@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
@@ -31,7 +31,7 @@
31
31
  "SRC1_TYPE": "f32",
32
32
  "BLOCK_SIZE": 32
33
33
  },
34
- "DECLS": ["BYTE_HELPERS", "Q4_0"]
34
+ "DECLS": ["BYTE_HELPERS", "Q4_0_T", "Q4_0"]
35
35
  },
36
36
  {
37
37
  "REPLS": {
@@ -39,7 +39,7 @@
39
39
  "SRC1_TYPE": "f32",
40
40
  "BLOCK_SIZE": 32
41
41
  },
42
- "DECLS": ["BYTE_HELPERS", "Q4_1"]
42
+ "DECLS": ["BYTE_HELPERS", "Q4_1_T", "Q4_1"]
43
43
  },
44
44
  {
45
45
  "REPLS": {
@@ -47,7 +47,7 @@
47
47
  "SRC1_TYPE": "f32",
48
48
  "BLOCK_SIZE": 32
49
49
  },
50
- "DECLS": ["BYTE_HELPERS", "Q5_0"]
50
+ "DECLS": ["BYTE_HELPERS", "Q5_0_T", "Q5_0"]
51
51
  },
52
52
  {
53
53
  "REPLS": {
@@ -55,7 +55,7 @@
55
55
  "SRC1_TYPE": "f32",
56
56
  "BLOCK_SIZE": 32
57
57
  },
58
- "DECLS": ["BYTE_HELPERS", "Q5_1"]
58
+ "DECLS": ["BYTE_HELPERS", "Q5_1_T", "Q5_1"]
59
59
  },
60
60
  {
61
61
  "REPLS": {
@@ -63,7 +63,7 @@
63
63
  "SRC1_TYPE": "f32",
64
64
  "BLOCK_SIZE": 32
65
65
  },
66
- "DECLS": ["BYTE_HELPERS", "Q8_0"]
66
+ "DECLS": ["BYTE_HELPERS", "Q8_0_T", "Q8_0"]
67
67
  },
68
68
  {
69
69
  "REPLS": {
@@ -71,7 +71,7 @@
71
71
  "SRC1_TYPE": "f32",
72
72
  "BLOCK_SIZE": 256
73
73
  },
74
- "DECLS": ["BYTE_HELPERS", "Q2_K"]
74
+ "DECLS": ["BYTE_HELPERS", "Q2_K_T", "Q2_K"]
75
75
  },
76
76
  {
77
77
  "REPLS": {
@@ -79,7 +79,7 @@
79
79
  "SRC1_TYPE": "f32",
80
80
  "BLOCK_SIZE": 256
81
81
  },
82
- "DECLS": ["BYTE_HELPERS", "Q3_K"]
82
+ "DECLS": ["BYTE_HELPERS", "Q3_K_T", "Q3_K"]
83
83
  },
84
84
  {
85
85
  "REPLS": {
@@ -87,7 +87,7 @@
87
87
  "SRC1_TYPE": "f32",
88
88
  "BLOCK_SIZE": 256
89
89
  },
90
- "DECLS": ["Q45_K_SCALE_MIN", "BYTE_HELPERS", "Q4_K"]
90
+ "DECLS": ["Q45_K_SCALE_MIN", "BYTE_HELPERS", "Q4_K_T", "Q4_K"]
91
91
  },
92
92
  {
93
93
  "REPLS": {
@@ -95,7 +95,7 @@
95
95
  "SRC1_TYPE": "f32",
96
96
  "BLOCK_SIZE": 256
97
97
  },
98
- "DECLS": ["Q45_K_SCALE_MIN", "BYTE_HELPERS", "Q5_K"]
98
+ "DECLS": ["Q45_K_SCALE_MIN", "BYTE_HELPERS", "Q5_K_T", "Q5_K"]
99
99
  },
100
100
  {
101
101
  "REPLS": {
@@ -103,7 +103,7 @@
103
103
  "SRC1_TYPE": "f32",
104
104
  "BLOCK_SIZE": 256
105
105
  },
106
- "DECLS": ["BYTE_HELPERS", "Q6_K"]
106
+ "DECLS": ["BYTE_HELPERS", "Q6_K_T", "Q6_K"]
107
107
  },
108
108
  {
109
109
  "REPLS": {
@@ -111,7 +111,7 @@
111
111
  "SRC1_TYPE": "f32",
112
112
  "BLOCK_SIZE": 256
113
113
  },
114
- "DECLS": ["BYTE_HELPERS", "IQ23_TABLES", "IQ2_XXS"]
114
+ "DECLS": ["BYTE_HELPERS", "IQ23_TABLES", "IQ2_XXS_GRID", "IQ2_XXS_T", "IQ2_XXS"]
115
115
  },
116
116
  {
117
117
  "REPLS": {
@@ -119,7 +119,7 @@
119
119
  "SRC1_TYPE": "f32",
120
120
  "BLOCK_SIZE": 256
121
121
  },
122
- "DECLS": ["BYTE_HELPERS", "IQ23_TABLES", "IQ2_XS"]
122
+ "DECLS": ["BYTE_HELPERS", "IQ23_TABLES", "IQ2_XS_GRID", "IQ2_XS_T", "IQ2_XS"]
123
123
  },
124
124
  {
125
125
  "REPLS": {
@@ -127,7 +127,7 @@
127
127
  "SRC1_TYPE": "f32",
128
128
  "BLOCK_SIZE": 256
129
129
  },
130
- "DECLS": ["BYTE_HELPERS", "IQ23_TABLES", "IQ2_S"]
130
+ "DECLS": ["BYTE_HELPERS", "IQ23_TABLES", "IQ2_S_GRID", "IQ2_S_T", "IQ2_S"]
131
131
  },
132
132
  {
133
133
  "REPLS": {
@@ -135,7 +135,7 @@
135
135
  "SRC1_TYPE": "f32",
136
136
  "BLOCK_SIZE": 256
137
137
  },
138
- "DECLS": ["BYTE_HELPERS", "IQ23_TABLES", "IQ3_XSS"]
138
+ "DECLS": ["BYTE_HELPERS", "IQ23_TABLES", "IQ3_XSS_GRID", "IQ3_XSS_T", "IQ3_XSS"]
139
139
  },
140
140
  {
141
141
  "REPLS": {
@@ -143,7 +143,7 @@
143
143
  "SRC1_TYPE": "f32",
144
144
  "BLOCK_SIZE": 256
145
145
  },
146
- "DECLS": ["BYTE_HELPERS", "IQ23_TABLES", "IQ3_S"]
146
+ "DECLS": ["BYTE_HELPERS", "IQ23_TABLES", "IQ3_S_GRID", "IQ3_S_T", "IQ3_S"]
147
147
  },
148
148
  {
149
149
  "REPLS": {
@@ -151,7 +151,7 @@
151
151
  "SRC1_TYPE": "f32",
152
152
  "BLOCK_SIZE": 256
153
153
  },
154
- "DECLS": ["BYTE_HELPERS", "IQ1_TABLE","IQ1_S"]
154
+ "DECLS": ["BYTE_HELPERS", "IQ1_GRID", "IQ1_S_T", "IQ1_S"]
155
155
  },
156
156
  {
157
157
  "REPLS": {
@@ -159,7 +159,7 @@
159
159
  "SRC1_TYPE": "f32",
160
160
  "BLOCK_SIZE": 256
161
161
  },
162
- "DECLS": ["BYTE_HELPERS", "IQ1_TABLE","IQ1_M"]
162
+ "DECLS": ["BYTE_HELPERS", "IQ1_GRID", "IQ1_M_T", "IQ1_M"]
163
163
  },
164
164
  {
165
165
  "REPLS": {
@@ -167,7 +167,7 @@
167
167
  "SRC1_TYPE": "f32",
168
168
  "BLOCK_SIZE": 32,
169
169
  },
170
- "DECLS": ["BYTE_HELPERS", "IQ4_TABLE", "IQ4_NL"]
170
+ "DECLS": ["BYTE_HELPERS", "IQ4_GRID", "IQ4_NL_T", "IQ4_NL"]
171
171
  },
172
172
  {
173
173
  "REPLS": {
@@ -175,7 +175,7 @@
175
175
  "SRC1_TYPE": "f32",
176
176
  "BLOCK_SIZE": 256,
177
177
  },
178
- "DECLS": ["BYTE_HELPERS", "IQ4_TABLE", "IQ4_XS"]
178
+ "DECLS": ["BYTE_HELPERS", "IQ4_GRID", "IQ4_XS_T", "IQ4_XS"]
179
179
  }
180
180
  ]
181
181
 
@@ -183,18 +183,6 @@
183
183
 
184
184
  #define(DECLS)
185
185
 
186
- #decl(BYTE_HELPERS)
187
-
188
- fn get_byte(value: u32, index: u32) -> u32 {
189
- return (value >> (index * 8)) & 0xFF;
190
- }
191
-
192
- fn get_byte_i32(value: u32, index: u32) -> i32 {
193
- return bitcast<i32>(((value >> (index * 8)) & 0xFF) << 24) >> 24;
194
- }
195
-
196
- #enddecl(BYTE_HELPERS)
197
-
198
186
  #decl(FLOAT)
199
187
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
200
188
  return f32(src0[src0_idx_base + offset]) * f32(src1[src1_idx_base + offset]);
@@ -202,11 +190,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
202
190
  #enddecl(FLOAT)
203
191
 
204
192
  #decl(Q4_0)
205
- struct q4_0 {
206
- d: f16,
207
- qs: array<f16, 8>
208
- };
209
-
210
193
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
211
194
  let block_q4_0 = src0[src0_idx_base + offset];
212
195
  let d = f32(block_q4_0.d);
@@ -227,12 +210,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
227
210
  #enddecl(Q4_0)
228
211
 
229
212
  #decl(Q4_1)
230
- struct q4_1 {
231
- d: f16,
232
- m: f16,
233
- qs: array<u32, 4>
234
- };
235
-
236
213
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
237
214
  let block_q4_1 = src0[src0_idx_base + offset];
238
215
  let d = f32(block_q4_1.d);
@@ -254,12 +231,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
254
231
  #enddecl(Q4_1)
255
232
 
256
233
  #decl(Q5_0)
257
- struct q5_0 {
258
- d: f16,
259
- qh: array<f16, 2>,
260
- qs: array<f16, 8>
261
- };
262
-
263
234
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
264
235
  let block_q5_0 = src0[src0_idx_base + offset];
265
236
  let d = f32(block_q5_0.d);
@@ -283,13 +254,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
283
254
  #enddecl(Q5_0)
284
255
 
285
256
  #decl(Q5_1)
286
- struct q5_1 {
287
- d: f16,
288
- m: f16,
289
- qh: u32,
290
- qs: array<u32, 4>
291
- };
292
-
293
257
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
294
258
  let block_q5_1 = src0[src0_idx_base + offset];
295
259
  let d = f32(block_q5_1.d);
@@ -313,11 +277,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
313
277
  #enddecl(Q5_1)
314
278
 
315
279
  #decl(Q8_0)
316
- struct q8_0 {
317
- d: f16,
318
- qs: array<f16, 16>
319
- };
320
-
321
280
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
322
281
  let block_q8_0 = src0[src0_idx_base + offset];
323
282
  let d = f32(block_q8_0.d);
@@ -336,12 +295,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
336
295
  #enddecl(Q8_0)
337
296
 
338
297
  #decl(Q8_1)
339
- struct q8_1 {
340
- d: f16,
341
- m: f16,
342
- qs: array<u32, 8>
343
- };
344
-
345
298
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
346
299
  let block_q8_1 = src0[src0_idx_base + offset];
347
300
  let d = f32(block_q8_1.d);
@@ -362,13 +315,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
362
315
 
363
316
  #decl(Q2_K)
364
317
  // 16 blocks of 16 elements each
365
- struct q2_k {
366
- scales: array<u32, 4>,
367
- qs: array<u32, 16>,
368
- d: f16,
369
- dmin: f16
370
- };
371
-
372
318
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
373
319
  let block = src0[src0_idx_base + offset];
374
320
  let d = f32(block.d);
@@ -403,13 +349,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
403
349
 
404
350
  #decl(Q3_K)
405
351
  // 16 blocks of 16 elements each
406
- struct q3_k {
407
- hmask: array<f16, 16>,
408
- qs: array<f16, 32>,
409
- scales: array<f16, 6>, // 6-bit quantized values
410
- d: f16
411
- };
412
-
413
352
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
414
353
  let block = src0[src0_idx_base + offset];
415
354
  let d = f32(block.d);
@@ -470,34 +409,8 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
470
409
 
471
410
  #enddecl(Q3_K)
472
411
 
473
- #decl(Q45_K_SCALE_MIN)
474
-
475
- fn get_scale_min(is: u32, scales: array<u32, 3>) -> vec2<f32> {
476
- if (is < 4) {
477
- let sc_byte = get_byte(scales[is / 4], is % 4);
478
- let min_byte = get_byte(scales[(is + 4) / 4], is % 4);
479
- return vec2(f32(sc_byte & 63), f32(min_byte & 63));
480
- } else {
481
- let sc_min_lo = get_byte(scales[(is + 4) / 4], (is + 4) % 4);
482
- let sc_hi = get_byte(scales[(is - 4) / 4], (is - 4) % 4);
483
- let min_hi = get_byte(scales[is / 4], is % 4);
484
- let sc = (sc_min_lo & 0xF) | ((sc_hi >> 6) << 4);
485
- let m = (sc_min_lo >> 4) | ((min_hi >> 6) << 4);
486
- return vec2(f32(sc), f32(m));
487
- }
488
- }
489
-
490
- #enddecl(Q45_K_SCALE_MIN)
491
-
492
412
  #decl(Q4_K)
493
413
  // 8 blocks of 32 elements each
494
- struct q4_k {
495
- d: f16,
496
- dmin: f16,
497
- scales: array<u32, 3>,
498
- qs: array<u32, 32>
499
- };
500
-
501
414
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
502
415
  let block = src0[src0_idx_base + offset];
503
416
  let d = f32(block.d);
@@ -528,14 +441,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
528
441
 
529
442
  #decl(Q5_K)
530
443
  // 8 blocks of 32 elements each
531
- struct q5_k {
532
- d: f16,
533
- dmin: f16,
534
- scales: array<u32, 3>,
535
- qh: array<u32, 8>,
536
- qs: array<u32, 32>
537
- };
538
-
539
444
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
540
445
  let block = src0[src0_idx_base + offset];
541
446
  let d = f32(block.d);
@@ -570,13 +475,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
570
475
 
571
476
  #decl(Q6_K)
572
477
  // 16 blocks of 16 elements each
573
- struct q6_k {
574
- ql: array<f16, 64>,
575
- qh: array<f16, 32>,
576
- scales: array<f16, 8>,
577
- d: f16
578
- };
579
-
580
478
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
581
479
  let block = src0[src0_idx_base + offset];
582
480
  let d = f32(block.d);
@@ -634,98 +532,7 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
634
532
 
