cui-llama.rn 1.4.3 → 1.4.6

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 (134) hide show
  1. package/README.md +93 -114
  2. package/android/src/main/CMakeLists.txt +5 -0
  3. package/android/src/main/java/com/rnllama/LlamaContext.java +91 -17
  4. package/android/src/main/java/com/rnllama/RNLlama.java +37 -4
  5. package/android/src/main/jni-utils.h +6 -0
  6. package/android/src/main/jni.cpp +289 -31
  7. package/android/src/main/jniLibs/arm64-v8a/librnllama.so +0 -0
  8. package/android/src/main/jniLibs/arm64-v8a/librnllama_v8.so +0 -0
  9. package/android/src/main/jniLibs/arm64-v8a/librnllama_v8_2.so +0 -0
  10. package/android/src/main/jniLibs/arm64-v8a/librnllama_v8_2_dotprod.so +0 -0
  11. package/android/src/main/jniLibs/arm64-v8a/librnllama_v8_2_dotprod_i8mm.so +0 -0
  12. package/android/src/main/jniLibs/arm64-v8a/librnllama_v8_2_i8mm.so +0 -0
  13. package/android/src/main/jniLibs/x86_64/librnllama.so +0 -0
  14. package/android/src/main/jniLibs/x86_64/librnllama_x86_64.so +0 -0
  15. package/android/src/newarch/java/com/rnllama/RNLlamaModule.java +7 -2
  16. package/android/src/oldarch/java/com/rnllama/RNLlamaModule.java +7 -2
  17. package/cpp/chat-template.hpp +529 -0
  18. package/cpp/chat.cpp +1779 -0
  19. package/cpp/chat.h +135 -0
  20. package/cpp/common.cpp +2064 -1873
  21. package/cpp/common.h +700 -699
  22. package/cpp/ggml-alloc.c +1039 -1042
  23. package/cpp/ggml-alloc.h +1 -1
  24. package/cpp/ggml-backend-impl.h +255 -255
  25. package/cpp/ggml-backend-reg.cpp +586 -582
  26. package/cpp/ggml-backend.cpp +2004 -2002
  27. package/cpp/ggml-backend.h +354 -354
  28. package/cpp/ggml-common.h +1851 -1853
  29. package/cpp/ggml-cpp.h +39 -39
  30. package/cpp/ggml-cpu-aarch64.cpp +4248 -4247
  31. package/cpp/ggml-cpu-aarch64.h +8 -8
  32. package/cpp/ggml-cpu-impl.h +531 -386
  33. package/cpp/ggml-cpu-quants.c +12527 -10920
  34. package/cpp/ggml-cpu-traits.cpp +36 -36
  35. package/cpp/ggml-cpu-traits.h +38 -38
  36. package/cpp/ggml-cpu.c +15766 -14391
  37. package/cpp/ggml-cpu.cpp +655 -635
  38. package/cpp/ggml-cpu.h +138 -135
  39. package/cpp/ggml-impl.h +567 -567
  40. package/cpp/ggml-metal-impl.h +235 -0
  41. package/cpp/ggml-metal.h +1 -1
  42. package/cpp/ggml-metal.m +5146 -4884
  43. package/cpp/ggml-opt.cpp +854 -854
  44. package/cpp/ggml-opt.h +216 -216
  45. package/cpp/ggml-quants.c +5238 -5238
  46. package/cpp/ggml-threading.h +14 -14
  47. package/cpp/ggml.c +6529 -6514
