }
319
+ for (auto i = pos + needle.size(); i < json_str.size(); ++i) {
320
+ unsigned char ch = static_cast<unsigned char>(json_str[i]);
321
+ if (ch != '\'' && ch != '"' && ch != '}' && ch != ':' && !std::isspace(ch)) {
322
+ return false;
323
+ }
324
+ }
325
+ if (pos != 0 && json_str[pos - 1] == '"') {
326
+ --pos;
327
+ }
328
+ json_str.resize(pos);
329
+ return true;
330
+ };
331
+ // Helper to generate a partial argument JSON
332
+ constexpr auto gen_partial_json = [partial_json](auto set_partial_arg, auto &arguments, auto &builder, auto &function_name) {
333
+ auto rest = builder.consume_rest();
334
+ utf8_truncate_safe_resize(rest);
335
+ set_partial_arg(rest, "XML_TOOL_CALL_PARTIAL_FLAG");
336
+ auto tool_str = arguments.dump();
337
+ if (partial_json(tool_str)) {
338
+ if (builder.add_tool_call(function_name, "", tool_str)) {
339
+ return;
340
+ }
341
+ }
342
+ LOG_DBG("Failed to parse partial XML-Style tool call, fallback to non-partial: %s\n", tool_str.c_str());
343
+ };
344
+ // Helper to find a close (because there may be form.last_val_end or form.last_tool_end)
345
+ constexpr auto try_find_close = [](
346
+ common_chat_msg_parser & builder,
347
+ const std::string & end,
348
+ const std::optional<std::string> & alt_end,
349
+ const std::string & end_next,
350
+ const std::optional<std::string> & alt_end_next
351
+ ) {
352
+ auto saved_pos = builder.pos();
353
+ auto tc = builder.try_find_literal(end);
354
+ auto val_end_size = end.size();
355
+ if (alt_end) {
356
+ auto pos_1 = builder.pos();
357
+ builder.move_to(saved_pos);
358
+ auto tc2 = try_find_2_literal_splited_by_spaces(builder, *alt_end, end_next);
359
+ if (alt_end_next) {
360
+ builder.move_to(saved_pos);
361
+ auto tc3 = try_find_2_literal_splited_by_spaces(builder, *alt_end, *alt_end_next);
362
+ if (tc3 && (!tc2 || tc2->prelude.size() > tc3->prelude.size())) {
363
+ tc2 = tc3;
364
+ }
365
+ }
366
+ if (tc2 && (!tc || tc->prelude.size() > tc2->prelude.size())) {
367
+ tc = tc2;
368
+ tc->groups[0].end = std::min(builder.input().size(), tc->groups[0].begin + alt_end->size());
369
+ builder.move_to(tc->groups[0].end);
370
+ val_end_size = alt_end->size();
371
+ } else {
372
+ builder.move_to(pos_1);
373
+ }
374
+ }
375
+ return std::make_pair(val_end_size, tc);
376
+ };
377
+ // Helper to find a val_end or last_val_end, returns matched pattern size
378
+ const auto try_find_val_end = [try_find_close, &builder, &form]() {
379
+ return try_find_close(builder, form.val_end, form.last_val_end, form.tool_end, form.last_tool_end);
380
+ };
381
+ // Helper to find a tool_end or last_tool_end, returns matched pattern size
382
+ const auto try_find_tool_end = [try_find_close, &builder, &form]() {
383
+ return try_find_close(builder, form.tool_end, form.last_tool_end, form.scope_end, std::nullopt);
384
+ };
385
+
386
+ bool recovery = true;
387
+ const auto start_pos = builder.pos();
388
+ if (!all_space(form.scope_start)) {
389
+ if (auto tc = builder.try_find_literal(form.scope_start)) {
390
+ if (all_space(tc->prelude)) {
391
+ if (form.scope_start.size() != tc->groups[0].end - tc->groups[0].begin)
392
+ throw common_chat_msg_partial_exception("Partial literal: " + gbnf_format_literal(form.scope_start));
393
+ } else {
394
+ builder.move_to(start_pos);
395
+ return false;
396
+ }
397
+ } else return false;
398
+ }
399
+ while (auto tc = builder.try_find_literal(form.tool_start)) {
400
+ if (!all_space(tc->prelude)) {
401
+ LOG_DBG("XML-Style tool call: Expected %s, but found %s, trying to match next pattern\n",
402
+ gbnf_format_literal(form.tool_start).c_str(),
403
+ gbnf_format_literal(tc->prelude).c_str()