635
533
  #enddecl(Q6_K)
636
534
 
637
- #decl(IQ23_TABLES)
638
- const kmask_iq2xs : array<u32, 2> = array<u32, 2>(
639
- 0x08040201u, // 1, 2, 4, 8
640
- 0x80402010u // 16, 32, 64, 128
641
- );
642
-
643
- const ksigns_iq2xs: array<u32, 32> = array<u32, 32>(
644
- 0x03828100,0x87060584,0x8b0a0988,0x0f8e8d0c,
645
- 0x93121190,0x17969514,0x1b9a9918,0x9f1e1d9c,
646
- 0xa32221a0,0x27a6a524,0x2baaa928,0xaf2e2dac,
647
- 0x33b2b130,0xb73635b4,0xbb3a39b8,0x3fbebd3c,
648
- 0xc34241c0,0x47c6c544,0x4bcac948,0xcf4e4dcc,
649
- 0x53d2d150,0xd75655d4,0xdb5a59d8,0x5fdedd5c,
650
- 0x63e2e160,0xe76665e4,0xeb6a69e8,0x6feeed6c,
651
- 0xf37271f0,0x77f6f574,0x7bfaf978,0xff7e7dfc
652
- );
653
- #enddecl(IQ23_TABLES)
654
-
655
535
  #decl(IQ2_XXS)
656
-
657
- const iq2xxs_grid = array<u32, 512>(
658
- 0x08080808, 0x08080808, 0x0808082b, 0x08080808, 0x08081919, 0x08080808, 0x08082b08, 0x08080808,
659
- 0x08082b2b, 0x08080808, 0x08190819, 0x08080808, 0x08191908, 0x08080808, 0x082b0808, 0x08080808,
660
- 0x082b082b, 0x08080808, 0x082b2b08, 0x08080808, 0x082b2b2b, 0x08080808, 0x19080819, 0x08080808,
661
- 0x19081908, 0x08080808, 0x19190808, 0x08080808, 0x19192b08, 0x08080808, 0x192b0819, 0x08080808,
662
- 0x192b1908, 0x08080808, 0x2b080808, 0x08080808, 0x2b08082b, 0x08080808, 0x2b082b2b, 0x08080808,
663
- 0x2b2b082b, 0x08080808, 0x08080819, 0x08080819, 0x08081908, 0x08080819, 0x08190808, 0x08080819,
664
- 0x08191919, 0x08080819, 0x19080808, 0x08080819, 0x2b081908, 0x08080819, 0x2b192b08, 0x08080819,
665
- 0x08080808, 0x0808082b, 0x0808082b, 0x0808082b, 0x082b082b, 0x0808082b, 0x2b08082b, 0x0808082b,
666
- 0x08080819, 0x08081908, 0x08081908, 0x08081908, 0x08190808, 0x08081908, 0x082b0819, 0x08081908,
667
- 0x082b1908, 0x08081908, 0x19080808, 0x08081908, 0x1908082b, 0x08081908, 0x19082b08, 0x08081908,
668
- 0x192b0808, 0x08081908, 0x2b080819, 0x08081908, 0x2b081908, 0x08081908, 0x2b190808, 0x08081908,
669
- 0x2b2b1908, 0x08081908, 0x08080808, 0x08081919, 0x0808082b, 0x08081919, 0x08082b08, 0x08081919,
670
- 0x082b0808, 0x08081919, 0x1908192b, 0x08081919, 0x192b2b19, 0x08081919, 0x2b080808, 0x08081919,
671
- 0x2b190819, 0x08081919, 0x08082b19, 0x0808192b, 0x08190808, 0x0808192b, 0x19080808, 0x0808192b,
672
- 0x2b081908, 0x0808192b, 0x2b2b1908, 0x0808192b, 0x08080808, 0x08082b08, 0x08081919, 0x08082b08,
673
- 0x08082b08, 0x08082b08, 0x08191908, 0x08082b08, 0x082b2b08, 0x08082b08, 0x19080819, 0x08082b08,
674
- 0x19081908, 0x08082b08, 0x19190808, 0x08082b08, 0x1919082b, 0x08082b08, 0x2b082b08, 0x08082b08,
675
- 0x08081908, 0x08082b19, 0x19080808, 0x08082b19, 0x0808082b, 0x08082b2b, 0x08191908, 0x08082b2b,
676
- 0x08080819, 0x08190808, 0x08081908, 0x08190808, 0x08190808, 0x08190808, 0x082b0819, 0x08190808,
677
- 0x19080808, 0x08190808, 0x192b0808, 0x08190808, 0x2b081908, 0x08190808, 0x2b190808, 0x08190808,
678
- 0x2b191919, 0x08190808, 0x08080808, 0x08190819, 0x08082b08, 0x08190819, 0x082b0808, 0x08190819,
679
- 0x19190808, 0x08190819, 0x19192b2b, 0x08190819, 0x2b080808, 0x08190819, 0x082b1908, 0x0819082b,
680
- 0x19081919, 0x0819082b, 0x08080808, 0x08191908, 0x08082b08, 0x08191908, 0x082b0808, 0x08191908,
681
- 0x082b1919, 0x08191908, 0x19082b19, 0x08191908, 0x2b080808, 0x08191908, 0x08192b08, 0x08191919,
682
- 0x192b082b, 0x08191919, 0x08080808, 0x0819192b, 0x0819192b, 0x0819192b, 0x08080819, 0x08192b08,
683
- 0x08081908, 0x08192b08, 0x08190808, 0x08192b08, 0x19080808, 0x08192b08, 0x2b080819, 0x08192b08,
684
- 0x08080808, 0x08192b19, 0x08081919, 0x08192b19, 0x2b2b0808, 0x08192b19, 0x19190819, 0x08192b2b,
685
- 0x08080808, 0x082b0808, 0x0808082b, 0x082b0808, 0x08082b2b, 0x082b0808, 0x19081908, 0x082b0808,
686
- 0x192b0819, 0x082b0808, 0x2b080808, 0x082b0808, 0x2b08082b, 0x082b0808, 0x082b2b19, 0x082b0819,
687
- 0x19082b08, 0x082b0819, 0x08080808, 0x082b082b, 0x0808082b, 0x082b082b, 0x08080819, 0x082b1908,
688
- 0x08081908, 0x082b1908, 0x08190808, 0x082b1908, 0x19080808, 0x082b1908, 0x1919192b, 0x082b1908,
689
- 0x08080808, 0x082b1919, 0x19080819, 0x082b1919, 0x192b1908, 0x082b1919, 0x2b190808, 0x082b192b,
690
- 0x08082b08, 0x082b2b08, 0x082b0808, 0x082b2b08, 0x2b191908, 0x082b2b08, 0x19081908, 0x082b2b2b,
691
- 0x08080819, 0x19080808, 0x08081908, 0x19080808, 0x08190808, 0x19080808, 0x08192b08, 0x19080808,
692
- 0x082b0819, 0x19080808, 0x082b1908, 0x19080808, 0x19080808, 0x19080808, 0x19082b08, 0x19080808,
693
- 0x1919192b, 0x19080808, 0x192b0808, 0x19080808, 0x2b080819, 0x19080808, 0x2b081908, 0x19080808,
694
- 0x2b190808, 0x19080808, 0x08080808, 0x19080819, 0x082b0808, 0x19080819, 0x192b0819, 0x19080819,
695
- 0x2b080808, 0x19080819, 0x2b081919, 0x19080819, 0x08080819, 0x1908082b, 0x08190808, 0x1908082b,
696
- 0x19082b08, 0x1908082b, 0x1919192b, 0x1908082b, 0x192b2b08, 0x1908082b, 0x08080808, 0x19081908,
697
- 0x08082b08, 0x19081908, 0x082b0808, 0x19081908, 0x2b080808, 0x19081908, 0x2b192b19, 0x19081908,
698
- 0x0819082b, 0x19081919, 0x082b1908, 0x19081919, 0x08080808, 0x1908192b, 0x08080819, 0x19082b08,
699
- 0x08081908, 0x19082b08, 0x08190808, 0x19082b08, 0x19080808, 0x19082b08, 0x19081919, 0x19082b08,
700
- 0x08080808, 0x19082b19, 0x19192b08, 0x19082b19, 0x192b0819, 0x19082b19, 0x2b08082b, 0x19082b19,
701
- 0x19081919, 0x19082b2b, 0x2b190808, 0x19082b2b, 0x08080808, 0x19190808, 0x08082b08, 0x19190808,
702
- 0x08190819, 0x19190808, 0x08192b19, 0x19190808, 0x082b0808, 0x19190808, 0x2b080808, 0x19190808,
703
- 0x2b082b08, 0x19190808, 0x08081908, 0x19190819, 0x1908082b, 0x19190819, 0x2b2b1908, 0x19190819,
704
- 0x2b190819, 0x1919082b, 0x2b190808, 0x19191908, 0x2b19082b, 0x19191908, 0x08082b2b, 0x19191919,
705
- 0x08080819, 0x1919192b, 0x19191908, 0x1919192b, 0x08080808, 0x19192b08, 0x08190819, 0x19192b08,
706
- 0x08192b19, 0x19192b08, 0x192b1908, 0x19192b08, 0x19080808, 0x19192b19, 0x08082b08, 0x19192b2b,
707
- 0x08081908, 0x192b0808, 0x08190808, 0x192b0808, 0x19080808, 0x192b0808, 0x192b2b08, 0x192b0808,
708
- 0x08080808, 0x192b0819, 0x19191919, 0x192b0819, 0x08192b08, 0x192b082b, 0x192b0808, 0x192b082b,
709
- 0x08080808, 0x192b1908, 0x08081919, 0x192b1908, 0x08190808, 0x192b1919, 0x0819082b, 0x192b1919,
710
- 0x2b081908, 0x192b1919, 0x1908082b, 0x192b2b08, 0x08080808, 0x2b080808, 0x0808082b, 0x2b080808,
711
- 0x08082b2b, 0x2b080808, 0x19080819, 0x2b080808, 0x2b08082b, 0x2b080808, 0x08081908, 0x2b080819,
712
- 0x08192b08, 0x2b080819, 0x19080808, 0x2b080819, 0x08190819, 0x2b08082b, 0x08080819, 0x2b081908,
713
- 0x08081908, 0x2b081908, 0x08190808, 0x2b081908, 0x08191919, 0x2b081908, 0x19080808, 0x2b081908,
714
- 0x192b0808, 0x2b081908, 0x08080808, 0x2b081919, 0x1908192b, 0x2b081919, 0x2b191908, 0x2b081919,
715
- 0x08082b19, 0x2b08192b, 0x19080808, 0x2b08192b, 0x192b0808, 0x2b08192b, 0x0808082b, 0x2b082b08,
716
- 0x08081908, 0x2b082b19, 0x08190819, 0x2b082b2b, 0x08081908, 0x2b190808, 0x08190808, 0x2b190808,
717
- 0x082b1908, 0x2b190808, 0x19080808, 0x2b190808, 0x2b2b0819, 0x2b190808, 0x0819192b, 0x2b190819,
718
- 0x2b080808, 0x2b190819, 0x19081919, 0x2b19082b, 0x08080808, 0x2b191908, 0x082b082b, 0x2b191908,
719
- 0x19081908, 0x2b191908, 0x19190819, 0x2b191919, 0x2b080819, 0x2b192b08, 0x082b0808, 0x2b192b19,
720
- 0x0808082b, 0x2b2b0808, 0x19190808, 0x2b2b0808, 0x2b081919, 0x2b2b0808, 0x08082b19, 0x2b2b0819,
721
- 0x08080808, 0x2b2b082b, 0x08192b08, 0x2b2b1908, 0x19190808, 0x2b2b2b08, 0x08081908, 0x2b2b2b19
722
- );
723
-
724
- struct iq2_xxs {
725
- d: f16,
726
- qs: array<f16, 32>
727
- };
728
-
729
536
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
730
537
  let block = src0[src0_idx_base + offset];
731
538
  let d = f32(block.d);
@@ -753,143 +560,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
753
560
  #enddecl(IQ2_XXS)
754
561
 