  48. package/cpp/ggml.h +2198 -2194
  49. package/cpp/gguf.cpp +1329 -1329
  50. package/cpp/gguf.h +202 -202
  51. package/cpp/json-schema-to-grammar.cpp +1024 -1045
  52. package/cpp/json-schema-to-grammar.h +21 -8
  53. package/cpp/json.hpp +24766 -24766
  54. package/cpp/llama-adapter.cpp +347 -347
  55. package/cpp/llama-adapter.h +74 -74
  56. package/cpp/llama-arch.cpp +1513 -1487
  57. package/cpp/llama-arch.h +403 -400
  58. package/cpp/llama-batch.cpp +368 -368
  59. package/cpp/llama-batch.h +88 -88
  60. package/cpp/llama-chat.cpp +588 -578
  61. package/cpp/llama-chat.h +53 -52
  62. package/cpp/llama-context.cpp +1775 -1775
  63. package/cpp/llama-context.h +128 -128
  64. package/cpp/llama-cparams.cpp +1 -1
  65. package/cpp/llama-cparams.h +37 -37
  66. package/cpp/llama-cpp.h +30 -30
  67. package/cpp/llama-grammar.cpp +1219 -1139
  68. package/cpp/llama-grammar.h +173 -143
  69. package/cpp/llama-hparams.cpp +71 -71
  70. package/cpp/llama-hparams.h +139 -139
  71. package/cpp/llama-impl.cpp +167 -167
  72. package/cpp/llama-impl.h +61 -61
  73. package/cpp/llama-kv-cache.cpp +718 -718
  74. package/cpp/llama-kv-cache.h +219 -218
  75. package/cpp/llama-mmap.cpp +600 -590
  76. package/cpp/llama-mmap.h +68 -67
  77. package/cpp/llama-model-loader.cpp +1124 -1124
  78. package/cpp/llama-model-loader.h +167 -167
  79. package/cpp/llama-model.cpp +4087 -3997
  80. package/cpp/llama-model.h +370 -370
  81. package/cpp/llama-sampling.cpp +2558 -2408
  82. package/cpp/llama-sampling.h +32 -32
  83. package/cpp/llama-vocab.cpp +3264 -3247
  84. package/cpp/llama-vocab.h +125 -125
  85. package/cpp/llama.cpp +10284 -10077
  86. package/cpp/llama.h +1354 -1323
  87. package/cpp/log.cpp +393 -401
  88. package/cpp/log.h +132 -121
  89. package/cpp/minja/chat-template.hpp +529 -0
  90. package/cpp/minja/minja.hpp +2915 -0
  91. package/cpp/minja.hpp +2915 -0
  92. package/cpp/rn-llama.cpp +66 -6
  93. package/cpp/rn-llama.h +26 -1
  94. package/cpp/sampling.cpp +570 -505
  95. package/cpp/sampling.h +3 -0
  96. package/cpp/sgemm.cpp +2598 -2597
  97. package/cpp/sgemm.h +14 -14
  98. package/cpp/speculative.cpp +278 -277
  99. package/cpp/speculative.h +28 -28
  100. package/cpp/unicode.cpp +9 -2
  101. package/ios/CMakeLists.txt +6 -0
  102. package/ios/RNLlama.h +0 -8
  103. package/ios/RNLlama.mm +27 -3
  104. package/ios/RNLlamaContext.h +10 -1
  105. package/ios/RNLlamaContext.mm +269 -57
  106. package/jest/mock.js +21 -2