404
+ );
405
+ builder.move_to(tc->groups[0].begin - tc->prelude.size());
406
+ break;
407
+ }
408
+
409
+ // Find tool name
410
+ auto func_name = builder.try_find_literal(all_space(form.tool_sep) ? form.key_start : form.tool_sep);
411
+ if (!func_name) {
412
+ auto [sz, tc] = try_find_tool_end();
413
+ func_name = tc;
414
+ }
415
+ if (!func_name) {
416
+ // Partial tool name not supported
417
+ throw common_chat_msg_partial_exception("incomplete tool_call");
418
+ }
419
+ // If the model generate multiple tool call and the first tool call has no argument
420
+ if (func_name->prelude.find(form.tool_end) != std::string::npos || (form.last_tool_end ? func_name->prelude.find(*form.last_tool_end) != std::string::npos : false)) {
421
+ builder.move_to(func_name->groups[0].begin - func_name->prelude.size());
422
+ auto [sz, tc] = try_find_tool_end();
423
+ func_name = tc;
424
+ }
425
+
426
+ // Parse tool name
427
+ builder.move_to(all_space(form.tool_sep) ? func_name->groups[0].begin : func_name->groups[0].end);
428
+ std::string function_name = string_strip(func_name->prelude);
429
+ // Kimi-K2 uses functions.{{ tool_call['function']['name'] }}:{{ loop.index }} as function name
430
+ if (builder.syntax().format == COMMON_CHAT_FORMAT_KIMI_K2) {
431
+ if (string_starts_with(function_name, "functions.")) {
432
+ static const std::regex re(":\\d+$");
433
+ if (std::regex_search(function_name, re)) {
434
+ function_name = function_name.substr(10, function_name.rfind(":") - 10);
435
+ }
436
+ }
437
+ }
438
+
439
+ // Argument JSON
440
+ json arguments = json::object();
441
+
442
+ // Helper to generate a partial argument JSON
443
+ const auto gen_partial_args = [&](auto set_partial_arg) {
444
+ gen_partial_json(set_partial_arg, arguments, builder, function_name);
445
+ };
446
+
447
+ // Parse all arg_key/arg_value pairs
448
+ while (auto tc = builder.try_find_literal(form.key_start)) {
449
+ if (!all_space(tc->prelude)) {
450
+ LOG_DBG("XML-Style tool call: Expected %s, but found %s, trying to match next pattern\n",
451
+ gbnf_format_literal(form.key_start).c_str(),
452
+ gbnf_format_literal(tc->prelude).c_str()
453
+ );
454
+ builder.move_to(tc->groups[0].begin - tc->prelude.size());
455
+ break;
456
+ }
457
+ if (tc->groups[0].end - tc->groups[0].begin != form.key_start.size()) {
458
+ auto tool_call_arg = arguments.dump();
459
+ if (tool_call_arg.size() != 0 && tool_call_arg[tool_call_arg.size() - 1] == '}') {
460
+ tool_call_arg.resize(tool_call_arg.size() - 1);
461
+ }
462
+ builder.add_tool_call(function_name, "", tool_call_arg);
463
+ throw common_chat_msg_partial_exception("Partial literal: " + gbnf_format_literal(form.key_start));
464
+ }
465
+
466
+ // Parse arg_key
467
+ auto key_res = builder.try_find_literal(form.key_val_sep);
468
+ if (!key_res) {
469
+ gen_partial_args([&](auto &rest, auto &needle) {arguments[rest + needle] = "";});
470
+ throw common_chat_msg_partial_exception("Expected " + gbnf_format_literal(form.key_val_sep) + " after " + gbnf_format_literal(form.key_start));
471
+ }
472
+ if (key_res->groups[0].end - key_res->groups[0].begin != form.key_val_sep.size()) {
473
+ gen_partial_args([&](auto &, auto &needle) {arguments[key_res->prelude + needle] = "";});
474
+ throw common_chat_msg_partial_exception("Partial literal: " + gbnf_format_literal(form.key_val_sep));
475
+ }
476
+ auto &key = key_res->prelude;
477
+ recovery = false;
478
+
479
+ // Parse arg_value
480
+ if (form.key_val_sep2) {
481
+ if (auto tc = builder.try_find_literal(*form.key_val_sep2)) {
482
+ if (!all_space(tc->prelude)) {
483