755
562
  #decl(IQ2_XS)
756
- const iq2xs_grid = array<u32, 1024>(
757
- 0x08080808, 0x08080808, 0x0808082b, 0x08080808, 0x08081919, 0x08080808, 0x08082b08, 0x08080808,
758
- 0x08082b2b, 0x08080808, 0x08190819, 0x08080808, 0x08191908, 0x08080808, 0x0819192b, 0x08080808,
759
- 0x08192b19, 0x08080808, 0x082b0808, 0x08080808, 0x082b082b, 0x08080808, 0x082b1919, 0x08080808,
760
- 0x082b2b08, 0x08080808, 0x19080819, 0x08080808, 0x19081908, 0x08080808, 0x1908192b, 0x08080808,
761
- 0x19082b19, 0x08080808, 0x19190808, 0x08080808, 0x1919082b, 0x08080808, 0x19191919, 0x08080808,
762
- 0x19192b08, 0x08080808, 0x192b0819, 0x08080808, 0x192b1908, 0x08080808, 0x2b080808, 0x08080808,
763
- 0x2b08082b, 0x08080808, 0x2b081919, 0x08080808, 0x2b082b08, 0x08080808, 0x2b190819, 0x08080808,
764
- 0x2b191908, 0x08080808, 0x2b192b19, 0x08080808, 0x2b2b0808, 0x08080808, 0x08080819, 0x08080819,
765
- 0x08081908, 0x08080819, 0x0808192b, 0x08080819, 0x08082b19, 0x08080819, 0x08190808, 0x08080819,
766
- 0x0819082b, 0x08080819, 0x08191919, 0x08080819, 0x08192b08, 0x08080819, 0x08192b2b, 0x08080819,
767
- 0x082b0819, 0x08080819, 0x082b1908, 0x08080819, 0x19080808, 0x08080819, 0x1908082b, 0x08080819,
768
- 0x19081919, 0x08080819, 0x19082b08, 0x08080819, 0x19190819, 0x08080819, 0x19191908, 0x08080819,
769
- 0x192b0808, 0x08080819, 0x192b2b08, 0x08080819, 0x2b080819, 0x08080819, 0x2b081908, 0x08080819,
770
- 0x2b190808, 0x08080819, 0x08080808, 0x0808082b, 0x0808082b, 0x0808082b, 0x08081919, 0x0808082b,
771
- 0x08082b08, 0x0808082b, 0x08190819, 0x0808082b, 0x08191908, 0x0808082b, 0x082b0808, 0x0808082b,
772
- 0x19080819, 0x0808082b, 0x19081908, 0x0808082b, 0x19190808, 0x0808082b, 0x19191919, 0x0808082b,
773
- 0x2b080808, 0x0808082b, 0x2b082b2b, 0x0808082b, 0x08080819, 0x08081908, 0x08081908, 0x08081908,
774
- 0x0808192b, 0x08081908, 0x08082b19, 0x08081908, 0x08190808, 0x08081908, 0x0819082b, 0x08081908,
775
- 0x08191919, 0x08081908, 0x08192b08, 0x08081908, 0x082b0819, 0x08081908, 0x082b1908, 0x08081908,
776
- 0x19080808, 0x08081908, 0x1908082b, 0x08081908, 0x19081919, 0x08081908, 0x19082b08, 0x08081908,
777
- 0x19190819, 0x08081908, 0x19191908, 0x08081908, 0x1919192b, 0x08081908, 0x192b0808, 0x08081908,
778
- 0x2b080819, 0x08081908, 0x2b081908, 0x08081908, 0x2b190808, 0x08081908, 0x08080808, 0x08081919,
779
- 0x0808082b, 0x08081919, 0x08081919, 0x08081919, 0x08082b08, 0x08081919, 0x08190819, 0x08081919,
780
- 0x08191908, 0x08081919, 0x082b0808, 0x08081919, 0x19080819, 0x08081919, 0x19081908, 0x08081919,
781
- 0x19190808, 0x08081919, 0x192b0819, 0x08081919, 0x2b080808, 0x08081919, 0x08080819, 0x0808192b,
782
- 0x08081908, 0x0808192b, 0x08190808, 0x0808192b, 0x082b192b, 0x0808192b, 0x19080808, 0x0808192b,
783
- 0x1908082b, 0x0808192b, 0x2b081908, 0x0808192b, 0x08080808, 0x08082b08, 0x0808082b, 0x08082b08,
784
- 0x08081919, 0x08082b08, 0x08082b08, 0x08082b08, 0x08082b2b, 0x08082b08, 0x08190819, 0x08082b08,
785
- 0x08191908, 0x08082b08, 0x082b0808, 0x08082b08, 0x082b1919, 0x08082b08, 0x19080819, 0x08082b08,
786
- 0x19081908, 0x08082b08, 0x19190808, 0x08082b08, 0x19192b08, 0x08082b08, 0x2b080808, 0x08082b08,
787
- 0x2b2b0808, 0x08082b08, 0x2b2b2b2b, 0x08082b08, 0x08080819, 0x08082b19, 0x08081908, 0x08082b19,
788
- 0x08190808, 0x08082b19, 0x19080808, 0x08082b19, 0x2b080819, 0x08082b19, 0x2b082b19, 0x08082b19,
789
- 0x08080808, 0x08082b2b, 0x082b0808, 0x08082b2b, 0x082b2b08, 0x08082b2b, 0x2b19192b, 0x08082b2b,
790
- 0x2b2b0808, 0x08082b2b, 0x08080819, 0x08190808, 0x08081908, 0x08190808, 0x0808192b, 0x08190808,
791
- 0x08082b19, 0x08190808, 0x08190808, 0x08190808, 0x0819082b, 0x08190808, 0x08191919, 0x08190808,
792
- 0x08192b08, 0x08190808, 0x082b0819, 0x08190808, 0x082b1908, 0x08190808, 0x19080808, 0x08190808,
793
- 0x1908082b, 0x08190808, 0x19081919, 0x08190808, 0x19082b08, 0x08190808, 0x19190819, 0x08190808,
794
- 0x19191908, 0x08190808, 0x192b0808, 0x08190808, 0x192b2b2b, 0x08190808, 0x2b080819, 0x08190808,
795
- 0x2b081908, 0x08190808, 0x2b190808, 0x08190808, 0x08080808, 0x08190819, 0x0808082b, 0x08190819,
796
- 0x08081919, 0x08190819, 0x08082b08, 0x08190819, 0x08190819, 0x08190819, 0x08191908, 0x08190819,
797
- 0x082b0808, 0x08190819, 0x19080819, 0x08190819, 0x19081908, 0x08190819, 0x19190808, 0x08190819,
798
- 0x2b080808, 0x08190819, 0x2b191908, 0x08190819, 0x2b19192b, 0x08190819, 0x08080819, 0x0819082b,
799
- 0x08081908, 0x0819082b, 0x0808192b, 0x0819082b, 0x08190808, 0x0819082b, 0x19080808, 0x0819082b,
800
- 0x192b0808, 0x0819082b, 0x08080808, 0x08191908, 0x0808082b, 0x08191908, 0x08081919, 0x08191908,
801
- 0x08082b08, 0x08191908, 0x08190819, 0x08191908, 0x08191908, 0x08191908, 0x082b0808, 0x08191908,
802
- 0x19080819, 0x08191908, 0x19081908, 0x08191908, 0x19082b19, 0x08191908, 0x19190808, 0x08191908,
803
- 0x192b1908, 0x08191908, 0x2b080808, 0x08191908, 0x08080819, 0x08191919, 0x08081908, 0x08191919,
804
- 0x08190808, 0x08191919, 0x19080808, 0x08191919, 0x08080808, 0x0819192b, 0x08191908, 0x0819192b,
805
- 0x19082b19, 0x0819192b, 0x08080819, 0x08192b08, 0x08081908, 0x08192b08, 0x08190808, 0x08192b08,
806
- 0x0819082b, 0x08192b08, 0x19080808, 0x08192b08, 0x19191908, 0x08192b08, 0x2b08192b, 0x08192b08,
807
- 0x08080808, 0x08192b19, 0x08081919, 0x08192b19, 0x192b192b, 0x08192b19, 0x19190819, 0x08192b2b,
808
- 0x2b2b2b19, 0x08192b2b, 0x08080808, 0x082b0808, 0x0808082b, 0x082b0808, 0x08081919, 0x082b0808,
809
- 0x08082b08, 0x082b0808, 0x08082b2b, 0x082b0808, 0x08190819, 0x082b0808, 0x08191908, 0x082b0808,
810
- 0x082b0808, 0x082b0808, 0x19080819, 0x082b0808, 0x19081908, 0x082b0808, 0x19190808, 0x082b0808,
811
- 0x2b080808, 0x082b0808, 0x2b2b0808, 0x082b0808, 0x08080819, 0x082b0819, 0x08081908, 0x082b0819,
812
- 0x08190808, 0x082b0819, 0x19080808, 0x082b0819, 0x19082b08, 0x082b0819, 0x192b1919, 0x082b0819,
813
- 0x08080808, 0x082b082b, 0x082b082b, 0x082b082b, 0x2b080808, 0x082b082b, 0x2b2b2b08, 0x082b082b,
814
- 0x08080819, 0x082b1908, 0x08081908, 0x082b1908, 0x08190808, 0x082b1908, 0x082b2b19, 0x082b1908,
815
- 0x19080808, 0x082b1908, 0x08080808, 0x082b1919, 0x19080819, 0x082b1919, 0x1919082b, 0x082b1919,
816
- 0x2b192b19, 0x082b1919, 0x08080819, 0x082b192b, 0x08192b2b, 0x082b192b, 0x2b2b192b, 0x082b192b,
817
- 0x08080808, 0x082b2b08, 0x08082b08, 0x082b2b08, 0x08082b2b, 0x082b2b08, 0x082b0808, 0x082b2b08,
818
- 0x19191919, 0x082b2b08, 0x2b082b08, 0x082b2b08, 0x2b2b082b, 0x082b2b08, 0x192b2b08, 0x082b2b19,
819
- 0x2b190808, 0x082b2b19, 0x08082b08, 0x082b2b2b, 0x082b0808, 0x082b2b2b, 0x2b08082b, 0x082b2b2b,
820
- 0x2b082b08, 0x082b2b2b, 0x2b082b2b, 0x082b2b2b, 0x08080819, 0x19080808, 0x08081908, 0x19080808,
821
- 0x0808192b, 0x19080808, 0x08082b19, 0x19080808, 0x08190808, 0x19080808, 0x0819082b, 0x19080808,
822
- 0x08191919, 0x19080808, 0x08192b08, 0x19080808, 0x082b0819, 0x19080808, 0x082b1908, 0x19080808,
823
- 0x19080808, 0x19080808, 0x1908082b, 0x19080808, 0x19081919, 0x19080808, 0x19082b08, 0x19080808,
824
- 0x19082b2b, 0x19080808, 0x19190819, 0x19080808, 0x19191908, 0x19080808, 0x192b0808, 0x19080808,
825
- 0x192b1919, 0x19080808, 0x2b080819, 0x19080808, 0x2b081908, 0x19080808, 0x2b190808, 0x19080808,
826
- 0x08080808, 0x19080819, 0x0808082b, 0x19080819, 0x08081919, 0x19080819, 0x08082b08, 0x19080819,
827
- 0x08190819, 0x19080819, 0x08191908, 0x19080819, 0x082b0808, 0x19080819, 0x19080819, 0x19080819,
828
- 0x19081908, 0x19080819, 0x19190808, 0x19080819, 0x2b080808, 0x19080819, 0x2b081919, 0x19080819,
829
- 0x2b2b082b, 0x19080819, 0x08080819, 0x1908082b, 0x08081908, 0x1908082b, 0x08190808, 0x1908082b,
830
- 0x0819082b, 0x1908082b, 0x082b2b19, 0x1908082b, 0x19080808, 0x1908082b, 0x08080808, 0x19081908,
831
- 0x0808082b, 0x19081908, 0x08081919, 0x19081908, 0x08082b08, 0x19081908, 0x08190819, 0x19081908,
832
- 0x08191908, 0x19081908, 0x08192b19, 0x19081908, 0x082b0808, 0x19081908, 0x19080819, 0x19081908,
833
- 0x19081908, 0x19081908, 0x19190808, 0x19081908, 0x2b080808, 0x19081908, 0x2b191908, 0x19081908,
834
- 0x08080819, 0x19081919, 0x08081908, 0x19081919, 0x08190808, 0x19081919, 0x082b1908, 0x19081919,
835
- 0x19080808, 0x19081919, 0x2b192b2b, 0x19081919, 0x08080808, 0x1908192b, 0x08082b2b, 0x1908192b,
836
- 0x19081908, 0x1908192b, 0x19190808, 0x1908192b, 0x08080819, 0x19082b08, 0x08081908, 0x19082b08,
837
- 0x08190808, 0x19082b08, 0x19080808, 0x19082b08, 0x19081919, 0x19082b08, 0x19191908, 0x19082b08,
838
- 0x192b082b, 0x19082b08, 0x08080808, 0x19082b19, 0x08190819, 0x19082b19, 0x19081908, 0x19082b19,
839
- 0x19190808, 0x19082b19, 0x192b2b19, 0x19082b19, 0x08081908, 0x19082b2b, 0x08080808, 0x19190808,
840
- 0x0808082b, 0x19190808, 0x08081919, 0x19190808, 0x08082b08, 0x19190808, 0x08190819, 0x19190808,
841
- 0x08191908, 0x19190808, 0x082b0808, 0x19190808, 0x082b2b08, 0x19190808, 0x19080819, 0x19190808,
842
- 0x19081908, 0x19190808, 0x19190808, 0x19190808, 0x2b080808, 0x19190808, 0x08080819, 0x19190819,
843
- 0x08081908, 0x19190819, 0x08190808, 0x19190819, 0x08191919, 0x19190819, 0x19080808, 0x19190819,
844
- 0x1908082b, 0x19190819, 0x08080808, 0x1919082b, 0x19081908, 0x1919082b, 0x2b2b2b2b, 0x1919082b,
845
- 0x08080819, 0x19191908, 0x08081908, 0x19191908, 0x08190808, 0x19191908, 0x082b0819, 0x19191908,
846
- 0x19080808, 0x19191908, 0x192b0808, 0x19191908, 0x2b080819, 0x19191908, 0x2b2b0819, 0x19191908,
847
- 0x08080808, 0x19191919, 0x08082b08, 0x19191919, 0x2b080808, 0x19191919, 0x2b082b08, 0x19191919,
848
- 0x082b0819, 0x1919192b, 0x192b2b08, 0x1919192b, 0x2b2b0819, 0x1919192b, 0x08080808, 0x19192b08,
849
- 0x08191908, 0x19192b08, 0x19080819, 0x19192b08, 0x19190808, 0x19192b08, 0x2b192b19, 0x19192b08,
850
- 0x08192b2b, 0x19192b19, 0x19080808, 0x19192b19, 0x1908082b, 0x19192b19, 0x2b081919, 0x19192b2b,
851
- 0x08080819, 0x192b0808, 0x08081908, 0x192b0808, 0x08190808, 0x192b0808, 0x19080808, 0x192b0808,
852
- 0x19191908, 0x192b0808, 0x192b082b, 0x192b0808, 0x2b08192b, 0x192b0808, 0x2b2b2b19, 0x192b0808,
853
- 0x08080808, 0x192b0819, 0x082b1908, 0x192b082b, 0x19082b2b, 0x192b082b, 0x2b19082b, 0x192b082b,
854
- 0x08080808, 0x192b1908, 0x0819192b, 0x192b1908, 0x08190808, 0x192b1919, 0x19080808, 0x192b1919,
855
- 0x19081919, 0x192b1919, 0x2b2b1908, 0x192b1919, 0x08080819, 0x192b2b08, 0x192b2b2b, 0x192b2b08,
856
- 0x082b1919, 0x192b2b19, 0x0808192b, 0x192b2b2b, 0x19191908, 0x192b2b2b, 0x192b082b, 0x192b2b2b,
857
- 0x08080808, 0x2b080808, 0x0808082b, 0x2b080808, 0x08081919, 0x2b080808, 0x08082b08, 0x2b080808,
858
- 0x08190819, 0x2b080808, 0x08191908, 0x2b080808, 0x082b0808, 0x2b080808, 0x082b2b2b, 0x2b080808,
859
- 0x19080819, 0x2b080808, 0x19081908, 0x2b080808, 0x19190808, 0x2b080808, 0x2b080808, 0x2b080808,
860
- 0x2b08082b, 0x2b080808, 0x2b2b2b08, 0x2b080808, 0x2b2b2b2b, 0x2b080808, 0x08080819, 0x2b080819,
861
- 0x08081908, 0x2b080819, 0x0808192b, 0x2b080819, 0x08190808, 0x2b080819, 0x19080808, 0x2b080819,
862
- 0x19190819, 0x2b080819, 0x19192b19, 0x2b080819, 0x08080808, 0x2b08082b, 0x082b0808, 0x2b08082b,
863
- 0x2b080808, 0x2b08082b, 0x2b08082b, 0x2b08082b, 0x2b2b0808, 0x2b08082b, 0x2b2b2b08, 0x2b08082b,
864
- 0x08080819, 0x2b081908, 0x08081908, 0x2b081908, 0x08190808, 0x2b081908, 0x0819082b, 0x2b081908,
865
- 0x08191919, 0x2b081908, 0x19080808, 0x2b081908, 0x192b0808, 0x2b081908, 0x2b082b19, 0x2b081908,
866
- 0x08080808, 0x2b081919, 0x19081908, 0x2b081919, 0x2b2b1919, 0x2b081919, 0x08192b08, 0x2b08192b,
867
- 0x192b2b2b, 0x2b08192b, 0x08080808, 0x2b082b08, 0x08082b08, 0x2b082b08, 0x082b1919, 0x2b082b08,
868
- 0x19192b2b, 0x2b082b08, 0x2b080808, 0x2b082b08, 0x2b08082b, 0x2b082b08, 0x2b2b2b08, 0x2b082b08,
869
- 0x0808192b, 0x2b082b19, 0x082b082b, 0x2b082b2b, 0x2b080808, 0x2b082b2b, 0x2b082b08, 0x2b082b2b,
870
- 0x2b19192b, 0x2b082b2b, 0x2b2b2b08, 0x2b082b2b, 0x08080819, 0x2b190808, 0x08081908, 0x2b190808,
871
- 0x08190808, 0x2b190808, 0x19080808, 0x2b190808, 0x1919192b, 0x2b190808, 0x2b081908, 0x2b190808,
872
- 0x08080808, 0x2b190819, 0x082b082b, 0x2b190819, 0x192b1908, 0x2b190819, 0x1919192b, 0x2b19082b,
873
- 0x2b082b19, 0x2b19082b, 0x08080808, 0x2b191908, 0x08081919, 0x2b191908, 0x19081908, 0x2b191908,
874
- 0x19190808, 0x2b191908, 0x19192b08, 0x2b191908, 0x082b2b19, 0x2b191919, 0x2b190808, 0x2b191919,
875
- 0x2b19082b, 0x2b191919, 0x19080819, 0x2b19192b, 0x19190819, 0x2b192b08, 0x2b2b192b, 0x2b192b08,
876
- 0x19082b19, 0x2b192b19, 0x08191919, 0x2b192b2b, 0x192b0808, 0x2b192b2b, 0x08080808, 0x2b2b0808,
877
- 0x0808082b, 0x2b2b0808, 0x08082b08, 0x2b2b0808, 0x08082b2b, 0x2b2b0808, 0x082b0808, 0x2b2b0808,
878
- 0x082b2b2b, 0x2b2b0808, 0x2b2b0808, 0x2b2b0808, 0x19190819, 0x2b2b0819, 0x19192b19, 0x2b2b0819,
879
- 0x2b2b192b, 0x2b2b0819, 0x08080808, 0x2b2b082b, 0x0808082b, 0x2b2b082b, 0x08082b08, 0x2b2b082b,
880
- 0x082b2b2b, 0x2b2b082b, 0x2b080808, 0x2b2b082b, 0x2b2b0808, 0x2b2b082b, 0x19080808, 0x2b2b1908,
881
- 0x2b191919, 0x2b2b1908, 0x192b1919, 0x2b2b192b, 0x2b192b08, 0x2b2b192b, 0x08082b2b, 0x2b2b2b08,
882
- 0x082b0808, 0x2b2b2b08, 0x082b082b, 0x2b2b2b08, 0x082b2b08, 0x2b2b2b08, 0x2b2b0808, 0x2b2b2b08,
883
- 0x2b2b2b08, 0x2b2b2b08, 0x08081908, 0x2b2b2b19, 0x2b081908, 0x2b2b2b19, 0x2b08192b, 0x2b2b2b19,
884
- 0x082b2b08, 0x2b2b2b2b, 0x082b2b2b, 0x2b2b2b2b, 0x2b190819, 0x2b2b2b2b, 0x2b2b2b2b, 0x2b2b2b2b
885
- );
886
-
887
- struct iq2_xs {
888
- d: f16,
889
- qs: array<f16, 32>,
890
- scales: array<f16, 4>
891
- };
892
-
893
563
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
894
564
  let block = src0[src0_idx_base + offset];
895
565
  let d = f32(block.d);
@@ -925,273 +595,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
925
595
  #enddecl(IQ2_XS)
926
596
 
