cui-llama.rn 1.6.0 → 1.6.1

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 (195) hide show
  1. package/README.md +35 -7
  2. package/android/src/main/CMakeLists.txt +16 -11
  3. package/android/src/main/java/com/rnllama/LlamaContext.java +4 -1
  4. package/android/src/main/jni.cpp +20 -4
  5. package/android/src/main/jniLibs/arm64-v8a/librnllama.so +0 -0
  6. package/android/src/main/jniLibs/arm64-v8a/librnllama_v8.so +0 -0
  7. package/android/src/main/jniLibs/arm64-v8a/librnllama_v8_2.so +0 -0
  8. package/android/src/main/jniLibs/arm64-v8a/librnllama_v8_2_dotprod.so +0 -0
  9. package/android/src/main/jniLibs/arm64-v8a/librnllama_v8_2_dotprod_i8mm.so +0 -0
  10. package/android/src/main/jniLibs/arm64-v8a/librnllama_v8_2_i8mm.so +0 -0
  11. package/android/src/main/jniLibs/x86_64/librnllama.so +0 -0
  12. package/android/src/main/jniLibs/x86_64/librnllama_x86_64.so +0 -0
  13. package/cpp/LICENSE +21 -0
  14. package/cpp/chat.cpp +1 -1
  15. package/cpp/common.cpp +17 -2
  16. package/cpp/common.h +7 -3
  17. package/cpp/ggml-alloc.c +4 -1
  18. package/cpp/ggml-cpp.h +1 -1
  19. package/cpp/ggml-cpu/amx/amx.cpp +221 -0
  20. package/cpp/ggml-cpu/amx/amx.h +8 -0
  21. package/cpp/ggml-cpu/amx/common.h +91 -0
  22. package/cpp/ggml-cpu/amx/mmq.cpp +2511 -0
  23. package/cpp/ggml-cpu/amx/mmq.h +10 -0
  24. package/{ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers → cpp/ggml-cpu}/binary-ops.h +1 -1
  25. package/cpp/ggml-cpu/common.h +72 -0
  26. package/cpp/{ggml-cpu-aarch64.cpp → ggml-cpu/ggml-cpu-aarch64.cpp} +809 -101
  27. package/cpp/{ggml-cpu.c → ggml-cpu/ggml-cpu.c} +109 -42
  28. package/cpp/{ggml-cpu.cpp → ggml-cpu/ggml-cpu.cpp} +3 -0
  29. package/cpp/{ops.cpp → ggml-cpu/ops.cpp} +246 -160
  30. package/{ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers → cpp/ggml-cpu}/ops.h +2 -20
  31. package/cpp/{sgemm.cpp → ggml-cpu/sgemm.cpp} +501 -0
  32. package/{ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers → cpp/ggml-cpu}/simd-mappings.h +7 -3
  33. package/{ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers → cpp/ggml-cpu}/unary-ops.h +1 -1
  34. package/cpp/ggml-cpu.h +5 -0
  35. package/cpp/ggml-impl.h +16 -9
  36. package/cpp/ggml-llama-sim.metallib +0 -0
  37. package/cpp/ggml-llama.metallib +0 -0
  38. package/cpp/ggml-metal.m +492 -47
  39. package/cpp/ggml.c +134 -244
  40. package/cpp/ggml.h +61 -94
  41. package/cpp/json-schema-to-grammar.cpp +3 -0
  42. package/cpp/llama-arch.cpp +46 -17
  43. package/cpp/llama-arch.h +9 -0
  44. package/cpp/llama-batch.cpp +5 -1
  45. package/cpp/llama-batch.h +2 -1
  46. package/cpp/llama-chat.cpp +31 -10
  47. package/cpp/llama-chat.h +3 -2
  48. package/cpp/llama-context.cpp +104 -489
  49. package/cpp/llama-context.h +14 -30
  50. package/cpp/llama-graph.cpp +69 -62
  51. package/cpp/llama-graph.h +21 -18
  52. package/cpp/llama-hparams.h +5 -0
  53. package/cpp/llama-kv-cache.cpp +1497 -391
  54. package/cpp/llama-kv-cache.h +272 -80
  55. package/cpp/llama-memory.h +11 -1
  56. package/cpp/llama-model.cpp +502 -176
  57. package/cpp/llama-model.h +13 -3
  58. package/cpp/llama-sampling.cpp +2 -1
  59. package/cpp/llama-vocab.cpp +8 -1
  60. package/cpp/llama.h +14 -11
  61. package/cpp/rn-llama.cpp +20 -172
  62. package/cpp/rn-llama.h +1 -5
  63. package/ios/CMakeLists.txt +13 -10
  64. package/ios/RNLlama.h +6 -0
  65. package/ios/RNLlama.mm +5 -0
  66. package/ios/RNLlamaContext.mm +26 -28
  67. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/common.h +7 -3
  68. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/ggml-cpp.h +1 -1
  69. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/ggml-cpu.h +5 -0
  70. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/ggml-impl.h +16 -9
  71. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/ggml.h +61 -94
  72. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/llama-arch.h +9 -0
  73. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/llama-batch.h +2 -1
  74. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/llama-chat.h +3 -2
  75. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/llama-context.h +14 -30
  76. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/llama-graph.h +21 -18
  77. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/llama-hparams.h +5 -0
  78. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/llama-kv-cache.h +272 -80
  79. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/llama-memory.h +11 -1
  80. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/llama-model.h +13 -3
  81. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/llama.h +14 -11
  82. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/rn-llama.h +1 -5
  83. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/ggml-llama.metallib +0 -0
  84. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/rnllama +0 -0
  85. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/common.h +7 -3
  86. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/ggml-cpp.h +1 -1