+ LOG_DBG("Failed to parse XML-Style tool call: Unexcepted %s between %s and %s\n",
484
+ gbnf_format_literal(tc->prelude).c_str(),
485
+ gbnf_format_literal(form.key_val_sep).c_str(),
486
+ gbnf_format_literal(*form.key_val_sep2).c_str()
487
+ );
488
+ return return_error(builder, start_pos, false);
489
+ }
490
+ if (tc->groups[0].end - tc->groups[0].begin != form.key_val_sep2->size()) {
491
+ gen_partial_args([&](auto &, auto &needle) {arguments[key] = needle;});
492
+ throw common_chat_msg_partial_exception("Partial literal: " + gbnf_format_literal(*form.key_val_sep2));
493
+ }
494
+ } else {
495
+ gen_partial_args([&](auto &, auto &needle) {arguments[key] = needle;});
496
+ throw common_chat_msg_partial_exception("Expected " + gbnf_format_literal(*form.key_val_sep2) + " after " + gbnf_format_literal(form.key_val_sep));
497
+ }
498
+ }
499
+ auto val_start = builder.pos();
500
+
501
+ // Test if arg_val is a partial JSON
502
+ std::optional<common_json> value_json = std::nullopt;
503
+ if (!form.raw_argval || !*form.raw_argval) {
504
+ try { value_json = builder.try_consume_json(); }
505
+ catch (const std::runtime_error&) { builder.move_to(val_start); }
506
+ // TODO: Delete this when json_partial adds top-level support for null/true/false
507
+ if (builder.pos() == val_start) {
508
+ const static std::regex number_regex(R"([0-9-][0-9]*(\.\d*)?([eE][+-]?\d*)?)");
509
+ builder.consume_spaces();
510
+ std::string_view sv = utf8_truncate_safe_view(builder.input());
511
+ sv.remove_prefix(builder.pos());
512
+ std::string rest = "a";
513
+ if (sv.size() < 6) rest = sv;
514
+ if (string_starts_with("null", rest) || string_starts_with("true", rest) || string_starts_with("false", rest) || std::regex_match(sv.begin(), sv.end(), number_regex)) {
515
+ value_json = {123, {"123", "123"}};
516
+ builder.consume_rest();
517
+ } else {
518
+ builder.move_to(val_start);
519
+ }
520
+ }
521
+ }
522
+
523
+ // If it is a JSON and followed by </arg_value>, parse as json
524
+ // cannot support streaming because it may be a plain text starting with JSON
525
+ if (value_json) {
526
+ auto json_end = builder.pos();
527
+ builder.consume_spaces();
528
+ if (builder.pos() == builder.input().size()) {
529
+ if (form.raw_argval && !*form.raw_argval && (value_json->json.is_string() || value_json->json.is_object() || value_json->json.is_array())) {
530
+ arguments[key] = value_json->json;
531
+ auto json_str = arguments.dump();
532
+ if (!value_json->healing_marker.json_dump_marker.empty()) {
533
+ GGML_ASSERT(std::string::npos != json_str.rfind(value_json->healing_marker.json_dump_marker));
534
+ json_str.resize(json_str.rfind(value_json->healing_marker.json_dump_marker));
535
+ } else {
536
+ GGML_ASSERT(json_str.back() == '}');
537
+ json_str.resize(json_str.size() - 1);
538
+ }
539
+ builder.add_tool_call(function_name, "", json_str);
540
+ } else {
541
+ gen_partial_args([&](auto &, auto &needle) {arguments[key] = needle;});
542
+ }
543
+ LOG_DBG("Possible JSON arg_value: %s\n", value_json->json.dump().c_str());
544
+ throw common_chat_msg_partial_exception("JSON arg_value detected. Waiting for more tokens for validations.");
545
+ }
546
+ builder.move_to(json_end);
547
+ auto [val_end_size, tc] = try_find_val_end();
548
+ if (tc && all_space(tc->prelude) && value_json->healing_marker.marker.empty()) {
549
+ if (tc->groups[0].end - tc->groups[0].begin != val_end_size) {
550
+ gen_partial_args([&](auto &, auto &needle) {arguments[key] = needle;});
551
+ LOG_DBG("Possible terminated JSON arg_value: %s\n", value_json->json.dump().c_str());
552