927
597
  #decl(IQ2_S)
928
-
929
- const iq2s_grid = array<u32, 2048>(
930
- 0x08080808, 0x08080808, 0x0808082b, 0x08080808, 0x08081919, 0x08080808, 0x08082b08, 0x08080808,
931
- 0x08082b2b, 0x08080808, 0x08190819, 0x08080808, 0x08191908, 0x08080808, 0x0819192b, 0x08080808,
932
- 0x08192b19, 0x08080808, 0x082b0808, 0x08080808, 0x082b082b, 0x08080808, 0x082b1919, 0x08080808,
933
- 0x082b2b08, 0x08080808, 0x19080819, 0x08080808, 0x19081908, 0x08080808, 0x1908192b, 0x08080808,
934
- 0x19082b19, 0x08080808, 0x19190808, 0x08080808, 0x1919082b, 0x08080808, 0x19191919, 0x08080808,
935
- 0x19192b08, 0x08080808, 0x192b0819, 0x08080808, 0x192b1908, 0x08080808, 0x192b192b, 0x08080808,
936
- 0x192b2b19, 0x08080808, 0x2b080808, 0x08080808, 0x2b08082b, 0x08080808, 0x2b081919, 0x08080808,
937
- 0x2b082b08, 0x08080808, 0x2b190819, 0x08080808, 0x2b191908, 0x08080808, 0x2b2b0808, 0x08080808,
938
- 0x2b2b1919, 0x08080808, 0x2b2b2b2b, 0x08080808, 0x08080819, 0x08080819, 0x08081908, 0x08080819,
939
- 0x0808192b, 0x08080819, 0x08082b19, 0x08080819, 0x08190808, 0x08080819, 0x0819082b, 0x08080819,
940
- 0x08191919, 0x08080819, 0x08192b08, 0x08080819, 0x082b0819, 0x08080819, 0x082b1908, 0x08080819,
941
- 0x19080808, 0x08080819, 0x1908082b, 0x08080819, 0x19081919, 0x08080819, 0x19082b08, 0x08080819,
942
- 0x19190819, 0x08080819, 0x19191908, 0x08080819, 0x1919192b, 0x08080819, 0x19192b19, 0x08080819,
943
- 0x192b0808, 0x08080819, 0x192b1919, 0x08080819, 0x192b2b08, 0x08080819, 0x2b080819, 0x08080819,
944
- 0x2b081908, 0x08080819, 0x2b190808, 0x08080819, 0x2b19082b, 0x08080819, 0x2b191919, 0x08080819,
945
- 0x2b2b0819, 0x08080819, 0x2b2b1908, 0x08080819, 0x08080808, 0x0808082b, 0x0808082b, 0x0808082b,
946
- 0x08081919, 0x0808082b, 0x08082b08, 0x0808082b, 0x08190819, 0x0808082b, 0x08191908, 0x0808082b,
947
- 0x082b0808, 0x0808082b, 0x082b2b2b, 0x0808082b, 0x19080819, 0x0808082b, 0x19081908, 0x0808082b,
948
- 0x1908192b, 0x0808082b, 0x19082b19, 0x0808082b, 0x19190808, 0x0808082b, 0x19191919, 0x0808082b,
949
- 0x2b080808, 0x0808082b, 0x2b081919, 0x0808082b, 0x2b082b2b, 0x0808082b, 0x2b191908, 0x0808082b,
950
- 0x2b2b082b, 0x0808082b, 0x08080819, 0x08081908, 0x08081908, 0x08081908, 0x0808192b, 0x08081908,
951
- 0x08082b19, 0x08081908, 0x08190808, 0x08081908, 0x0819082b, 0x08081908, 0x08191919, 0x08081908,
952
- 0x08192b08, 0x08081908, 0x082b0819, 0x08081908, 0x082b1908, 0x08081908, 0x082b192b, 0x08081908,
953
- 0x082b2b19, 0x08081908, 0x19080808, 0x08081908, 0x1908082b, 0x08081908, 0x19081919, 0x08081908,
954
- 0x19082b08, 0x08081908, 0x19082b2b, 0x08081908, 0x19190819, 0x08081908, 0x19191908, 0x08081908,
955
- 0x1919192b, 0x08081908, 0x19192b19, 0x08081908, 0x192b0808, 0x08081908, 0x192b082b, 0x08081908,
956
- 0x192b1919, 0x08081908, 0x2b080819, 0x08081908, 0x2b081908, 0x08081908, 0x2b08192b, 0x08081908,
957
- 0x2b082b19, 0x08081908, 0x2b190808, 0x08081908, 0x2b191919, 0x08081908, 0x2b192b08, 0x08081908,
958
- 0x2b2b0819, 0x08081908, 0x2b2b1908, 0x08081908, 0x08080808, 0x08081919, 0x0808082b, 0x08081919,
959
- 0x08081919, 0x08081919, 0x08082b08, 0x08081919, 0x08082b2b, 0x08081919, 0x08190819, 0x08081919,
960
- 0x08191908, 0x08081919, 0x0819192b, 0x08081919, 0x08192b19, 0x08081919, 0x082b0808, 0x08081919,
961
- 0x082b1919, 0x08081919, 0x082b2b08, 0x08081919, 0x19080819, 0x08081919, 0x19081908, 0x08081919,
962
- 0x1908192b, 0x08081919, 0x19082b19, 0x08081919, 0x19190808, 0x08081919, 0x1919082b, 0x08081919,
963
- 0x19191919, 0x08081919, 0x19192b08, 0x08081919, 0x192b0819, 0x08081919, 0x192b1908, 0x08081919,
964
- 0x2b080808, 0x08081919, 0x2b08082b, 0x08081919, 0x2b081919, 0x08081919, 0x2b082b08, 0x08081919,
965
- 0x2b190819, 0x08081919, 0x2b191908, 0x08081919, 0x2b2b0808, 0x08081919, 0x08080819, 0x0808192b,
966
- 0x08081908, 0x0808192b, 0x0808192b, 0x0808192b, 0x08082b19, 0x0808192b, 0x08190808, 0x0808192b,
967
- 0x08191919, 0x0808192b, 0x19080808, 0x0808192b, 0x19081919, 0x0808192b, 0x19082b08, 0x0808192b,
968
- 0x19190819, 0x0808192b, 0x19191908, 0x0808192b, 0x192b0808, 0x0808192b, 0x2b080819, 0x0808192b,
969
- 0x2b081908, 0x0808192b, 0x2b190808, 0x0808192b, 0x08080808, 0x08082b08, 0x0808082b, 0x08082b08,
970
- 0x08081919, 0x08082b08, 0x08082b08, 0x08082b08, 0x08190819, 0x08082b08, 0x08191908, 0x08082b08,
971
- 0x0819192b, 0x08082b08, 0x08192b19, 0x08082b08, 0x082b0808, 0x08082b08, 0x082b1919, 0x08082b08,
972
- 0x082b2b2b, 0x08082b08, 0x19080819, 0x08082b08, 0x19081908, 0x08082b08, 0x1908192b, 0x08082b08,
973
- 0x19082b19, 0x08082b08, 0x19190808, 0x08082b08, 0x1919082b, 0x08082b08, 0x19191919, 0x08082b08,
974
- 0x19192b08, 0x08082b08, 0x192b0819, 0x08082b08, 0x192b1908, 0x08082b08, 0x2b080808, 0x08082b08,
975
- 0x2b081919, 0x08082b08, 0x2b191908, 0x08082b08, 0x2b2b2b2b, 0x08082b08, 0x08080819, 0x08082b19,
976
- 0x08081908, 0x08082b19, 0x08190808, 0x08082b19, 0x0819082b, 0x08082b19, 0x08191919, 0x08082b19,
977
- 0x08192b08, 0x08082b19, 0x082b0819, 0x08082b19, 0x19080808, 0x08082b19, 0x19081919, 0x08082b19,
978
- 0x19082b08, 0x08082b19, 0x19190819, 0x08082b19, 0x19191908, 0x08082b19, 0x192b0808, 0x08082b19,
979
- 0x2b080819, 0x08082b19, 0x2b190808, 0x08082b19, 0x08080808, 0x08082b2b, 0x08190819, 0x08082b2b,
980
- 0x08191908, 0x08082b2b, 0x082b082b, 0x08082b2b, 0x082b2b08, 0x08082b2b, 0x082b2b2b, 0x08082b2b,
981
- 0x19190808, 0x08082b2b, 0x2b192b19, 0x08082b2b, 0x08080819, 0x08190808, 0x08081908, 0x08190808,
982
- 0x0808192b, 0x08190808, 0x08082b19, 0x08190808, 0x08190808, 0x08190808, 0x0819082b, 0x08190808,
983
- 0x08191919, 0x08190808, 0x08192b08, 0x08190808, 0x082b0819, 0x08190808, 0x082b1908, 0x08190808,
984
- 0x082b192b, 0x08190808, 0x19080808, 0x08190808, 0x1908082b, 0x08190808, 0x19081919, 0x08190808,
985
- 0x19082b08, 0x08190808, 0x19190819, 0x08190808, 0x19191908, 0x08190808, 0x1919192b, 0x08190808,
986
- 0x19192b19, 0x08190808, 0x192b0808, 0x08190808, 0x192b082b, 0x08190808, 0x192b1919, 0x08190808,
987
- 0x192b2b08, 0x08190808, 0x2b080819, 0x08190808, 0x2b081908, 0x08190808, 0x2b08192b, 0x08190808,
988
- 0x2b190808, 0x08190808, 0x2b191919, 0x08190808, 0x2b192b08, 0x08190808, 0x2b2b0819, 0x08190808,
989
- 0x2b2b1908, 0x08190808, 0x08080808, 0x08190819, 0x0808082b, 0x08190819, 0x08081919, 0x08190819,
990
- 0x08082b08, 0x08190819, 0x08082b2b, 0x08190819, 0x08190819, 0x08190819, 0x08191908, 0x08190819,
991
- 0x0819192b, 0x08190819, 0x08192b19, 0x08190819, 0x082b0808, 0x08190819, 0x082b082b, 0x08190819,
992
- 0x082b1919, 0x08190819, 0x082b2b08, 0x08190819, 0x19080819, 0x08190819, 0x19081908, 0x08190819,
993
- 0x1908192b, 0x08190819, 0x19082b19, 0x08190819, 0x19190808, 0x08190819, 0x1919082b, 0x08190819,
994
- 0x19191919, 0x08190819, 0x19192b08, 0x08190819, 0x192b0819, 0x08190819, 0x192b1908, 0x08190819,
995
- 0x2b080808, 0x08190819, 0x2b08082b, 0x08190819, 0x2b081919, 0x08190819, 0x2b082b08, 0x08190819,
996
- 0x2b190819, 0x08190819, 0x2b191908, 0x08190819, 0x08080819, 0x0819082b, 0x08081908, 0x0819082b,
997
- 0x08082b19, 0x0819082b, 0x08190808, 0x0819082b, 0x08191919, 0x0819082b, 0x082b0819, 0x0819082b,
998
- 0x082b1908, 0x0819082b, 0x19080808, 0x0819082b, 0x19081919, 0x0819082b, 0x19190819, 0x0819082b,
999
- 0x19191908, 0x0819082b, 0x2b080819, 0x0819082b, 0x2b081908, 0x0819082b, 0x2b190808, 0x0819082b,
1000
- 0x08080808, 0x08191908, 0x0808082b, 0x08191908, 0x08081919, 0x08191908, 0x08082b08, 0x08191908,
1001
- 0x08190819, 0x08191908, 0x08191908, 0x08191908, 0x0819192b, 0x08191908, 0x08192b19, 0x08191908,
1002
- 0x082b0808, 0x08191908, 0x082b1919, 0x08191908, 0x082b2b08, 0x08191908, 0x19080819, 0x08191908,
1003
- 0x19081908, 0x08191908, 0x1908192b, 0x08191908, 0x19082b19, 0x08191908, 0x19190808, 0x08191908,
1004
- 0x1919082b, 0x08191908, 0x19191919, 0x08191908, 0x19192b08, 0x08191908, 0x192b0819, 0x08191908,
1005
- 0x192b1908, 0x08191908, 0x2b080808, 0x08191908, 0x2b08082b, 0x08191908, 0x2b081919, 0x08191908,
1006
- 0x2b082b08, 0x08191908, 0x2b190819, 0x08191908, 0x2b191908, 0x08191908, 0x2b2b0808, 0x08191908,
1007
- 0x08080819, 0x08191919, 0x08081908, 0x08191919, 0x0808192b, 0x08191919, 0x08082b19, 0x08191919,
1008
- 0x08190808, 0x08191919, 0x0819082b, 0x08191919, 0x08191919, 0x08191919, 0x08192b08, 0x08191919,
1009
- 0x082b0819, 0x08191919, 0x082b1908, 0x08191919, 0x19080808, 0x08191919, 0x1908082b, 0x08191919,
1010
- 0x19081919, 0x08191919, 0x19082b08, 0x08191919, 0x19190819, 0x08191919, 0x19191908, 0x08191919,
1011
- 0x192b0808, 0x08191919, 0x2b080819, 0x08191919, 0x2b081908, 0x08191919, 0x2b190808, 0x08191919,
1012
- 0x08080808, 0x0819192b, 0x08081919, 0x0819192b, 0x08082b08, 0x0819192b, 0x08190819, 0x0819192b,
1013
- 0x08191908, 0x0819192b, 0x082b0808, 0x0819192b, 0x19080819, 0x0819192b, 0x19081908, 0x0819192b,
1014
- 0x19190808, 0x0819192b, 0x2b080808, 0x0819192b, 0x2b2b2b2b, 0x0819192b, 0x08080819, 0x08192b08,
1015
- 0x08081908, 0x08192b08, 0x0808192b, 0x08192b08, 0x08082b19, 0x08192b08, 0x08190808, 0x08192b08,
1016
- 0x08191919, 0x08192b08, 0x08192b08, 0x08192b08, 0x082b0819, 0x08192b08, 0x19080808, 0x08192b08,
1017
- 0x1908082b, 0x08192b08, 0x19081919, 0x08192b08, 0x19082b08, 0x08192b08, 0x19190819, 0x08192b08,
1018
- 0x19191908, 0x08192b08, 0x192b0808, 0x08192b08, 0x2b080819, 0x08192b08, 0x2b081908, 0x08192b08,
1019
- 0x08080808, 0x08192b19, 0x0808082b, 0x08192b19, 0x08081919, 0x08192b19, 0x08082b08, 0x08192b19,
1020
- 0x08190819, 0x08192b19, 0x08191908, 0x08192b19, 0x082b0808, 0x08192b19, 0x19080819, 0x08192b19,
1021
- 0x19081908, 0x08192b19, 0x19190808, 0x08192b19, 0x192b2b19, 0x08192b19, 0x2b2b082b, 0x08192b19,
1022
- 0x08081908, 0x08192b2b, 0x08190808, 0x08192b2b, 0x19080808, 0x08192b2b, 0x1919192b, 0x08192b2b,
1023
- 0x08080808, 0x082b0808, 0x0808082b, 0x082b0808, 0x08081919, 0x082b0808, 0x08082b08, 0x082b0808,
1024
- 0x08190819, 0x082b0808, 0x08191908, 0x082b0808, 0x0819192b, 0x082b0808, 0x08192b19, 0x082b0808,
1025
- 0x082b0808, 0x082b0808, 0x082b1919, 0x082b0808, 0x082b2b2b, 0x082b0808, 0x19080819, 0x082b0808,
1026
- 0x19081908, 0x082b0808, 0x19190808, 0x082b0808, 0x1919082b, 0x082b0808, 0x19191919, 0x082b0808,
1027
- 0x192b1908, 0x082b0808, 0x2b080808, 0x082b0808, 0x2b082b2b, 0x082b0808, 0x2b191908, 0x082b0808,
1028
- 0x2b2b2b2b, 0x082b0808, 0x08080819, 0x082b0819, 0x08081908, 0x082b0819, 0x08190808, 0x082b0819,
1029
- 0x0819082b, 0x082b0819, 0x08191919, 0x082b0819, 0x082b0819, 0x082b0819, 0x19080808, 0x082b0819,
1030
- 0x1908082b, 0x082b0819, 0x19081919, 0x082b0819, 0x19190819, 0x082b0819, 0x19191908, 0x082b0819,
1031
- 0x192b0808, 0x082b0819, 0x2b080819, 0x082b0819, 0x2b081908, 0x082b0819, 0x2b190808, 0x082b0819,
1032
- 0x08080808, 0x082b082b, 0x08082b2b, 0x082b082b, 0x082b082b, 0x082b082b, 0x082b2b08, 0x082b082b,
1033
- 0x082b2b2b, 0x082b082b, 0x19081908, 0x082b082b, 0x19190808, 0x082b082b, 0x2b082b08, 0x082b082b,
1034
- 0x2b082b2b, 0x082b082b, 0x2b2b2b08, 0x082b082b, 0x08080819, 0x082b1908, 0x08081908, 0x082b1908,
1035
- 0x0808192b, 0x082b1908, 0x08082b19, 0x082b1908, 0x08190808, 0x082b1908, 0x08191919, 0x082b1908,
1036
- 0x08192b08, 0x082b1908, 0x082b0819, 0x082b1908, 0x082b1908, 0x082b1908, 0x19080808, 0x082b1908,
1037
- 0x1908082b, 0x082b1908, 0x19081919, 0x082b1908, 0x19082b08, 0x082b1908, 0x19190819, 0x082b1908,
1038
- 0x19191908, 0x082b1908, 0x192b0808, 0x082b1908, 0x2b080819, 0x082b1908, 0x2b081908, 0x082b1908,
1039
- 0x2b190808, 0x082b1908, 0x08080808, 0x082b1919, 0x08081919, 0x082b1919, 0x08082b08, 0x082b1919,
1040
- 0x08190819, 0x082b1919, 0x08191908, 0x082b1919, 0x082b0808, 0x082b1919, 0x19080819, 0x082b1919,
1041
- 0x19081908, 0x082b1919, 0x19190808, 0x082b1919, 0x192b192b, 0x082b1919, 0x2b080808, 0x082b1919,
1042
- 0x08080819, 0x082b192b, 0x08081908, 0x082b192b, 0x08190808, 0x082b192b, 0x19080808, 0x082b192b,
1043
- 0x19192b19, 0x082b192b, 0x08080808, 0x082b2b08, 0x08081919, 0x082b2b08, 0x08190819, 0x082b2b08,
1044
- 0x08191908, 0x082b2b08, 0x19080819, 0x082b2b08, 0x19081908, 0x082b2b08, 0x19190808, 0x082b2b08,
1045
- 0x2b082b2b, 0x082b2b08, 0x2b2b2b2b, 0x082b2b08, 0x08080819, 0x082b2b19, 0x08081908, 0x082b2b19,
1046
- 0x08190808, 0x082b2b19, 0x2b191919, 0x082b2b19, 0x08082b2b, 0x082b2b2b, 0x082b082b, 0x082b2b2b,
1047
- 0x192b1908, 0x082b2b2b, 0x2b082b08, 0x082b2b2b, 0x2b082b2b, 0x082b2b2b, 0x08080819, 0x19080808,
1048
- 0x08081908, 0x19080808, 0x0808192b, 0x19080808, 0x08082b19, 0x19080808, 0x08190808, 0x19080808,
1049
- 0x0819082b, 0x19080808, 0x08191919, 0x19080808, 0x08192b08, 0x19080808, 0x08192b2b, 0x19080808,
1050
- 0x082b0819, 0x19080808, 0x082b1908, 0x19080808, 0x082b192b, 0x19080808, 0x19080808, 0x19080808,
1051
- 0x1908082b, 0x19080808, 0x19081919, 0x19080808, 0x19082b08, 0x19080808, 0x19082b2b, 0x19080808,
1052
- 0x19190819, 0x19080808, 0x19191908, 0x19080808, 0x1919192b, 0x19080808, 0x19192b19, 0x19080808,
1053
- 0x192b0808, 0x19080808, 0x192b082b, 0x19080808, 0x192b1919, 0x19080808, 0x2b080819, 0x19080808,
1054
- 0x2b081908, 0x19080808, 0x2b190808, 0x19080808, 0x2b191919, 0x19080808, 0x2b192b08, 0x19080808,
1055
- 0x2b2b0819, 0x19080808, 0x2b2b1908, 0x19080808, 0x08080808, 0x19080819, 0x0808082b, 0x19080819,
1056
- 0x08081919, 0x19080819, 0x08082b08, 0x19080819, 0x08190819, 0x19080819, 0x08191908, 0x19080819,
1057
- 0x0819192b, 0x19080819, 0x08192b19, 0x19080819, 0x082b0808, 0x19080819, 0x082b082b, 0x19080819,
1058
- 0x082b1919, 0x19080819, 0x19080819, 0x19080819, 0x19081908, 0x19080819, 0x1908192b, 0x19080819,
1059
- 0x19082b19, 0x19080819, 0x19190808, 0x19080819, 0x1919082b, 0x19080819, 0x19191919, 0x19080819,
1060
- 0x19192b08, 0x19080819, 0x192b0819, 0x19080819, 0x192b1908, 0x19080819, 0x2b080808, 0x19080819,
1061
- 0x2b08082b, 0x19080819, 0x2b081919, 0x19080819, 0x2b082b08, 0x19080819, 0x2b190819, 0x19080819,
1062
- 0x2b191908, 0x19080819, 0x2b2b0808, 0x19080819, 0x08080819, 0x1908082b, 0x08081908, 0x1908082b,
1063
- 0x08190808, 0x1908082b, 0x0819082b, 0x1908082b, 0x08191919, 0x1908082b, 0x08192b08, 0x1908082b,
1064
- 0x082b1908, 0x1908082b, 0x19080808, 0x1908082b, 0x19081919, 0x1908082b, 0x19082b08, 0x1908082b,
1065
- 0x19190819, 0x1908082b, 0x19191908, 0x1908082b, 0x192b0808, 0x1908082b, 0x2b080819, 0x1908082b,
1066
- 0x2b081908, 0x1908082b, 0x08080808, 0x19081908, 0x0808082b, 0x19081908, 0x08081919, 0x19081908,
1067
- 0x08082b08, 0x19081908, 0x08082b2b, 0x19081908, 0x08190819, 0x19081908, 0x08191908, 0x19081908,
1068
- 0x0819192b, 0x19081908, 0x08192b19, 0x19081908, 0x082b0808, 0x19081908, 0x082b082b, 0x19081908,
1069
- 0x082b1919, 0x19081908, 0x082b2b08, 0x19081908, 0x19080819, 0x19081908, 0x19081908, 0x19081908,
1070
- 0x1908192b, 0x19081908, 0x19082b19, 0x19081908, 0x19190808, 0x19081908, 0x1919082b, 0x19081908,
1071
- 0x19191919, 0x19081908, 0x19192b08, 0x19081908, 0x192b0819, 0x19081908, 0x192b1908, 0x19081908,
1072
- 0x2b080808, 0x19081908, 0x2b08082b, 0x19081908, 0x2b081919, 0x19081908, 0x2b082b08, 0x19081908,
1073
- 0x2b190819, 0x19081908, 0x2b191908, 0x19081908, 0x2b2b0808, 0x19081908, 0x08080819, 0x19081919,
1074
- 0x08081908, 0x19081919, 0x0808192b, 0x19081919, 0x08082b19, 0x19081919, 0x08190808, 0x19081919,
1075
- 0x0819082b, 0x19081919, 0x08191919, 0x19081919, 0x08192b08, 0x19081919, 0x082b0819, 0x19081919,
1076
- 0x082b1908, 0x19081919, 0x19080808, 0x19081919, 0x1908082b, 0x19081919, 0x19081919, 0x19081919,
1077
- 0x19082b08, 0x19081919, 0x19190819, 0x19081919, 0x19191908, 0x19081919, 0x192b0808, 0x19081919,
1078
- 0x192b2b2b, 0x19081919, 0x2b080819, 0x19081919, 0x2b081908, 0x19081919, 0x2b190808, 0x19081919,
1079
- 0x08080808, 0x1908192b, 0x0808082b, 0x1908192b, 0x08081919, 0x1908192b, 0x08082b08, 0x1908192b,
1080
- 0x08190819, 0x1908192b, 0x08191908, 0x1908192b, 0x082b0808, 0x1908192b, 0x19080819, 0x1908192b,
1081
- 0x19081908, 0x1908192b, 0x19190808, 0x1908192b, 0x2b080808, 0x1908192b, 0x2b2b1919, 0x1908192b,
1082
- 0x08080819, 0x19082b08, 0x08081908, 0x19082b08, 0x08082b19, 0x19082b08, 0x08190808, 0x19082b08,
1083
- 0x0819082b, 0x19082b08, 0x08191919, 0x19082b08, 0x08192b08, 0x19082b08, 0x082b0819, 0x19082b08,
1084
- 0x082b1908, 0x19082b08, 0x19080808, 0x19082b08, 0x1908082b, 0x19082b08, 0x19081919, 0x19082b08,
1085
- 0x19082b08, 0x19082b08, 0x19190819, 0x19082b08, 0x19191908, 0x19082b08, 0x192b0808, 0x19082b08,
1086
- 0x2b081908, 0x19082b08, 0x2b190808, 0x19082b08, 0x08080808, 0x19082b19, 0x0808082b, 0x19082b19,
1087
- 0x08081919, 0x19082b19, 0x08082b08, 0x19082b19, 0x08190819, 0x19082b19, 0x08191908, 0x19082b19,
1088
- 0x082b0808, 0x19082b19, 0x19080819, 0x19082b19, 0x19081908, 0x19082b19, 0x19190808, 0x19082b19,
1089
- 0x2b080808, 0x19082b19, 0x2b19192b, 0x19082b19, 0x08080819, 0x19082b2b, 0x08081908, 0x19082b2b,
1090
- 0x08190808, 0x19082b2b, 0x19080808, 0x19082b2b, 0x08080808, 0x19190808, 0x0808082b, 0x19190808,
1091
- 0x08081919, 0x19190808, 0x08082b08, 0x19190808, 0x08190819, 0x19190808, 0x08191908, 0x19190808,
1092
- 0x0819192b, 0x19190808, 0x08192b19, 0x19190808, 0x082b0808, 0x19190808, 0x082b082b, 0x19190808,
1093
- 0x082b1919, 0x19190808, 0x082b2b08, 0x19190808, 0x19080819, 0x19190808, 0x19081908, 0x19190808,
1094
- 0x1908192b, 0x19190808, 0x19082b19, 0x19190808, 0x19190808, 0x19190808, 0x1919082b, 0x19190808,
1095
- 0x19191919, 0x19190808, 0x19192b08, 0x19190808, 0x192b0819, 0x19190808, 0x192b1908, 0x19190808,
1096
- 0x2b080808, 0x19190808, 0x2b08082b, 0x19190808, 0x2b081919, 0x19190808, 0x2b082b08, 0x19190808,
1097
- 0x2b190819, 0x19190808, 0x2b191908, 0x19190808, 0x08080819, 0x19190819, 0x08081908, 0x19190819,
1098
- 0x0808192b, 0x19190819, 0x08082b19, 0x19190819, 0x08190808, 0x19190819, 0x0819082b, 0x19190819,
1099
- 0x08191919, 0x19190819, 0x08192b08, 0x19190819, 0x082b0819, 0x19190819, 0x082b1908, 0x19190819,
1100
- 0x19080808, 0x19190819, 0x1908082b, 0x19190819, 0x19081919, 0x19190819, 0x19082b08, 0x19190819,
1101
- 0x19190819, 0x19190819, 0x19191908, 0x19190819, 0x192b0808, 0x19190819, 0x2b080819, 0x19190819,
1102
- 0x2b081908, 0x19190819, 0x2b190808, 0x19190819, 0x08080808, 0x1919082b, 0x08081919, 0x1919082b,
1103
- 0x08082b08, 0x1919082b, 0x08190819, 0x1919082b, 0x08191908, 0x1919082b, 0x082b0808, 0x1919082b,
1104
- 0x19080819, 0x1919082b, 0x19081908, 0x1919082b, 0x19190808, 0x1919082b, 0x192b2b19, 0x1919082b,
1105
- 0x2b080808, 0x1919082b, 0x08080819, 0x19191908, 0x08081908, 0x19191908, 0x0808192b, 0x19191908,
1106
- 0x08082b19, 0x19191908, 0x08190808, 0x19191908, 0x0819082b, 0x19191908, 0x08191919, 0x19191908,
1107
- 0x08192b08, 0x19191908, 0x082b0819, 0x19191908, 0x082b1908, 0x19191908, 0x19080808, 0x19191908,
1108
- 0x1908082b, 0x19191908, 0x19081919, 0x19191908, 0x19082b08, 0x19191908, 0x19190819, 0x19191908,
1109
- 0x19191908, 0x19191908, 0x192b0808, 0x19191908, 0x2b080819, 0x19191908, 0x2b081908, 0x19191908,
1110
- 0x2b190808, 0x19191908, 0x08080808, 0x19191919, 0x0808082b, 0x19191919, 0x08081919, 0x19191919,
1111
- 0x08082b08, 0x19191919, 0x08190819, 0x19191919, 0x08191908, 0x19191919, 0x082b0808, 0x19191919,
1112
- 0x19080819, 0x19191919, 0x19081908, 0x19191919, 0x19190808, 0x19191919, 0x2b080808, 0x19191919,
1113
- 0x08080819, 0x1919192b, 0x08081908, 0x1919192b, 0x08190808, 0x1919192b, 0x082b192b, 0x1919192b,
1114
- 0x19080808, 0x1919192b, 0x08080808, 0x19192b08, 0x0808082b, 0x19192b08, 0x08081919, 0x19192b08,
1115
- 0x08082b08, 0x19192b08, 0x08190819, 0x19192b08, 0x08191908, 0x19192b08, 0x082b0808, 0x19192b08,
1116
- 0x19080819, 0x19192b08, 0x19081908, 0x19192b08, 0x19190808, 0x19192b08, 0x19192b2b, 0x19192b08,
1117
- 0x2b080808, 0x19192b08, 0x08080819, 0x19192b19, 0x08081908, 0x19192b19, 0x08190808, 0x19192b19,
1118
- 0x19080808, 0x19192b19, 0x08080808, 0x19192b2b, 0x08192b19, 0x19192b2b, 0x2b081919, 0x19192b2b,
1119
- 0x2b2b2b08, 0x19192b2b, 0x08080819, 0x192b0808, 0x08081908, 0x192b0808, 0x0808192b, 0x192b0808,
1120
- 0x08190808, 0x192b0808, 0x0819082b, 0x192b0808, 0x08191919, 0x192b0808, 0x08192b08, 0x192b0808,
1121
- 0x082b0819, 0x192b0808, 0x082b1908, 0x192b0808, 0x19080808, 0x192b0808, 0x19081919, 0x192b0808,
1122
- 0x19082b08, 0x192b0808, 0x19190819, 0x192b0808, 0x19191908, 0x192b0808, 0x192b0808, 0x192b0808,
1123
- 0x2b081908, 0x192b0808, 0x2b190808, 0x192b0808, 0x08080808, 0x192b0819, 0x0808082b, 0x192b0819,
1124
- 0x08081919, 0x192b0819, 0x08082b08, 0x192b0819, 0x08190819, 0x192b0819, 0x08191908, 0x192b0819,
1125
- 0x082b0808, 0x192b0819, 0x19080819, 0x192b0819, 0x19081908, 0x192b0819, 0x19190808, 0x192b0819,
1126
- 0x2b080808, 0x192b0819, 0x2b192b19, 0x192b0819, 0x08081908, 0x192b082b, 0x08190808, 0x192b082b,
1127
- 0x19080808, 0x192b082b, 0x1919192b, 0x192b082b, 0x2b2b0819, 0x192b082b, 0x08080808, 0x192b1908,
1128
- 0x08081919, 0x192b1908, 0x08082b08, 0x192b1908, 0x08190819, 0x192b1908, 0x08191908, 0x192b1908,
1129
- 0x082b0808, 0x192b1908, 0x19080819, 0x192b1908, 0x19081908, 0x192b1908, 0x19190808, 0x192b1908,
1130
- 0x2b080808, 0x192b1908, 0x08080819, 0x192b1919, 0x08081908, 0x192b1919, 0x08190808, 0x192b1919,
1131
- 0x19080808, 0x192b1919, 0x19082b2b, 0x192b1919, 0x192b2b08, 0x192b1919, 0x2b19082b, 0x192b1919,
1132
- 0x08080808, 0x192b192b, 0x2b191908, 0x192b192b, 0x08080819, 0x192b2b08, 0x08081908, 0x192b2b08,
1133
- 0x08190808, 0x192b2b08, 0x192b1919, 0x192b2b08, 0x2b192b08, 0x192b2b08, 0x08080808, 0x192b2b19,
1134
- 0x082b2b2b, 0x192b2b19, 0x1908082b, 0x192b2b2b, 0x2b2b0819, 0x192b2b2b, 0x08080808, 0x2b080808,
1135
- 0x0808082b, 0x2b080808, 0x08081919, 0x2b080808, 0x08082b08, 0x2b080808, 0x08190819, 0x2b080808,
1136
- 0x08191908, 0x2b080808, 0x08192b19, 0x2b080808, 0x082b0808, 0x2b080808, 0x082b1919, 0x2b080808,
1137
- 0x19080819, 0x2b080808, 0x19081908, 0x2b080808, 0x19190808, 0x2b080808, 0x1919082b, 0x2b080808,
1138
- 0x19191919, 0x2b080808, 0x19192b08, 0x2b080808, 0x192b0819, 0x2b080808, 0x2b080808, 0x2b080808,
1139
- 0x2b081919, 0x2b080808, 0x2b190819, 0x2b080808, 0x2b191908, 0x2b080808, 0x08080819, 0x2b080819,
1140
- 0x08081908, 0x2b080819, 0x08082b19, 0x2b080819, 0x08190808, 0x2b080819, 0x0819082b, 0x2b080819,
1141
- 0x08191919, 0x2b080819, 0x08192b08, 0x2b080819, 0x082b0819, 0x2b080819, 0x082b1908, 0x2b080819,
1142
- 0x19080808, 0x2b080819, 0x1908082b, 0x2b080819, 0x19081919, 0x2b080819, 0x19082b08, 0x2b080819,
1143
- 0x19190819, 0x2b080819, 0x19191908, 0x2b080819, 0x2b080819, 0x2b080819, 0x2b081908, 0x2b080819,
1144
- 0x2b190808, 0x2b080819, 0x2b2b2b19, 0x2b080819, 0x08080808, 0x2b08082b, 0x08081919, 0x2b08082b,
1145
- 0x08082b2b, 0x2b08082b, 0x08190819, 0x2b08082b, 0x08191908, 0x2b08082b, 0x19080819, 0x2b08082b,
1146
- 0x19081908, 0x2b08082b, 0x19190808, 0x2b08082b, 0x08080819, 0x2b081908, 0x08081908, 0x2b081908,
1147
- 0x0808192b, 0x2b081908, 0x08082b19, 0x2b081908, 0x08190808, 0x2b081908, 0x0819082b, 0x2b081908,
1148
- 0x08191919, 0x2b081908, 0x08192b08, 0x2b081908, 0x082b0819, 0x2b081908, 0x19080808, 0x2b081908,
1149
- 0x1908082b, 0x2b081908, 0x19081919, 0x2b081908, 0x19082b08, 0x2b081908, 0x19190819, 0x2b081908,
1150
- 0x19191908, 0x2b081908, 0x192b0808, 0x2b081908, 0x2b080819, 0x2b081908, 0x2b081908, 0x2b081908,
1151
- 0x2b190808, 0x2b081908, 0x08080808, 0x2b081919, 0x0808082b, 0x2b081919, 0x08081919, 0x2b081919,
1152
- 0x08082b08, 0x2b081919, 0x08190819, 0x2b081919, 0x08191908, 0x2b081919, 0x082b0808, 0x2b081919,
1153
- 0x19080819, 0x2b081919, 0x19081908, 0x2b081919, 0x19190808, 0x2b081919, 0x2b080808, 0x2b081919,
1154
- 0x2b082b2b, 0x2b081919, 0x08080819, 0x2b08192b, 0x08081908, 0x2b08192b, 0x08190808, 0x2b08192b,
1155
- 0x082b2b19, 0x2b08192b, 0x19080808, 0x2b08192b, 0x08080808, 0x2b082b08, 0x08081919, 0x2b082b08,
1156
- 0x08190819, 0x2b082b08, 0x08191908, 0x2b082b08, 0x19080819, 0x2b082b08, 0x19081908, 0x2b082b08,
1157
- 0x19190808, 0x2b082b08, 0x2b2b082b, 0x2b082b08, 0x08080819, 0x2b082b19, 0x08081908, 0x2b082b19,
1158
- 0x19080808, 0x2b082b19, 0x192b1919, 0x2b082b19, 0x082b082b, 0x2b082b2b, 0x19192b08, 0x2b082b2b,
1159
- 0x19192b2b, 0x2b082b2b, 0x2b08082b, 0x2b082b2b, 0x2b2b082b, 0x2b082b2b, 0x08080819, 0x2b190808,
1160
- 0x08081908, 0x2b190808, 0x08082b19, 0x2b190808, 0x08190808, 0x2b190808, 0x0819082b, 0x2b190808,
1161
- 0x08191919, 0x2b190808, 0x08192b08, 0x2b190808, 0x082b1908, 0x2b190808, 0x19080808, 0x2b190808,
1162
- 0x1908082b, 0x2b190808, 0x19081919, 0x2b190808, 0x19082b08, 0x2b190808, 0x19190819, 0x2b190808,
1163
- 0x19191908, 0x2b190808, 0x192b0808, 0x2b190808, 0x2b080819, 0x2b190808, 0x2b081908, 0x2b190808,
1164
- 0x2b190808, 0x2b190808, 0x08080808, 0x2b190819, 0x08081919, 0x2b190819, 0x08190819, 0x2b190819,
1165
- 0x08191908, 0x2b190819, 0x19080819, 0x2b190819, 0x19081908, 0x2b190819, 0x19190808, 0x2b190819,
1166
- 0x19192b2b, 0x2b190819, 0x08080819, 0x2b19082b, 0x08081908, 0x2b19082b, 0x08190808, 0x2b19082b,
1167
- 0x19080808, 0x2b19082b, 0x2b2b192b, 0x2b19082b, 0x08080808, 0x2b191908, 0x0808082b, 0x2b191908,
1168
- 0x08081919, 0x2b191908, 0x08082b08, 0x2b191908, 0x08190819, 0x2b191908, 0x08191908, 0x2b191908,
1169
- 0x082b0808, 0x2b191908, 0x19080819, 0x2b191908, 0x19081908, 0x2b191908, 0x19190808, 0x2b191908,
1170
- 0x2b080808, 0x2b191908, 0x2b19192b, 0x2b191908, 0x08080819, 0x2b191919, 0x08081908, 0x2b191919,
1171
- 0x08190808, 0x2b191919, 0x19080808, 0x2b191919, 0x2b192b08, 0x2b191919, 0x2b2b0819, 0x2b191919,
1172
- 0x08080808, 0x2b19192b, 0x1908192b, 0x2b19192b, 0x192b1908, 0x2b19192b, 0x08080819, 0x2b192b08,
1173
- 0x08081908, 0x2b192b08, 0x08190808, 0x2b192b08, 0x082b192b, 0x2b192b08, 0x19080808, 0x2b192b08,
1174
- 0x2b2b2b19, 0x2b192b08, 0x08080808, 0x2b192b19, 0x19082b19, 0x2b192b19, 0x1919082b, 0x2b192b19,
1175
- 0x2b190808, 0x2b192b2b, 0x08080808, 0x2b2b0808, 0x08081919, 0x2b2b0808, 0x08082b2b, 0x2b2b0808,
1176
- 0x08191908, 0x2b2b0808, 0x082b082b, 0x2b2b0808, 0x082b2b2b, 0x2b2b0808, 0x19080819, 0x2b2b0808,
1177
- 0x19081908, 0x2b2b0808, 0x19190808, 0x2b2b0808, 0x2b2b082b, 0x2b2b0808, 0x2b2b2b2b, 0x2b2b0808,
1178
- 0x19080808, 0x2b2b0819, 0x192b1919, 0x2b2b0819, 0x0808082b, 0x2b2b082b, 0x08082b2b, 0x2b2b082b,
1179
- 0x082b082b, 0x2b2b082b, 0x082b2b08, 0x2b2b082b, 0x082b2b2b, 0x2b2b082b, 0x2b08082b, 0x2b2b082b,
1180
- 0x2b082b08, 0x2b2b082b, 0x2b082b2b, 0x2b2b082b, 0x2b2b2b08, 0x2b2b082b, 0x08080819, 0x2b2b1908,
1181
- 0x08081908, 0x2b2b1908, 0x08190808, 0x2b2b1908, 0x19080808, 0x2b2b1908, 0x2b082b19, 0x2b2b1908,
1182
- 0x2b2b1908, 0x2b2b1908, 0x08080808, 0x2b2b1919, 0x08192b19, 0x2b2b1919, 0x19190819, 0x2b2b192b,
1183
- 0x08082b2b, 0x2b2b2b08, 0x082b2b08, 0x2b2b2b08, 0x2b2b082b, 0x2b2b2b08, 0x19191908, 0x2b2b2b19,
1184
- 0x2b08192b, 0x2b2b2b19, 0x08082b08, 0x2b2b2b2b, 0x08082b2b, 0x2b2b2b2b, 0x082b0808, 0x2b2b2b2b,
1185
- 0x082b082b, 0x2b2b2b2b, 0x082b2b08, 0x2b2b2b2b, 0x2b082b08, 0x2b2b2b2b, 0x2b2b2b2b, 0x2b2b2b2b
1186
- );
1187
-
1188
- struct iq2_s {
1189
- d: f16,
1190
- qs: array<f16, 32>,
1191
- qh: array<f16, 4>,
1192
- scales: array<f16, 4>
1193
- };
1194
-
1195
598
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
1196
599
  let block = src0[src0_idx_base + offset];
1197
600
  let d = f32(block.d);
@@ -1236,47 +639,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
1236
639
  #enddecl(IQ2_S)
1237
640
 