  87. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/ggml-cpu.h +5 -0
  88. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/ggml-impl.h +16 -9
  89. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/ggml.h +61 -94
  90. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/llama-arch.h +9 -0
  91. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/llama-batch.h +2 -1
  92. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/llama-chat.h +3 -2
  93. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/llama-context.h +14 -30
  94. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/llama-graph.h +21 -18
  95. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/llama-hparams.h +5 -0
  96. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/llama-kv-cache.h +272 -80
  97. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/llama-memory.h +11 -1
  98. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/llama-model.h +13 -3
  99. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/llama.h +14 -11
  100. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/rn-llama.h +1 -5
  101. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/ggml-llama-sim.metallib +0 -0
  102. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/rnllama +0 -0
  103. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/common.h +7 -3
  104. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/ggml-cpp.h +1 -1
  105. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/ggml-cpu.h +5 -0
  106. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/ggml-impl.h +16 -9
  107. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/ggml.h +61 -94
  108. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/llama-arch.h +9 -0
  109. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/llama-batch.h +2 -1
  110. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/llama-chat.h +3 -2
  111. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/llama-context.h +14 -30
  112. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/llama-graph.h +21 -18
  113. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/llama-hparams.h +5 -0
  114. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/llama-kv-cache.h +272 -80
  115. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/llama-memory.h +11 -1
  116. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/llama-model.h +13 -3
  117. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/llama.h +14 -11
  118. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/rn-llama.h +1 -5
  119. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/ggml-llama.metallib +0 -0
  120. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/rnllama +0 -0
  121. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/common.h +7 -3
  122. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/ggml-cpp.h +1 -1
  123. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/ggml-cpu.h +5 -0
  124. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/ggml-impl.h +16 -9
  125. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/ggml.h +61 -94
  126. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/llama-arch.h +9 -0
  127. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/llama-batch.h +2 -1
  128. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/llama-chat.h +3 -2
  129. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/llama-context.h +14 -30
  130. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/llama-graph.h +21 -18
  131. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/llama-hparams.h +5 -0
  132. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/llama-kv-cache.h +272 -80
  133. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/llama-memory.h +11 -1
  134. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/llama-model.h +13 -3
  135. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/llama.h +14 -11
  136. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/rn-llama.h +1 -5
  137. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/ggml-llama-sim.metallib +0 -0
  138. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/rnllama +0 -0
  139. package/lib/commonjs/NativeRNLlama.js.map +1 -1
  140. package/lib/module/NativeRNLlama.js.map +1 -1
  141. package/lib/typescript/NativeRNLlama.d.ts +4 -0
  142. package/lib/typescript/NativeRNLlama.d.ts.map +1 -1
  143. package/package.json +1 -1
  144. package/src/NativeRNLlama.ts +5 -0
  145. package/cpp/binary-ops.h +0 -16
  146. package/cpp/ops.h +0 -128
  147. package/cpp/simd-mappings.h +0 -888
  148. package/cpp/unary-ops.h +0 -28
  149. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/binary-ops.h +0 -16
  150. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/ggml-cpu-aarch64.h +0 -8
  151. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/ggml-cpu-impl.h +0 -512
  152. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/ggml-cpu-quants.h +0 -63
  153. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/ggml-cpu-traits.h +0 -38
  154. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/ops.h +0 -128