+ throw common_chat_msg_partial_exception("Partial literal: " + gbnf_format_literal(form.val_end) + (form.last_val_end ? gbnf_format_literal(*form.last_val_end) : ""));
553
+ } else arguments[key] = value_json->json;
554
+ } else builder.move_to(val_start);
555
+ }
556
+
557
+ // If not, parse as plain text
558
+ if (val_start == builder.pos()) {
559
+ if (auto [val_end_size, value_plain] = try_find_val_end(); value_plain) {
560
+ auto &value_str = value_plain->prelude;
561
+ if (form.trim_raw_argval) value_str = string_strip(value_str);
562
+ if (value_plain->groups[0].end - value_plain->groups[0].begin != val_end_size) {
563
+ gen_partial_args([&](auto &, auto &needle) {arguments[key] = value_str + needle;});
564
+ throw common_chat_msg_partial_exception(
565
+ "Expected " + gbnf_format_literal(form.val_end) +
566
+ " after " + gbnf_format_literal(form.key_val_sep) +
567
+ (form.key_val_sep2 ? " " + gbnf_format_literal(*form.key_val_sep2) : "")
568
+ );
569
+ }
570
+ arguments[key] = value_str;
571
+ } else {
572
+ if (form.trim_raw_argval) {
573
+ gen_partial_args([&](auto &rest, auto &needle) {arguments[key] = string_strip(rest) + needle;});
574
+ } else {
575
+ gen_partial_args([&](auto &rest, auto &needle) {arguments[key] = rest + needle;});
576
+ }
577
+ throw common_chat_msg_partial_exception(
578
+ "Expected " + gbnf_format_literal(form.val_end) +
579
+ " after " + gbnf_format_literal(form.key_val_sep) +
580
+ (form.key_val_sep2 ? " " + gbnf_format_literal(*form.key_val_sep2) : "")
581
+ );
582
+ }
583
+ }
584
+ }
585
+
586
+ // Consume closing tag
587
+ if (auto [tool_end_size, tc] = try_find_tool_end(); tc) {
588
+ if (!all_space(tc->prelude)) {
589
+ LOG_DBG("Failed to parse XML-Style tool call: Expected %s, but found %s\n",
590
+ gbnf_format_literal(form.tool_end).c_str(),
591
+ gbnf_format_literal(tc->prelude).c_str()
592
+ );
593
+ return return_error(builder, start_pos, recovery);
594
+ }
595
+ if (tc->groups[0].end - tc->groups[0].begin == tool_end_size) {
596
+ // Add the parsed tool call
597
+ if (!builder.add_tool_call(function_name, "", arguments.dump())) {
598
+ throw common_chat_msg_partial_exception("Failed to add XML-Style tool call");
599
+ }
600
+ recovery = false;
601
+ continue;
602
+ }
603
+ }
604
+
605
+ auto tool_call_arg = arguments.dump();
606
+ if (tool_call_arg.size() != 0 && tool_call_arg[tool_call_arg.size() - 1] == '}') {
607
+ tool_call_arg.resize(tool_call_arg.size() - 1);
608
+ }
609
+ builder.add_tool_call(function_name, "", tool_call_arg);
610
+ throw common_chat_msg_partial_exception("Expected " + gbnf_format_literal(form.tool_end) + " after " + gbnf_format_literal(form.val_end));
611
+ }
612
+ if (auto tc = builder.try_find_literal(form.scope_end)) {
613
+ if (!all_space(tc->prelude)) {
614
+ LOG_DBG("Failed to parse XML-Style tool call: Expected %s, but found %s\n",
615
+ gbnf_format_literal(form.scope_end).c_str(),
616
+ gbnf_format_literal(tc->prelude).c_str()
617
+ );
618
+ return return_error(builder, start_pos, recovery);
619
+ }
620
+ } else {
621
+ if (all_space(form.scope_end)) return true;
622
+ builder.consume_spaces();
623
+ if (builder.pos() == builder.input().size())
624
+ throw common_chat_msg_partial_exception("incomplete tool calls");
625
+ LOG_DBG("Failed to parse XML-Style tool call: Expected %s, but found %s\n",
626
+ gbnf_format_literal(form.scope_end).c_str(),
627
+ gbnf_format_literal(builder.consume_rest()).c_str()
628
+ );
629
+ return return_error(builder, start_pos, recovery);
630
+ }
631
+
632
+ return true;
633
+ }
634
+
635
+ /**
636
+ * Parse XML-Style tool call for given xml_tool_call_format. Return false for invalid syntax and get the position untouched.