1238
641
  #decl(IQ3_XSS)
1239
-
1240
- const iq3xxs_grid = array<u32, 256>(
1241
- 0x04040404, 0x04040414, 0x04040424, 0x04040c0c, 0x04040c1c, 0x04040c3e, 0x04041404, 0x04041414,
1242
- 0x04041c0c, 0x04042414, 0x04043e1c, 0x04043e2c, 0x040c040c, 0x040c041c, 0x040c0c04, 0x040c0c14,
1243
- 0x040c140c, 0x040c142c, 0x040c1c04, 0x040c1c14, 0x040c240c, 0x040c2c24, 0x040c3e04, 0x04140404,
1244
- 0x04140414, 0x04140424, 0x04140c0c, 0x04141404, 0x04141414, 0x04141c0c, 0x04141c1c, 0x04141c3e,
1245
- 0x04142c0c, 0x04142c3e, 0x04143e2c, 0x041c040c, 0x041c043e, 0x041c0c04, 0x041c0c14, 0x041c142c,
1246
- 0x041c3e04, 0x04240c1c, 0x04241c3e, 0x04242424, 0x04242c3e, 0x04243e1c, 0x04243e2c, 0x042c040c,
1247
- 0x042c043e, 0x042c1c14, 0x042c2c14, 0x04341c2c, 0x04343424, 0x043e0c04, 0x043e0c24, 0x043e0c34,
1248
- 0x043e241c, 0x043e340c, 0x0c04040c, 0x0c04041c, 0x0c040c04, 0x0c040c14, 0x0c04140c, 0x0c04141c,
1249
- 0x0c041c04, 0x0c041c14, 0x0c041c24, 0x0c04243e, 0x0c042c04, 0x0c0c0404, 0x0c0c0414, 0x0c0c0c0c,
1250
- 0x0c0c1404, 0x0c0c1414, 0x0c14040c, 0x0c14041c, 0x0c140c04, 0x0c140c14, 0x0c14140c, 0x0c141c04,
1251
- 0x0c143e14, 0x0c1c0404, 0x0c1c0414, 0x0c1c1404, 0x0c1c1c0c, 0x0c1c2434, 0x0c1c3434, 0x0c24040c,
1252
- 0x0c24042c, 0x0c242c04, 0x0c2c1404, 0x0c2c1424, 0x0c2c2434, 0x0c2c3e0c, 0x0c34042c, 0x0c3e1414,
1253
- 0x0c3e2404, 0x14040404, 0x14040414, 0x14040c0c, 0x14040c1c, 0x14041404, 0x14041414, 0x14041434,
1254
- 0x14041c0c, 0x14042414, 0x140c040c, 0x140c041c, 0x140c042c, 0x140c0c04, 0x140c0c14, 0x140c140c,
1255
- 0x140c1c04, 0x140c341c, 0x140c343e, 0x140c3e04, 0x14140404, 0x14140414, 0x14140c0c, 0x14140c3e,
1256
- 0x14141404, 0x14141414, 0x14141c3e, 0x14142404, 0x14142c2c, 0x141c040c, 0x141c0c04, 0x141c0c24,
1257
- 0x141c3e04, 0x141c3e24, 0x14241c2c, 0x14242c1c, 0x142c041c, 0x142c143e, 0x142c240c, 0x142c3e24,
1258
- 0x143e040c, 0x143e041c, 0x143e0c34, 0x143e242c, 0x1c04040c, 0x1c040c04, 0x1c040c14, 0x1c04140c,
1259
- 0x1c04141c, 0x1c042c04, 0x1c04342c, 0x1c043e14, 0x1c0c0404, 0x1c0c0414, 0x1c0c1404, 0x1c0c1c0c,
1260
- 0x1c0c2424, 0x1c0c2434, 0x1c14040c, 0x1c14041c, 0x1c140c04, 0x1c14142c, 0x1c142c14, 0x1c143e14,
1261
- 0x1c1c0c0c, 0x1c1c1c1c, 0x1c241c04, 0x1c24243e, 0x1c243e14, 0x1c2c0404, 0x1c2c0434, 0x1c2c1414,
1262
- 0x1c2c2c2c, 0x1c340c24, 0x1c341c34, 0x1c34341c, 0x1c3e1c1c, 0x1c3e3404, 0x24040424, 0x24040c3e,
1263
- 0x24041c2c, 0x24041c3e, 0x24042c1c, 0x24042c3e, 0x240c3e24, 0x24141404, 0x24141c3e, 0x24142404,
1264
- 0x24143404, 0x24143434, 0x241c043e, 0x241c242c, 0x24240424, 0x24242c0c, 0x24243424, 0x242c142c,
1265
- 0x242c241c, 0x242c3e04, 0x243e042c, 0x243e0c04, 0x243e0c14, 0x243e1c04, 0x2c040c14, 0x2c04240c,
1266
- 0x2c043e04, 0x2c0c0404, 0x2c0c0434, 0x2c0c1434, 0x2c0c2c2c, 0x2c140c24, 0x2c141c14, 0x2c143e14,
1267
- 0x2c1c0414, 0x2c1c2c1c, 0x2c240c04, 0x2c24141c, 0x2c24143e, 0x2c243e14, 0x2c2c0414, 0x2c2c1c0c,
1268
- 0x2c342c04, 0x2c3e1424, 0x2c3e2414, 0x34041424, 0x34042424, 0x34042434, 0x34043424, 0x340c140c,
1269
- 0x340c340c, 0x34140c3e, 0x34143424, 0x341c1c04, 0x341c1c34, 0x34242424, 0x342c042c, 0x342c2c14,
1270
- 0x34341c1c, 0x343e041c, 0x343e140c, 0x3e04041c, 0x3e04042c, 0x3e04043e, 0x3e040c04, 0x3e041c14,
1271
- 0x3e042c14, 0x3e0c1434, 0x3e0c2404, 0x3e140c14, 0x3e14242c, 0x3e142c14, 0x3e1c0404, 0x3e1c0c2c,
1272
- 0x3e1c1c1c, 0x3e1c3404, 0x3e24140c, 0x3e24240c, 0x3e2c0404, 0x3e2c0414, 0x3e2c1424, 0x3e341c04
1273
- );
1274
-
1275
- struct iq3_xxs {
1276
- d: f16,
1277
- qs: array<f16, 48>
1278
- };
1279
-
1280
642
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
1281
643
  let block = src0[src0_idx_base + offset];
1282
644
  let d = f32(block.d);
@@ -1309,82 +671,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
1309
671
  #enddecl(IQ3_XSS)
1310
672
 