  155. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/sgemm.h +0 -14
  156. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/simd-mappings.h +0 -888
  157. package/ios/rnllama.xcframework/ios-arm64/rnllama.framework/Headers/vec.h +0 -802
  158. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/ggml-cpu-aarch64.h +0 -8
  159. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/ggml-cpu-impl.h +0 -512
  160. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/ggml-cpu-quants.h +0 -63
  161. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/ggml-cpu-traits.h +0 -38
  162. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/sgemm.h +0 -14
  163. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/unary-ops.h +0 -28
  164. package/ios/rnllama.xcframework/ios-arm64_x86_64-simulator/rnllama.framework/Headers/vec.h +0 -802
  165. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/binary-ops.h +0 -16
  166. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/ggml-cpu-aarch64.h +0 -8
  167. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/ggml-cpu-impl.h +0 -512
  168. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/ggml-cpu-quants.h +0 -63
  169. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/ggml-cpu-traits.h +0 -38
  170. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/ops.h +0 -128
  171. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/sgemm.h +0 -14
  172. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/simd-mappings.h +0 -888
  173. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/unary-ops.h +0 -28
  174. package/ios/rnllama.xcframework/tvos-arm64/rnllama.framework/Headers/vec.h +0 -802
  175. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/binary-ops.h +0 -16
  176. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/ggml-cpu-aarch64.h +0 -8
  177. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/ggml-cpu-impl.h +0 -512
  178. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/ggml-cpu-quants.h +0 -63
  179. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/ggml-cpu-traits.h +0 -38
  180. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/ops.h +0 -128
  181. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/sgemm.h +0 -14
  182. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/simd-mappings.h +0 -888
  183. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/unary-ops.h +0 -28
  184. package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/vec.h +0 -802
  185. /package/cpp/{binary-ops.cpp → ggml-cpu/binary-ops.cpp} +0 -0
  186. /package/cpp/{ggml-cpu-aarch64.h → ggml-cpu/ggml-cpu-aarch64.h} +0 -0
  187. /package/cpp/{ggml-cpu-impl.h → ggml-cpu/ggml-cpu-impl.h} +0 -0
  188. /package/cpp/{ggml-cpu-quants.c → ggml-cpu/ggml-cpu-quants.c} +0 -0
  189. /package/cpp/{ggml-cpu-quants.h → ggml-cpu/ggml-cpu-quants.h} +0 -0
  190. /package/cpp/{ggml-cpu-traits.cpp → ggml-cpu/ggml-cpu-traits.cpp} +0 -0
  191. /package/cpp/{ggml-cpu-traits.h → ggml-cpu/ggml-cpu-traits.h} +0 -0
  192. /package/cpp/{sgemm.h → ggml-cpu/sgemm.h} +0 -0
  193. /package/cpp/{unary-ops.cpp → ggml-cpu/unary-ops.cpp} +0 -0
  194. /package/cpp/{vec.cpp → ggml-cpu/vec.cpp} +0 -0
  195. /package/cpp/{vec.h → ggml-cpu/vec.h} +0 -0
package/ios/rnllama.xcframework/tvos-arm64_x86_64-simulator/rnllama.framework/Headers/unary-ops.h DELETED
@@ -1,28 +0,0 @@
1
- #pragma once
2
-
3
- #include "cpu-common.h"
4
-
5
- #ifdef __cplusplus
6
- extern "C" {
7
- #endif
8
-
9
- void lm_ggml_compute_forward_abs(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
10
- void lm_ggml_compute_forward_sgn(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
11
- void lm_ggml_compute_forward_neg(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
12
- void lm_ggml_compute_forward_step(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
13
- void lm_ggml_compute_forward_tanh(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
14
- void lm_ggml_compute_forward_elu(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
15
- void lm_ggml_compute_forward_relu(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
16
- void lm_ggml_compute_forward_sigmoid(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
17
- void lm_ggml_compute_forward_hardsigmoid(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
18
- void lm_ggml_compute_forward_exp(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
19
- void lm_ggml_compute_forward_hardswish(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
20
- void lm_ggml_compute_forward_sqr(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
21
- void lm_ggml_compute_forward_sqrt(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
22
- void lm_ggml_compute_forward_sin(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
23
- void lm_ggml_compute_forward_cos(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
24
- void lm_ggml_compute_forward_log(const struct lm_ggml_compute_params * params, struct lm_ggml_tensor * dst);
25
-
26
- #ifdef __cplusplus
27
- }
28
- #endif