  107. package/lib/commonjs/NativeRNLlama.js.map +1 -1
  108. package/lib/commonjs/grammar.js +3 -0
  109. package/lib/commonjs/grammar.js.map +1 -1
  110. package/lib/commonjs/index.js +87 -13
  111. package/lib/commonjs/index.js.map +1 -1
  112. package/lib/module/NativeRNLlama.js.map +1 -1
  113. package/lib/module/grammar.js +3 -0
  114. package/lib/module/grammar.js.map +1 -1
  115. package/lib/module/index.js +86 -13
  116. package/lib/module/index.js.map +1 -1
  117. package/lib/typescript/NativeRNLlama.d.ts +107 -2
  118. package/lib/typescript/NativeRNLlama.d.ts.map +1 -1
  119. package/lib/typescript/grammar.d.ts.map +1 -1
  120. package/lib/typescript/index.d.ts +32 -7
  121. package/lib/typescript/index.d.ts.map +1 -1
  122. package/llama-rn.podspec +1 -1
  123. package/package.json +3 -2
  124. package/src/NativeRNLlama.ts +115 -3
  125. package/src/grammar.ts +3 -0
  126. package/src/index.ts +138 -21
  127. package/android/src/main/build-arm64/CMakeFiles/3.31.4/CMakeCCompiler.cmake +0 -81
  128. package/android/src/main/build-arm64/CMakeFiles/3.31.4/CMakeSystem.cmake +0 -15
  129. package/android/src/main/build-arm64/CMakeFiles/3.31.4/CompilerIdC/CMakeCCompilerId.c +0 -904
  130. package/android/src/main/build-arm64/CMakeFiles/3.31.4/CompilerIdC/CMakeCCompilerId.o +0 -0
  131. package/android/src/main/build-arm64/CMakeFiles/3.31.4/CompilerIdCXX/CMakeCXXCompilerId.cpp +0 -919
  132. package/android/src/main/build-arm64/CMakeFiles/3.31.4/CompilerIdCXX/CMakeCXXCompilerId.o +0 -0
  133. package/android/src/main/build-arm64/CMakeFiles/CMakeConfigureLog.yaml +0 -55
  134. package/cpp/rn-llama.hpp +0 -913
package/cpp/llama-chat.h CHANGED
@@ -1,52 +1,53 @@
1
- #pragma once
2
-
3
- #include <string>
4
- #include <vector>
5
- #include <cstdint>
6
-
7
- enum llm_chat_template {
8
- LLM_CHAT_TEMPLATE_CHATML,
9
- LLM_CHAT_TEMPLATE_LLAMA_2,
10
- LLM_CHAT_TEMPLATE_LLAMA_2_SYS,
11
- LLM_CHAT_TEMPLATE_LLAMA_2_SYS_BOS,
12
- LLM_CHAT_TEMPLATE_LLAMA_2_SYS_STRIP,
13
- LLM_CHAT_TEMPLATE_MISTRAL_V1,
14
- LLM_CHAT_TEMPLATE_MISTRAL_V3,
15
- LLM_CHAT_TEMPLATE_MISTRAL_V3_TEKKEN,
16
- LLM_CHAT_TEMPLATE_MISTRAL_V7,
17
- LLM_CHAT_TEMPLATE_PHI_3,
18
- LLM_CHAT_TEMPLATE_PHI_4,
19
- LLM_CHAT_TEMPLATE_FALCON_3,
20
- LLM_CHAT_TEMPLATE_ZEPHYR,
21
- LLM_CHAT_TEMPLATE_MONARCH,
22
- LLM_CHAT_TEMPLATE_GEMMA,
23
- LLM_CHAT_TEMPLATE_ORION,
24
- LLM_CHAT_TEMPLATE_OPENCHAT,
25
- LLM_CHAT_TEMPLATE_VICUNA,
26
- LLM_CHAT_TEMPLATE_VICUNA_ORCA,
27