1311
673
  #decl(IQ3_S)
1312
-
1313
- const iq3s_grid = array<u32, 512>(
1314
- 0x01010101, 0x01010103, 0x01010105, 0x0101010b, 0x0101010f, 0x01010301, 0x01010303, 0x01010305,
1315
- 0x01010309, 0x0101030d, 0x01010501, 0x01010503, 0x0101050b, 0x01010707, 0x01010901, 0x01010905,
1316
- 0x0101090b, 0x0101090f, 0x01010b03, 0x01010b07, 0x01010d01, 0x01010d05, 0x01010f03, 0x01010f09,
1317
- 0x01010f0f, 0x01030101, 0x01030103, 0x01030105, 0x01030109, 0x01030301, 0x01030303, 0x0103030b,
1318
- 0x01030501, 0x01030507, 0x0103050f, 0x01030703, 0x0103070b, 0x01030909, 0x01030d03, 0x01030d0b,
1319
- 0x01030f05, 0x01050101, 0x01050103, 0x0105010b, 0x0105010f, 0x01050301, 0x01050307, 0x0105030d,
1320
- 0x01050503, 0x0105050b, 0x01050701, 0x01050709, 0x01050905, 0x0105090b, 0x0105090f, 0x01050b03,
1321
- 0x01050b07, 0x01050f01, 0x01050f07, 0x01070107, 0x01070303, 0x0107030b, 0x01070501, 0x01070505,
1322
- 0x01070703, 0x01070707, 0x0107070d, 0x01070909, 0x01070b01, 0x01070b05, 0x01070d0f, 0x01070f03,
1323
- 0x01070f0b, 0x01090101, 0x01090307, 0x0109030f, 0x01090503, 0x01090509, 0x01090705, 0x01090901,
1324
- 0x01090907, 0x01090b03, 0x01090f01, 0x010b0105, 0x010b0109, 0x010b0501, 0x010b0505, 0x010b050d,
1325
- 0x010b0707, 0x010b0903, 0x010b090b, 0x010b090f, 0x010b0d0d, 0x010b0f07, 0x010d010d, 0x010d0303,
1326
- 0x010d0307, 0x010d0703, 0x010d0b05, 0x010d0f03, 0x010f0101, 0x010f0105, 0x010f0109, 0x010f0501,
1327
- 0x010f0505, 0x010f050d, 0x010f0707, 0x010f0b01, 0x010f0b09, 0x03010101, 0x03010103, 0x03010105,
1328
- 0x03010109, 0x03010301, 0x03010303, 0x03010307, 0x0301030b, 0x0301030f, 0x03010501, 0x03010505,
1329
- 0x03010703, 0x03010709, 0x0301070d, 0x03010b09, 0x03010b0d, 0x03010d03, 0x03010f05, 0x03030101,
1330
- 0x03030103, 0x03030107, 0x0303010d, 0x03030301, 0x03030309, 0x03030503, 0x03030701, 0x03030707,
1331
- 0x03030903, 0x03030b01, 0x03030b05, 0x03030f01, 0x03030f0d, 0x03050101, 0x03050305, 0x0305030b,
1332
- 0x0305030f, 0x03050501, 0x03050509, 0x03050705, 0x03050901, 0x03050907, 0x03050b0b, 0x03050d01,
1333
- 0x03050f05, 0x03070103, 0x03070109, 0x0307010f, 0x03070301, 0x03070307, 0x03070503, 0x0307050f,
1334
- 0x03070701, 0x03070709, 0x03070903, 0x03070d05, 0x03070f01, 0x03090107, 0x0309010b, 0x03090305,
1335
- 0x03090309, 0x03090703, 0x03090707, 0x03090905, 0x0309090d, 0x03090b01, 0x03090b09, 0x030b0103,
1336
- 0x030b0301, 0x030b0307, 0x030b0503, 0x030b0701, 0x030b0705, 0x030b0b03, 0x030d0501, 0x030d0509,
1337
- 0x030d050f, 0x030d0909, 0x030d090d, 0x030f0103, 0x030f0107, 0x030f0301, 0x030f0305, 0x030f0503,
1338
- 0x030f070b, 0x030f0903, 0x030f0d05, 0x030f0f01, 0x05010101, 0x05010103, 0x05010107, 0x0501010b,
1339
- 0x0501010f, 0x05010301, 0x05010305, 0x05010309, 0x0501030d, 0x05010503, 0x05010507, 0x0501050f,
1340
- 0x05010701, 0x05010705, 0x05010903, 0x05010907, 0x0501090b, 0x05010b01, 0x05010b05, 0x05010d0f,
1341
- 0x05010f01, 0x05010f07, 0x05010f0b, 0x05030101, 0x05030105, 0x05030301, 0x05030307, 0x0503030f,
1342
- 0x05030505, 0x0503050b, 0x05030703, 0x05030709, 0x05030905, 0x05030b03, 0x05050103, 0x05050109,
1343
- 0x0505010f, 0x05050503, 0x05050507, 0x05050701, 0x0505070f, 0x05050903, 0x05050b07, 0x05050b0f,
1344
- 0x05050f03, 0x05050f09, 0x05070101, 0x05070105, 0x0507010b, 0x05070303, 0x05070505, 0x05070509,
1345
- 0x05070703, 0x05070707, 0x05070905, 0x05070b01, 0x05070d0d, 0x05090103, 0x0509010f, 0x05090501,
1346
- 0x05090507, 0x05090705, 0x0509070b, 0x05090903, 0x05090f05, 0x05090f0b, 0x050b0109, 0x050b0303,
1347
- 0x050b0505, 0x050b070f, 0x050b0901, 0x050b0b07, 0x050b0f01, 0x050d0101, 0x050d0105, 0x050d010f,
1348
- 0x050d0503, 0x050d0b0b, 0x050d0d03, 0x050f010b, 0x050f0303, 0x050f050d, 0x050f0701, 0x050f0907,
1349
- 0x050f0b01, 0x07010105, 0x07010303, 0x07010307, 0x0701030b, 0x0701030f, 0x07010505, 0x07010703,
1350
- 0x07010707, 0x0701070b, 0x07010905, 0x07010909, 0x0701090f, 0x07010b03, 0x07010d07, 0x07010f03,
1351
- 0x07030103, 0x07030107, 0x0703010b, 0x07030309, 0x07030503, 0x07030507, 0x07030901, 0x07030d01,
1352
- 0x07030f05, 0x07030f0d, 0x07050101, 0x07050305, 0x07050501, 0x07050705, 0x07050709, 0x07050b01,
1353
- 0x07070103, 0x07070301, 0x07070309, 0x07070503, 0x07070507, 0x0707050f, 0x07070701, 0x07070903,
1354
- 0x07070907, 0x0707090f, 0x07070b0b, 0x07070f07, 0x07090107, 0x07090303, 0x0709030d, 0x07090505,
1355
- 0x07090703, 0x07090b05, 0x07090d01, 0x07090d09, 0x070b0103, 0x070b0301, 0x070b0305, 0x070b050b,
1356
- 0x070b0705, 0x070b0909, 0x070b0b0d, 0x070b0f07, 0x070d030d, 0x070d0903, 0x070f0103, 0x070f0107,
1357
- 0x070f0501, 0x070f0505, 0x070f070b, 0x09010101, 0x09010109, 0x09010305, 0x09010501, 0x09010509,
1358
- 0x0901050f, 0x09010705, 0x09010903, 0x09010b01, 0x09010f01, 0x09030105, 0x0903010f, 0x09030303,
1359
- 0x09030307, 0x09030505, 0x09030701, 0x0903070b, 0x09030907, 0x09030b03, 0x09030b0b, 0x09050103,
1360
- 0x09050107, 0x09050301, 0x0905030b, 0x09050503, 0x09050707, 0x09050901, 0x09050b0f, 0x09050d05,
1361
- 0x09050f01, 0x09070109, 0x09070303, 0x09070307, 0x09070501, 0x09070505, 0x09070703, 0x0907070b,
1362
- 0x09090101, 0x09090105, 0x09090509, 0x0909070f, 0x09090901, 0x09090f03, 0x090b010b, 0x090b010f,
1363
- 0x090b0503, 0x090b0d05, 0x090d0307, 0x090d0709, 0x090d0d01, 0x090f0301, 0x090f030b, 0x090f0701,
1364
- 0x090f0907, 0x090f0b03, 0x0b010105, 0x0b010301, 0x0b010309, 0x0b010505, 0x0b010901, 0x0b010909,
1365
- 0x0b01090f, 0x0b010b05, 0x0b010d0d, 0x0b010f09, 0x0b030103, 0x0b030107, 0x0b03010b, 0x0b030305,
1366
- 0x0b030503, 0x0b030705, 0x0b030f05, 0x0b050101, 0x0b050303, 0x0b050507, 0x0b050701, 0x0b05070d,
1367
- 0x0b050b07, 0x0b070105, 0x0b07010f, 0x0b070301, 0x0b07050f, 0x0b070909, 0x0b070b03, 0x0b070d0b,
1368
- 0x0b070f07, 0x0b090103, 0x0b090109, 0x0b090501, 0x0b090705, 0x0b09090d, 0x0b0b0305, 0x0b0b050d,
1369
- 0x0b0b0b03, 0x0b0b0b07, 0x0b0d0905, 0x0b0f0105, 0x0b0f0109, 0x0b0f0505, 0x0d010303, 0x0d010307,
1370
- 0x0d01030b, 0x0d010703, 0x0d010707, 0x0d010d01, 0x0d030101, 0x0d030501, 0x0d03050f, 0x0d030d09,
1371
- 0x0d050305, 0x0d050709, 0x0d050905, 0x0d050b0b, 0x0d050d05, 0x0d050f01, 0x0d070101, 0x0d070309,
1372
- 0x0d070503, 0x0d070901, 0x0d09050b, 0x0d090907, 0x0d090d05, 0x0d0b0101, 0x0d0b0107, 0x0d0b0709,
1373
- 0x0d0b0d01, 0x0d0d010b, 0x0d0d0901, 0x0d0f0303, 0x0d0f0307, 0x0f010101, 0x0f010109, 0x0f01010f,
1374
- 0x0f010501, 0x0f010505, 0x0f01070d, 0x0f010901, 0x0f010b09, 0x0f010d05, 0x0f030105, 0x0f030303,
1375
- 0x0f030509, 0x0f030907, 0x0f03090b, 0x0f050103, 0x0f050109, 0x0f050301, 0x0f05030d, 0x0f050503,
1376
- 0x0f050701, 0x0f050b03, 0x0f070105, 0x0f070705, 0x0f07070b, 0x0f070b07, 0x0f090103, 0x0f09010b,
1377
- 0x0f090307, 0x0f090501, 0x0f090b01, 0x0f0b0505, 0x0f0b0905, 0x0f0d0105, 0x0f0d0703, 0x0f0f0101
1378
- );
1379
-
1380
- struct iq3_s {
1381
- d: f16,
1382
- qs: array<f16, 32>,
1383
- qh: array<f16, 4>,
1384
- signs: array<f16, 16>,
1385
- scales: array<f16, 2>
1386
- };
1387
-
1388
674
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
1389
675
  let block = src0[src0_idx_base + offset];
1390
676
  let d = f32(block.d);
@@ -1431,151 +717,7 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
1431
717
  }
1432
718
  #enddecl(IQ3_S)
1433
719
 