637
+ * May cause std::runtime_error if there is invalid syntax because partial valid tool call is already sent out to client.
638
+ * form.scope_start, form.tool_sep and form.scope_end can be empty.
639
+ */
640
+ bool common_chat_msg_parser::try_consume_xml_tool_calls(const struct xml_tool_call_format & form) {
641
+ auto pos = pos_;
642
+ auto tsize = result_.tool_calls.size();
643
+ try { return parse_xml_tool_calls(*this, form); }
644
+ catch (const xml_toolcall_syntax_exception&) {}
645
+ move_to(pos);
646
+ result_.tool_calls.resize(tsize);
647
+ return false;
648
+ }
649
+
650
+ /**
651
+ * Parse content uses reasoning and XML-Style tool call
652
+ * TODO: Note that form.allow_toolcall_in_think is not tested yet. If anyone confirms it works, this comment can be removed.
653
+ */
654
+ inline void parse_msg_with_xml_tool_calls(common_chat_msg_parser & builder, const struct xml_tool_call_format & form, const std::string & start_think = "<think>", const std::string & end_think = "</think>") {
655
+ constexpr auto rstrip = [](std::string &s) {
656
+ s.resize(std::distance(s.begin(), std::find_if(s.rbegin(), s.rend(), [](unsigned char ch) { return !std::isspace(ch); }).base()));
657
+ };
658
+ // Erase substring from l to r, along with additional spaces nearby
659
+ constexpr auto erase_spaces = [](auto &str, size_t l, size_t r) {
660
+ while (/* l > -1 && */ --l < str.size() && std::isspace(static_cast<unsigned char>(str[l])));
661
+ ++l;
662
+ while (++r < str.size() && std::isspace(static_cast<unsigned char>(str[r])));
663
+ if (l < r) str[l] = '\n';
664
+ if (l + 1 < r) str[l + 1] = '\n';
665
+ if (l != 0) l += 2;
666
+ str.erase(l, r - l);
667
+ return l;
668
+ };
669
+ constexpr auto trim_suffix = [](std::string &content, std::initializer_list<std::string_view> list) {
670
+ auto best_match = content.size();
671
+ for (auto pattern: list) {
672
+ if (pattern.size() == 0) continue;
673
+ for (auto match_idx = content.size() - std::min(pattern.size(), content.size()); content.size() > match_idx; match_idx++) {
674
+ auto match_len = content.size() - match_idx;
675
+ if (content.compare(match_idx, match_len, pattern.data(), match_len) == 0 && best_match > match_idx) {
676
+ best_match = match_idx;
677
+ }
678
+ }
679
+ }
680
+ if (content.size() > best_match) {
681
+ content.erase(best_match);
682
+ }
683
+ };
684
+ const auto trim_potential_partial_word = [&start_think, &end_think, &form, trim_suffix](std::string &content) {
685
+ return trim_suffix(content, {
686
+ start_think, end_think, form.scope_start, form.tool_start, form.tool_sep, form.key_start,
687
+ form.key_val_sep, form.key_val_sep2 ? form.key_val_sep2->c_str() : "",
688
+ form.val_end, form.last_val_end ? form.last_val_end->c_str() : "",
689
+ form.tool_end, form.last_tool_end ? form.last_tool_end->c_str() : "",
690
+ form.scope_end
691
+ });
692
+ };
693
+
694
+
695
+ // Trim leading spaces without affecting keyword matching
696
+ static const common_regex spaces_regex("\\s*");
697
+ {
698
+ auto tc = builder.consume_regex(spaces_regex);
699
+ auto spaces = builder.str(tc.groups[0]);
700
+ auto s1 = spaces.size();
701
+ trim_potential_partial_word(spaces);
702
+ auto s2 = spaces.size();
703