- LLM_CHAT_TEMPLATE_DEEPSEEK,
28
- LLM_CHAT_TEMPLATE_DEEPSEEK_2,
29
- LLM_CHAT_TEMPLATE_DEEPSEEK_3,
30
- LLM_CHAT_TEMPLATE_COMMAND_R,
31
- LLM_CHAT_TEMPLATE_LLAMA_3,
32
- LLM_CHAT_TEMPLATE_CHATGML_3,
33
- LLM_CHAT_TEMPLATE_CHATGML_4,
34
- LLM_CHAT_TEMPLATE_MINICPM,
35
- LLM_CHAT_TEMPLATE_EXAONE_3,
36
- LLM_CHAT_TEMPLATE_RWKV_WORLD,
37
- LLM_CHAT_TEMPLATE_GRANITE,
38
- LLM_CHAT_TEMPLATE_GIGACHAT,
39
- LLM_CHAT_TEMPLATE_MEGREZ,
40
- LLM_CHAT_TEMPLATE_UNKNOWN,
41
- };
42
-
43
- struct llama_chat_message;
44
-
45
- llm_chat_template llm_chat_template_from_str(const std::string & name);
46
-
47
- llm_chat_template llm_chat_detect_template(const std::string & tmpl);
48
-
49
- int32_t llm_chat_apply_template(
50
- llm_chat_template tmpl,
51
- const std::vector<const llama_chat_message *> & chat,
52
- std::string & dest, bool add_ass);
1
+ #pragma once
2
+
3
+ #include <string>
4
+ #include <vector>
5
+ #include <cstdint>
6
+
7
+ enum llm_chat_template {
8
+ LLM_CHAT_TEMPLATE_CHATML,
9
+ LLM_CHAT_TEMPLATE_LLAMA_2,
10
+ LLM_CHAT_TEMPLATE_LLAMA_2_SYS,
11
+ LLM_CHAT_TEMPLATE_LLAMA_2_SYS_BOS,
12
+ LLM_CHAT_TEMPLATE_LLAMA_2_SYS_STRIP,
13
+ LLM_CHAT_TEMPLATE_MISTRAL_V1,
14
+ LLM_CHAT_TEMPLATE_MISTRAL_V3,
15
+ LLM_CHAT_TEMPLATE_MISTRAL_V3_TEKKEN,
16
+ LLM_CHAT_TEMPLATE_MISTRAL_V7,
17
+ LLM_CHAT_TEMPLATE_PHI_3,
18
+ LLM_CHAT_TEMPLATE_PHI_4,
19
+ LLM_CHAT_TEMPLATE_FALCON_3,
20
+ LLM_CHAT_TEMPLATE_ZEPHYR,
21
+ LLM_CHAT_TEMPLATE_MONARCH,
22
+ LLM_CHAT_TEMPLATE_GEMMA,
23
+ LLM_CHAT_TEMPLATE_ORION,
24
+ LLM_CHAT_TEMPLATE_OPENCHAT,
25
+ LLM_CHAT_TEMPLATE_VICUNA,
26
+ LLM_CHAT_TEMPLATE_VICUNA_ORCA,
27
+ LLM_CHAT_TEMPLATE_DEEPSEEK,
28
+ LLM_CHAT_TEMPLATE_DEEPSEEK_2,
29
+ LLM_CHAT_TEMPLATE_DEEPSEEK_3,
30
+ LLM_CHAT_TEMPLATE_COMMAND_R,
31
+ LLM_CHAT_TEMPLATE_LLAMA_3,
32
+ LLM_CHAT_TEMPLATE_CHATGML_3,
33
+ LLM_CHAT_TEMPLATE_CHATGML_4,
34
+ LLM_CHAT_TEMPLATE_GLMEDGE,
35
+ LLM_CHAT_TEMPLATE_MINICPM,
36
+ LLM_CHAT_TEMPLATE_EXAONE_3,
37
+ LLM_CHAT_TEMPLATE_RWKV_WORLD,
38
+ LLM_CHAT_TEMPLATE_GRANITE,
39
+ LLM_CHAT_TEMPLATE_GIGACHAT,
40
+ LLM_CHAT_TEMPLATE_MEGREZ,
41
+ LLM_CHAT_TEMPLATE_UNKNOWN,
42
+ };
43
+
44
+ struct llama_chat_message;
45
+
46
+ llm_chat_template llm_chat_template_from_str(const std::string & name);
47
+
48
+ llm_chat_template llm_chat_detect_template(const std::string & tmpl);
49
+
50
+ int32_t llm_chat_apply_template(
51
+ llm_chat_template tmpl,
52
+ const std::vector<const llama_chat_message *> & chat,
53
+ std::string & dest, bool add_ass);