1434
- #decl(IQ1_TABLE)
1435
-
1436
- const IQ1_DELTA: f32 = 0.125;
1437
-
1438
- const iq1_grid = array<u32, 1024>(
1439
- 0xfffdffff, 0xfff7fff0, 0xffccfff5, 0xffdfffc0, 0xffd7ffdd, 0xff30ffd5, 0xff03ff0c, 0xff10ff01,
1440
- 0xff7dff7f, 0xff75ff77, 0xff5fff40, 0xff57ff5d, 0xfcf3ff55, 0xfcccfcf0, 0xfcc1fcc3, 0xfcc5fcc4,
1441
- 0xfc3cfcd0, 0xfc34fc31, 0xfc00fc0d, 0xfc1cfc05, 0xfc11fc13, 0xfc70fc17, 0xfc43fc4c, 0xfc50fc41,
1442
- 0xfdfdfdff, 0xfdf5fdf7, 0xfddffdc0, 0xfdd7fddd, 0xfd30fdd5, 0xfd04fd0c, 0xfd14fd13, 0xfd7dfd7f,
1443
- 0xfd75fd77, 0xfd40fd4c, 0xfd5ffd44, 0xfd57fd5d, 0xf3ccfd55, 0xf3c1f3c3, 0xf33cf3d0, 0xf300f334,
1444
- 0xf313f305, 0xf34cf310, 0xf350f344, 0xf0f3f0fc, 0xf0f1f0f0, 0xf0c7f0c0, 0xf0d4f0c5, 0xf030f03f,
1445
- 0xf00ff035, 0xf003f00c, 0xf001f000, 0xf01ff004, 0xf010f01d, 0xf015f017, 0xf04cf07c, 0xf047f040,
1446
- 0xf05cf045, 0xf050f053, 0xf054f051, 0xf1c4f1c3, 0xf133f13c, 0xf10df10f, 0xf107f100, 0xf11cf11f,
1447
- 0xf114f111, 0xf14cf170, 0xf144f143, 0xf7fdf7ff, 0xf7f5f7f7, 0xf7dff7c0, 0xf7d7f7dd, 0xf730f7d5,
1448
- 0xf701f70c, 0xf77ff710, 0xf777f77d, 0xf740f775, 0xf75df75f, 0xf755f757, 0xf4ccf4f0, 0xf4c4f4c3,
1449
- 0xf4d0f4d3, 0xf40ff43c, 0xf400f40c, 0xf413f41c, 0xf44cf414, 0xf441f443, 0xf450f444, 0xf5fdf5ff,
1450
- 0xf5f5f5f7, 0xf5dff5c0, 0xf5d7f5dd, 0xf530f5d5, 0xf504f50c, 0xf510f51c, 0xf57df57f, 0xf577f570,
1451
- 0xf540f575, 0xf55df55f, 0xf555f557, 0xcfcccfcf, 0xcfc4cfc3, 0xcfd0cfd3, 0xcf33cf3c, 0xcf00cf0f,
1452
- 0xcf1ccf07, 0xcf10cf13, 0xcf4ccf14, 0xcf41cf43, 0xcf50cf5c, 0xccf3ccfc, 0xccf4ccf1, 0xcccdcccf,
1453
- 0xccc7ccc0, 0xccd3ccdc, 0xcc30ccd4, 0xcc0fcc35, 0xcc0dcc0c, 0xcc00cc03, 0xcc04cc01, 0xcc10cc1f,
1454
- 0xcc4dcc73, 0xcc5ccc40, 0xcdcccc53, 0xcdc1cdc3, 0xcd3fcdd0, 0xcd34cd31, 0xcd00cd0d, 0xcd05cd07,
1455
- 0xcd11cd13, 0xcd4ccd70, 0xcd41cd43, 0xc3fccd50, 0xc3f4c3f1, 0xc3c0c3c3, 0xc3c4c3c7, 0xc3d1c3dc,
1456
- 0xc330c33c, 0xc337c331, 0xc30cc335, 0xc300c303, 0xc304c301, 0xc310c31d, 0xc373c317, 0xc34fc374,
1457
- 0xc340c343, 0xc344c347, 0xc35cc345, 0xc350c353, 0xc0fdc354, 0xc0f5c0f0, 0xc0c3c0cc, 0xc0c1c0c0,
1458
- 0xc0dfc0c4, 0xc0d0c0dd, 0xc0d5c0d7, 0xc033c03c, 0xc031c030, 0xc00dc00c, 0xc000c003, 0xc004c001,
1459
- 0xc01cc005, 0xc010c013, 0xc014c011, 0xc07dc07f, 0xc070c073, 0xc075c077, 0xc04cc04f, 0xc040c043,
1460
- 0xc044c041, 0xc05fc045, 0xc050c05d, 0xc1f3c1fc, 0xc1f1c1f0, 0xc1c1c1c0, 0xc1c5c1c7, 0xc1d1c1dc,
1461
- 0xc13dc13f, 0xc130c133, 0xc135c137, 0xc100c10c, 0xc107c101, 0xc11cc104, 0xc110c113, 0xc114c117,
1462
- 0xc171c115, 0xc14dc175, 0xc153c140, 0xc7ccc154, 0xc7d0c7c1, 0xc733c73c, 0xc734c731, 0xc700c70f,
1463
- 0xc705c707, 0xc71cc71f, 0xc711c713, 0xc770c714, 0xc743c74c, 0xc4cfc750, 0xc4c0c4cd, 0xc4dcc4c5,
1464
- 0xc43dc4d0, 0xc430c433, 0xc40cc437, 0xc400c403, 0xc404c401, 0xc41fc405, 0xc415c410, 0xc44cc474,
1465
- 0xc440c44d, 0xc45cc447, 0xc454c451, 0xc5c1c5f4, 0xc5d1c5d3, 0xc531c533, 0xc50fc534, 0xc500c50d,
1466
- 0xc51cc507, 0xc514c511, 0xc54cc570, 0xc545c541, 0xdffddfff, 0xdff5dff7, 0xdfdfdfc0, 0xdfd0dfdd,
1467
- 0xdfd5dfd7, 0xdf0cdf30, 0xdf1cdf04, 0xdf7fdf10, 0xdf77df7d, 0xdf40df75, 0xdf5ddf5f, 0xdf57df50,
1468
- 0xdcf0df55, 0xdcc3dccc, 0xdcd0dcc4, 0xdc33dc3d, 0xdc00dc34, 0xdc05dc07, 0xdc13dc1c, 0xdc11dc10,
1469
- 0xdc4fdc70, 0xdc44dc41, 0xddfcdc50, 0xddf5ddf7, 0xddc0ddcc, 0xdddddddf, 0xddd5ddd7, 0xdd0cdd30,
1470
- 0xdd04dd01, 0xdd7cdd10, 0xdd75dd77, 0xdd40dd4c, 0xdd5ddd5f, 0xdd55dd57, 0xd3c3d3f0, 0xd3c4d3c1,
1471
- 0xd333d3d0, 0xd331d330, 0xd30dd334, 0xd307d300, 0xd311d305, 0xd34cd370, 0xd344d343, 0xd350d35c,
1472
- 0xd0c0d0f4, 0xd0d4d0dc, 0xd030d03f, 0xd00cd037, 0xd000d003, 0xd01dd004, 0xd017d010, 0xd04fd074,
1473
- 0xd040d043, 0xd045d047, 0xd053d05c, 0xd054d051, 0xd1cfd1f0, 0xd1c4d1cd, 0xd13cd1d0, 0xd100d134,
1474
- 0xd11cd11f, 0xd173d114, 0xd14fd171, 0xd7ffd145, 0xd7f7d7fd, 0xd7c0d7f5, 0xd7ddd7df, 0xd7d5d7d7,
1475
- 0xd70cd730, 0xd710d703, 0xd77dd77f, 0xd775d777, 0xd75dd75f, 0xd755d757, 0xd4ccd4f4, 0xd4c4d4c3,
1476
- 0xd431d4d0, 0xd40dd434, 0xd41cd400, 0xd411d413, 0xd470d414, 0xd441d44f, 0xd453d444, 0xd5ffd450,
1477
- 0xd5f7d5fd, 0xd5dfd5f5, 0xd5d7d5dd, 0xd530d5d5, 0xd501d50c, 0xd510d504, 0xd57dd57f, 0xd575d577,
1478
- 0xd55fd540, 0xd557d55d, 0x3ff0d555, 0x3fc13fcc, 0x3f343fd0, 0x3f003f0d, 0x3f053f07, 0x3f133f1c,
1479
- 0x3f433f11, 0x3f5c3f44, 0x3cff3f51, 0x3cf33cfc, 0x3cf43cf1, 0x3cc03ccd, 0x3cc73cc1, 0x3cdc3cc5,
1480
- 0x3cd43cd1, 0x3c373c30, 0x3c0c3c35, 0x3c003c03, 0x3c043c01, 0x3c103c05, 0x3c153c17, 0x3c733c7c,
1481
- 0x3c4f3c71, 0x3c403c4d, 0x3c5c3c5f, 0x3df03c5d, 0x3dc33dcc, 0x3dd03dc1, 0x3d0d3d3c, 0x3d053d00,
1482
- 0x3d143d13, 0x3d433d74, 0x33fc3d50, 0x33c433c0, 0x333033d4, 0x33353337, 0x3303330c, 0x33013300,
1483
- 0x331d331c, 0x33173310, 0x337c3315, 0x33743371, 0x334d334f, 0x335f3340, 0x3354335c, 0x30fd30fc,
1484
- 0x30f530f0, 0x30c330cc, 0x30c130c0, 0x30df30c4, 0x30d530d0, 0x3033303c, 0x30313030, 0x300f3034,
1485
- 0x3003300c, 0x30013000, 0x30043007, 0x3013301c, 0x30113010, 0x307d3014, 0x30703073, 0x304c3077,
1486
- 0x30403043, 0x30443041, 0x30503045, 0x30553057, 0x31f031fc, 0x31c331f4, 0x31c731c0, 0x31dc31c5,
1487
- 0x31d431d3, 0x313d313f, 0x31373130, 0x310c310f, 0x3100310d, 0x31043101, 0x3110311d, 0x317c3117,
1488
- 0x31753170, 0x31403143, 0x3153315c, 0x37f03151, 0x37c037cc, 0x37d037c5, 0x3734373d, 0x3700370f,
1489
- 0x371c3707, 0x37113713, 0x37703714, 0x3743374c, 0x37443741, 0x34fc3750, 0x34f134f0, 0x34cf34f5,
1490
- 0x34c034c3, 0x34dc34c7, 0x34d134d3, 0x3430343f, 0x340c3435, 0x3403340d, 0x34013400, 0x341f3404,
1491
- 0x3410341d, 0x34153411, 0x34743471, 0x3440344d, 0x34473441, 0x3453345c, 0x34543451, 0x353335c1,
1492
- 0x35343531, 0x35073500, 0x35133505, 0x35433514, 0x0ffc3550, 0x0ff00ff3, 0x0ff40ff1, 0x0fc00fcd,
1493
- 0x0fdc0fc5, 0x0fd40fd3, 0x0f300f3f, 0x0f0c0f37, 0x0f000f03, 0x0f040f01, 0x0f170f10, 0x0f740f71,
1494
- 0x0f470f40, 0x0f5c0f5f, 0x0f540f51, 0x0cf70cf0, 0x0cf50cf4, 0x0cc30ccc, 0x0cc10cc0, 0x0cc40cc7,
1495
- 0x0cd00cdf, 0x0cd70cd1, 0x0c3c0cd5, 0x0c300c33, 0x0c340c31, 0x0c0c0c0f, 0x0c030c0d, 0x0c010c00,
1496
- 0x0c040c07, 0x0c1c0c05, 0x0c100c13, 0x0c140c11, 0x0c700c7d, 0x0c430c4c, 0x0c410c40, 0x0c5f0c44,
1497
- 0x0c550c50, 0x0df10dfc, 0x0dc00dcd, 0x0ddc0dc5, 0x0d3d0dd3, 0x0d350d30, 0x0d030d0c, 0x0d010d00,
1498
- 0x0d1d0d04, 0x0d700d10, 0x0d4d0d4f, 0x0d440d40, 0x0d530d45, 0x03f003f3, 0x03c303cc, 0x03c103c0,
1499
- 0x03c403c7, 0x03d003dc, 0x03d503d7, 0x0333033c, 0x03310330, 0x03350334, 0x030c030f, 0x03000303,
1500
- 0x03070301, 0x03050304, 0x031d031c, 0x03100313, 0x03140311, 0x0377037f, 0x034c0375, 0x03400343,
1501
- 0x03440341, 0x0353035c, 0x03550350, 0x00fd00fc, 0x00f000f3, 0x00f400f1, 0x00cc00cf, 0x00c300cd,
1502
- 0x00c100c0, 0x00c500c4, 0x00d300dc, 0x00d100d0, 0x003f00d4, 0x003d003c, 0x00300033, 0x00370031,
1503
- 0x000f0034, 0x000d000c, 0x00000003, 0x00070001, 0x00050004, 0x001c001f, 0x00100013, 0x00170011,
1504
- 0x00150014, 0x0073007c, 0x00740070, 0x004f0075, 0x0043004c, 0x00410040, 0x00440047, 0x0053005c,
1505
- 0x00510050, 0x01ff0054, 0x01fd01fc, 0x01f101f3, 0x01f401f7, 0x01c301cc, 0x01c701c0, 0x01df01c4,
1506
- 0x01dd01dc, 0x01d001d3, 0x01d701d1, 0x013c01d4, 0x01310130, 0x01340137, 0x010f0135, 0x010d010c,
1507
- 0x01000103, 0x01070101, 0x01050104, 0x0113011c, 0x01140110, 0x0170017d, 0x01770171, 0x01750174,
1508
- 0x0140014c, 0x015d0145, 0x01510150, 0x01540157, 0x07f007f3, 0x07f407f1, 0x07c007cf, 0x07dc07c7,
1509
- 0x073007d5, 0x07350737, 0x0703070c, 0x07010700, 0x07040707, 0x071d071f, 0x07100713, 0x0774077d,
1510
- 0x074d074f, 0x07470740, 0x0754075c, 0x04fd04fc, 0x04f504f0, 0x04c304cc, 0x04c104c0, 0x04d004c4,
1511
- 0x0433043c, 0x04310430, 0x040f0434, 0x040d040c, 0x04000403, 0x04070401, 0x04050404, 0x0413041c,
1512
- 0x04110410, 0x047c0414, 0x04740470, 0x0443044c, 0x04410440, 0x04440447, 0x05f30450, 0x05c005f7,
1513
- 0x05df05c5, 0x05d105d0, 0x053005d4, 0x05340537, 0x0500050c, 0x05070501, 0x051d0504, 0x05170510,
1514
- 0x057c0515, 0x054d0575, 0x05410540, 0x05450547, 0x1ff0055c, 0x1fc11fc3, 0x1fd01fc4, 0x1f0f1f33,
1515
- 0x1f011f00, 0x1f051f07, 0x1f131f1c, 0x1f141f11, 0x1f411f7c, 0x1cfc1f50, 0x1cf11cf3, 0x1ccd1cf4,
1516
- 0x1cdc1cc0, 0x1cd11cdd, 0x1c301cd4, 0x1c0c1c34, 0x1c011c00, 0x1c101c04, 0x1c151c11, 0x1c751c73,
1517
- 0x1c401c4d, 0x1c511c5c, 0x1dcc1c54, 0x1dc41dc1, 0x1d3c1d3f, 0x1d001d31, 0x1d071d01, 0x1d701d1f,
1518
- 0x1d411d4c, 0x13cc1d50, 0x13c013cd, 0x13c513c1, 0x13d113dc, 0x133f13d4, 0x1330133d, 0x13351337,
1519
- 0x1303130c, 0x13011300, 0x13051304, 0x131d131f, 0x13731310, 0x13741370, 0x134d134f, 0x13401343,
1520
- 0x13471341, 0x135c1345, 0x13541353, 0x10f710f0, 0x10cc10f5, 0x10c110c0, 0x103310c4, 0x10311030,
1521
- 0x100f1034, 0x1003100c, 0x10011000, 0x101c1004, 0x10101013, 0x10141011, 0x10741071, 0x104c1075,
1522
- 0x10411040, 0x10451044, 0x1050105d, 0x10571051, 0x11f411fd, 0x11df11c0, 0x11d711d1, 0x113f11d4,
1523
- 0x11371130, 0x110c1135, 0x11001103, 0x11071101, 0x111f1105, 0x11171110, 0x117d117f, 0x11751170,
1524
- 0x11411143, 0x11441147, 0x1153115f, 0x11551151, 0x17c417c1, 0x173c17d0, 0x1700170d, 0x171c1705,
1525
- 0x17701714, 0x1747174c, 0x14fc1751, 0x14cf14f3, 0x14dc14c0, 0x14d114d3, 0x143f14d4, 0x1430143c,
1526
- 0x14371431, 0x1403140c, 0x14011400, 0x141f1404, 0x14151410, 0x1473147d, 0x14401475, 0x1453145c,
1527
- 0x14541450, 0x15c115cc, 0x153c15c7, 0x15341533, 0x1500150f, 0x15051507, 0x15101513, 0x15711514,
1528
- 0x15471543, 0x15511545, 0x7ffd7fff, 0x7ff57ff7, 0x7fdd7fdf, 0x7fd57fd7, 0x7f0f7f30, 0x7f037f0c,
1529
- 0x7f047f01, 0x7f7f7f10, 0x7f777f7d, 0x7f407f75, 0x7f5d7f5f, 0x7f557f57, 0x7ccc7cf0, 0x7cc17cc3,
1530
- 0x7cd07cc4, 0x7c337c3c, 0x7c0f7c34, 0x7c007c0d, 0x7c077c01, 0x7c137c04, 0x7c147c11, 0x7c747c70,
1531
- 0x7c417c43, 0x7c507c44, 0x7dfd7dff, 0x7df57df7, 0x7ddf7dc0, 0x7dd77ddd, 0x7d0c7dd5, 0x7d047d03,
1532
- 0x7d7f7d10, 0x7d777d7d, 0x7d407d75, 0x7d5d7d5f, 0x7d557d57, 0x73c473c3, 0x7333733c, 0x7300730c,
1533
- 0x731c7305, 0x73147313, 0x73447343, 0x70f470fc, 0x70c070cd, 0x70d170c5, 0x703f70d4, 0x7030703c,
1534
- 0x700c7037, 0x70007003, 0x70047001, 0x70107005, 0x70177011, 0x707c7015, 0x70717073, 0x704f7074,
1535
- 0x7040704d, 0x70517047, 0x71c171cc, 0x71d071c4, 0x7133713c, 0x71357134, 0x7100710f, 0x71057104,
1536
- 0x7111711c, 0x71707115, 0x7145714c, 0x77ff7153, 0x77f777fd, 0x77c077f5, 0x77dd77df, 0x77d577d7,
1537
- 0x7730773c, 0x7703770c, 0x77107704, 0x777f7714, 0x7777777d, 0x77407775, 0x775d775f, 0x77557757,
1538
- 0x74f174f0, 0x74c374cc, 0x74d074c1, 0x7433743c, 0x74347431, 0x740d740f, 0x74057400, 0x7413741c,
1539
- 0x74417470, 0x74507444, 0x75fd75ff, 0x75f575f7, 0x75df75c0, 0x75d775dd, 0x753075d5, 0x7503750c,
1540
- 0x757f7501, 0x7577757d, 0x75407575, 0x755d755f, 0x75557557, 0x4fcc4ff0, 0x4fc74fc1, 0x4fd04fc4,
1541
- 0x4f314f3c, 0x4f004f34, 0x4f054f07, 0x4f154f14, 0x4f4c4f70, 0x4f414f43, 0x4f504f44, 0x4cf34cfc,
1542
- 0x4cf44cf1, 0x4cc04ccf, 0x4cc54cc7, 0x4cd34cdc, 0x4cd44cd1, 0x4c304c3f, 0x4c0c4c0f, 0x4c004c03,
1543
- 0x4c044c01, 0x4c104c1d, 0x4c714c73, 0x4c404c4d, 0x4c5c4c47, 0x4c514c53, 0x4df04c54, 0x4dc34dcc,
1544
- 0x4dd04dc4, 0x4d314d33, 0x4d0f4d34, 0x4d004d0d, 0x4d114d07, 0x4d704d14, 0x4d414d43, 0x43fc4d54,
1545
- 0x43f143f3, 0x43c043cf, 0x43d143c7, 0x4335433f, 0x4303430c, 0x43014300, 0x43044307, 0x431c431f,
1546
- 0x4310431d, 0x43714373, 0x4343434d, 0x43474340, 0x4354435c, 0x40f040ff, 0x40f540f7, 0x40cc40cf,
1547
- 0x40c040c3, 0x40c440c1, 0x40d040dc, 0x40d540d4, 0x4033403c, 0x40314030, 0x400f4034, 0x400d400c,
1548
- 0x40004003, 0x40074001, 0x40054004, 0x4013401c, 0x40114010, 0x407c4014, 0x40774070, 0x404d404c,
1549
- 0x40404043, 0x40444041, 0x405f4045, 0x4050405d, 0x40554057, 0x41f341fc, 0x41c041cf, 0x41df41c4,
1550
- 0x41d441d1, 0x41374130, 0x410c4134, 0x4100410d, 0x41044101, 0x41174110, 0x4173417d, 0x41754174,
1551
- 0x4143414d, 0x41534140, 0x41544151, 0x47c147f0, 0x47d047c4, 0x4731473c, 0x470d470f, 0x47014700,
1552
- 0x47134705, 0x47704710, 0x4741474c, 0x47504744, 0x44f144f3, 0x44cf44f4, 0x44c044cd, 0x44c544c7,
1553
- 0x44dc44df, 0x44d144d3, 0x443d443f, 0x44374430, 0x440c4435, 0x44004403, 0x44044401, 0x4410441d,
1554
- 0x44154411, 0x4473447c, 0x444d444f, 0x44454440, 0x4451445c, 0x45c045f0, 0x453345d0, 0x45344531,
1555
- 0x4500450f, 0x451c4507, 0x454c4570, 0x45404543, 0x5fff4541, 0x5ff75ffd, 0x5fc05ff5, 0x5fdd5fdf,
1556
- 0x5fd55fd7, 0x5f0c5f30, 0x5f015f03, 0x5f7f5f04, 0x5f775f7d, 0x5f405f75, 0x5f5d5f5f, 0x5f555f57,
1557
- 0x5cf45cf0, 0x5cc35ccc, 0x5cc45cc1, 0x5c315cc5, 0x5c0c5c34, 0x5c075c00, 0x5c1c5c05, 0x5c705c13,
1558
- 0x5c4d5c4f, 0x5c445c41, 0x5df75dfd, 0x5dcf5df5, 0x5ddd5dc4, 0x5dd55dd7, 0x5d0c5d30, 0x5d045d01,
1559
- 0x5d7f5d10, 0x5d775d7d, 0x5d405d75, 0x5d5d5d5f, 0x5d555d57, 0x53d053c4, 0x5333533c, 0x5303530f,
1560
- 0x53075300, 0x531c5305, 0x53115310, 0x53145317, 0x50f15370, 0x50cf50f4, 0x50c050cd, 0x50d150c7,
1561
- 0x503d50d4, 0x500c5030, 0x50005003, 0x50045001, 0x50155010, 0x5073507c, 0x50715070, 0x504d5074,
1562
- 0x50475040, 0x51cc51f0, 0x51c551c1, 0x51d051dc, 0x51315133, 0x510d5135, 0x51015100, 0x511f5107,
1563
- 0x5171511d, 0x5140514f, 0x51445141, 0x5153515c, 0x57ff5151, 0x57f757fd, 0x57df57f5, 0x57d757dd,
1564
- 0x570c57d5, 0x57015703, 0x577f5704, 0x5777577d, 0x57405775, 0x575d575f, 0x57555757, 0x54c354f0,
1565
- 0x54dc54c4, 0x543c54d0, 0x5400540f, 0x541c5405, 0x54145411, 0x5441544f, 0x55fd55ff, 0x55f555f7,
1566
- 0x55dd55df, 0x55d555d7, 0x5503550c, 0x557f5501, 0x5577557d, 0x55405575, 0x555d555f, 0x55555557
1567
- );
1568
-
1569
- #enddecl(IQ1_TABLE)
1570
-
1571
720
  #decl(IQ1_S)
1572
-
1573
- struct iq1_s {
1574
- d: f16,
1575
- qs: array<f16, 16>,
1576
- qh: array<f16, 8>
1577
- };
1578
-
1579
721
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
1580
722
  let block = src0[src0_idx_base + offset];
1581
723
  let d = f32(block.d);
@@ -1603,13 +745,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
1603
745
  #enddecl(IQ1_S)
1604
746
 