+ builder.move_to(builder.pos() - (s1 - s2));
704
+ }
705
+
706
+ // Parse content
707
+ bool reasoning_unclosed = builder.syntax().thinking_forced_open;
708
+ std::string unclosed_reasoning_content("");
709
+ for (;;) {
710
+ auto tc = try_find_2_literal_splited_by_spaces(builder, form.scope_start, form.tool_start);
711
+ std::string content;
712
+ std::string tool_call_start;
713
+
714
+ if (tc) {
715
+ content = std::move(tc->prelude);
716
+ tool_call_start = builder.str(tc->groups[0]);
717
+ LOG_DBG("Matched tool start: %s\n", gbnf_format_literal(tool_call_start).c_str());
718
+ } else {
719
+ content = builder.consume_rest();
720
+ utf8_truncate_safe_resize(content);
721
+ }
722
+
723
+ // Handle unclosed think block
724
+ if (reasoning_unclosed) {
725
+ if (auto pos = content.find(end_think); pos == std::string::npos && builder.pos() != builder.input().size()) {
726
+ unclosed_reasoning_content += content;
727
+ if (form.allow_toolcall_in_think) {
728
+ builder.move_to(tc->groups[0].begin);
729
+ if (!builder.try_consume_xml_tool_calls(form)) {
730
+ unclosed_reasoning_content += tool_call_start;
731
+ builder.move_to(tc->groups[0].end);
732
+ }
733
+ } else {
734
+ unclosed_reasoning_content += tool_call_start;
735
+ }
736
+ continue;
737
+ } else {
738
+ reasoning_unclosed = false;
739
+ std::string reasoning_content;
740
+ if (pos == std::string::npos) {
741
+ reasoning_content = std::move(content);
742
+ } else {
743
+ reasoning_content = content.substr(0, pos);
744
+ content.erase(0, pos + end_think.size());
745
+ }
746
+ if (builder.pos() == builder.input().size() && all_space(content)) {
747
+ rstrip(reasoning_content);
748
+ trim_potential_partial_word(reasoning_content);
749
+ rstrip(reasoning_content);
750
+ if (reasoning_content.empty()) {
751
+ rstrip(unclosed_reasoning_content);
752
+ trim_potential_partial_word(unclosed_reasoning_content);
753
+ rstrip(unclosed_reasoning_content);
754
+ if (unclosed_reasoning_content.empty()) continue;
755
+ }
756
+ }
757
+ if (builder.syntax().reasoning_format == COMMON_REASONING_FORMAT_NONE || builder.syntax().reasoning_in_content) {
758
+ builder.add_content(start_think);
759
+ builder.add_content(unclosed_reasoning_content);
760
+ builder.add_content(reasoning_content);
761
+ if (builder.pos() != builder.input().size() || !all_space(content))
762
+ builder.add_content(end_think);
763
+ } else {
764
+ builder.add_reasoning_content(unclosed_reasoning_content);
765
+ builder.add_reasoning_content(reasoning_content);
766
+ }
767
+ unclosed_reasoning_content.clear();
768
+ }
769
+ }
770
+
771
+ // Handle multiple think block
772
+ bool toolcall_in_think = false;
773
+ for (auto think_start = content.find(start_think); think_start != std::string::npos; think_start = content.find(start_think, think_start)) {
774
+ if (auto think_end = content.find(end_think, think_start + start_think.size()); think_end != std::string::npos) {
775
+ if (builder.syntax().reasoning_format != COMMON_REASONING_FORMAT_NONE && !builder.syntax().reasoning_in_content) {
776
+ auto reasoning_content = content.substr(think_start + start_think.size(), think_end - think_start - start_think.size());
777
+ builder.add_reasoning_content(reasoning_content);
778
+ think_start = erase_spaces(content, think_start, think_end + end_think.size() - 1);
779
+ } else {
780
+ think_start = think_end + end_think.size() - 1;
781
+ }
782
+ } else {
783
+ // This <tool_call> start is in thinking block, skip this tool call
784
+ auto pos = think_start + start_think.size();
785
+ unclosed_reasoning_content = content.substr(pos) + tool_call_start;
786
+ reasoning_unclosed = true;
787
+ content.resize(think_start);
788
+ toolcall_in_think = true;
789
+ }
790
+ }
791
+
792
+ if (builder.syntax().reasoning_format != COMMON_REASONING_FORMAT_NONE && !builder.syntax().reasoning_in_content) {
793
+ rstrip(content);
794
+ // Handle unclosed </think> token from content: delete all </think> token
795
+ if (auto pos = content.rfind(end_think); pos != std::string::npos) {
796
+ while (pos != std::string::npos) {
797
+ pos = erase_spaces(content, pos, pos + end_think.size() - 1);
798
+ pos = content.rfind(end_think, pos);
799
+ }
800
+ }
801
+ // Strip if needed
802
+ if (content.size() > 0 && std::isspace(static_cast<unsigned char>(content[0]))) {
803
+ content = string_strip(content);
804
+ }
805
+ }
806
+
807
+ // remove potential partial suffix
808
+ if (content.size() > 0 && builder.pos() == builder.input().size() && unclosed_reasoning_content.empty()) {
809
+ rstrip(content);
810
+ trim_potential_partial_word(content);
811
+ rstrip(content);
812
+ }
813
+
814
+ // Add content
815
+ if (content.size() != 0) {
816
+ // If there are multiple content blocks
817
+ if (builder.syntax().reasoning_format != COMMON_REASONING_FORMAT_NONE && !builder.syntax().reasoning_in_content && builder.result().content.size() != 0) {
818
+ builder.add_content("\n\n");
819
+ }
820
+ builder.add_content(content);
821
+ }
822
+
823
+ // This <tool_call> start is in thinking block, skip this tool call
824
+ if (toolcall_in_think && !form.allow_toolcall_in_think) {
825
+ continue;
826
+ }
827
+
828
+ // There is no tool call and all content is parsed
829
+ if (!tc) {
830
+ GGML_ASSERT(builder.pos() == builder.input().size());
831
+ GGML_ASSERT(unclosed_reasoning_content.empty());
832
+ GGML_ASSERT(!reasoning_unclosed);
833
+ break;
834
+ }
835
+
836
+ builder.move_to(tc->groups[0].begin);
837
+ if (builder.try_consume_xml_tool_calls(form)) {
838
+ auto end_of_tool = builder.pos();
839
+ builder.consume_spaces();
840
+ if (builder.pos() != builder.input().size()) {
841
+ builder.move_to(end_of_tool);
842
+ if (!builder.result().content.empty()) {
843
+ builder.add_content("\n\n");
844
+ }
845
+ }
846
+ } else {
847
+ static const common_regex next_char_regex(".");
848
+ auto c = builder.str(builder.consume_regex(next_char_regex).groups[0]);
849
+ rstrip(c);
850
+ builder.add_content(c);
851
+ }
852
+ }
853
+ }
854
+
855
+ /**
856
+ * Parse content uses reasoning and XML-Style tool call
857
+ * TODO: Note that form.allow_toolcall_in_think is not tested yet. If anyone confirms it works, this comment can be removed.
858
+ */
859
+ void common_chat_msg_parser::consume_reasoning_with_xml_tool_calls(const struct xml_tool_call_format & form, const std::string & start_think, const std::string & end_think) {
860
+ parse_msg_with_xml_tool_calls(*this, form, start_think, end_think);
861
+ }