1605
747
  #decl(IQ1_M)
1606
-
1607
- struct iq1_m {
1608
- qs: array<u32, 8>,
1609
- qh: array<u32, 4>,
1610
- scales: array<u32, 2>
1611
- };
1612
-
1613
748
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
1614
749
  let block = src0[src0_idx_base + offset];
1615
750
 
@@ -1655,21 +790,7 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
1655
790
 
1656
791
  #enddecl(IQ1_M)
1657
792
 
1658
- #decl(IQ4_TABLE)
1659
-
1660
- const kvalues_iq4nl = array<i32, 16>(
1661
- -127, -104, -83, -65, -49, -35, -22, -10, 1, 13, 25, 38, 53, 69, 89, 113
1662
- );
1663
-
1664
- #enddecl(IQ4_TABLE)
1665
-
1666
793
  #decl(IQ4_NL)
1667
-
1668
- struct iq4_nl {
1669
- d: f16,
1670
- qs: array<f16, 8>,
1671
- }
1672
-
1673
794
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
1674
795
  let block = src0[src0_idx_base + offset];
1675
796
  let d = f32(block.d);
@@ -1691,14 +812,6 @@ fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
1691
812
  #enddecl(IQ4_NL)
1692
813
 
1693
814
  #decl(IQ4_XS)
1694
-
1695
- struct iq4_xs {
1696
- d: f16,
1697
- scales_h: f16,
1698
- scales_l: u32,
1699
- qs: array<u32, 32>
1700
- };
1701
-
1702
815
  fn multiply_add(src0_idx_base: u32, src1_idx_base: u32, offset: u32) -> f32 {
1703
816
  let block = src0[src0_idx_base + offset];
1704
817
  let d = f32(block.d);
@@ -1751,13 +864,13 @@ struct MulMatParams {
1751
864
  broadcast3: u32
1752
865
  };
1753
866
 
1754
- @group(0) @binding(0) var<storage, read_write> src0: array<{{SRC0_TYPE}}>; // N rows, K columns
1755
- @group(0) @binding(1) var<storage, read_write> src1: array<{{SRC1_TYPE}}>; // M rows, K columns (transposed)
867
+ @group(0) @binding(0) var<storage, read_write> src0: array<{{SRC0_TYPE}}>; // M rows, K columns
868
+ @group(0) @binding(1) var<storage, read_write> src1: array<{{SRC1_TYPE}}>; // K rows, N columns (transposed)
1756
869
  @group(0) @binding(2) var<storage, read_write> dst: array<f32>; // M rows, N columns
1757
870
 
1758
871
  @group(0) @binding(3) var<uniform> params: MulMatParams;
1759
872
 
1760
- @compute @workgroup_size(64)
873
+ @compute @workgroup_size(256)
1761
874
  fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
1762
875
  let total = params.m * params.n * params.bs02 * params.broadcast2 * params.bs03 * params.broadcast3;
1763
876
  if (global_id.x >= total) {
@@ -1778,8 +891,8 @@ fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
1778
891
 
1779
892
  let dst2_rem = dst3_rem % dst2_stride;
1780
893
 
1781
- let row = dst2_rem / params.n; // output row
1782
- let col = dst2_rem % params.n; // output column
894
+ let row = dst2_rem / params.m; // output row
895
+ let col = dst2_rem % params.m; // output column
1783
896
 
1784
897
  let src0_idx_base = params.offset_src0 + src03_idx * params.stride_03 + src02_idx * params.stride_02 + col * params.stride_01;
1785
898
  let src1_idx_base = params.offset_src1 + src13_idx * params.stride_13 + src12_idx * params.stride_12 + row * params.stride_11;
@@ -1788,7 +901,7 @@ fn main(@builtin(global_invocation_id) global_id: vec3<u32>) {
1788
901
  for (var i: u32 = 0u; i < params.k/{{BLOCK_SIZE}}; i = i + 1u) {
1789
902
  sum += multiply_add(src0_idx_base, src1_idx_base, i);
1790
903
  }
1791
- dst[params.offset_dst + dst3_idx * dst3_stride + dst2_idx * dst2_stride + row * params.n + col] = sum;
904
+ dst[params.offset_dst + dst3_idx * dst3_stride + dst2_idx * dst2_stride + row * params.m + col] = sum;
1792
905
  }
1793
906
 
1794
907
  #end(SHADER)