whispercpp 1.2.0.2 → 1.3.1

Sign up to get free protection for your applications and to get access to all the features.
Files changed (135) hide show
  1. checksums.yaml +4 -4
  2. data/.gitignore +5 -0
  3. data/LICENSE +1 -1
  4. data/README.md +165 -434
  5. data/Rakefile +46 -86
  6. data/ext/.gitignore +13 -0
  7. data/ext/cpu.mk +9 -0
  8. data/ext/{dr_wav.h → examples/dr_wav.h} +3560 -1179
  9. data/ext/extconf.rb +185 -7
  10. data/ext/ggml/include/ggml-alloc.h +76 -0
  11. data/ext/ggml/include/ggml-backend.h +352 -0
  12. data/ext/ggml/include/ggml-blas.h +25 -0
  13. data/ext/ggml/include/ggml-cann.h +123 -0
  14. data/ext/ggml/include/ggml-cpp.h +38 -0
  15. data/ext/ggml/include/ggml-cpu.h +135 -0
  16. data/ext/ggml/include/ggml-cuda.h +47 -0
  17. data/ext/ggml/include/ggml-kompute.h +50 -0
  18. data/ext/ggml/include/ggml-metal.h +66 -0
  19. data/ext/ggml/include/ggml-opencl.h +26 -0
  20. data/ext/ggml/include/ggml-opt.h +216 -0
  21. data/ext/ggml/include/ggml-rpc.h +28 -0
  22. data/ext/ggml/include/ggml-sycl.h +49 -0
  23. data/ext/ggml/include/ggml-vulkan.h +31 -0
  24. data/ext/ggml/include/ggml.h +2285 -0
  25. data/ext/ggml/src/ggml-alloc.c +1037 -0
  26. data/ext/ggml/src/ggml-amx/common.h +94 -0
  27. data/ext/ggml/src/ggml-amx/ggml-amx.cpp +446 -0
  28. data/ext/ggml/src/ggml-amx/mmq.cpp +2510 -0
  29. data/ext/ggml/src/ggml-amx/mmq.h +17 -0
  30. data/ext/ggml/src/ggml-backend-impl.h +256 -0
  31. data/ext/ggml/src/ggml-backend-reg.cpp +552 -0
  32. data/ext/ggml/src/ggml-backend.cpp +1999 -0
  33. data/ext/ggml/src/ggml-blas/ggml-blas.cpp +517 -0
  34. data/ext/ggml/src/ggml-cann/acl_tensor.cpp +175 -0
  35. data/ext/ggml/src/ggml-cann/acl_tensor.h +258 -0
  36. data/ext/ggml/src/ggml-cann/aclnn_ops.cpp +3427 -0
  37. data/ext/ggml/src/ggml-cann/aclnn_ops.h +592 -0
  38. data/ext/ggml/src/ggml-cann/common.h +286 -0
  39. data/ext/ggml/src/ggml-cann/ggml-cann.cpp +2188 -0
  40. data/ext/ggml/src/ggml-cann/kernels/ascendc_kernels.h +19 -0
  41. data/ext/ggml/src/ggml-cann/kernels/dup.cpp +236 -0
  42. data/ext/ggml/src/ggml-cann/kernels/get_row_f16.cpp +197 -0
  43. data/ext/ggml/src/ggml-cann/kernels/get_row_f32.cpp +190 -0
  44. data/ext/ggml/src/ggml-cann/kernels/get_row_q4_0.cpp +204 -0
  45. data/ext/ggml/src/ggml-cann/kernels/get_row_q8_0.cpp +191 -0
  46. data/ext/ggml/src/ggml-cann/kernels/quantize_f16_q8_0.cpp +218 -0
  47. data/ext/ggml/src/ggml-cann/kernels/quantize_f32_q8_0.cpp +216 -0
  48. data/ext/ggml/src/ggml-cann/kernels/quantize_float_to_q4_0.cpp +295 -0
  49. data/ext/ggml/src/ggml-common.h +1853 -0
  50. data/ext/ggml/src/ggml-cpu/amx/amx.cpp +220 -0
  51. data/ext/ggml/src/ggml-cpu/amx/amx.h +8 -0
  52. data/ext/ggml/src/ggml-cpu/amx/common.h +91 -0
  53. data/ext/ggml/src/ggml-cpu/amx/mmq.cpp +2511 -0
  54. data/ext/ggml/src/ggml-cpu/amx/mmq.h +10 -0
  55. data/ext/ggml/src/ggml-cpu/cpu-feats-x86.cpp +323 -0
  56. data/ext/ggml/src/ggml-cpu/ggml-cpu-aarch64.cpp +4262 -0
  57. data/ext/ggml/src/ggml-cpu/ggml-cpu-aarch64.h +8 -0
  58. data/ext/ggml/src/ggml-cpu/ggml-cpu-hbm.cpp +55 -0
  59. data/ext/ggml/src/ggml-cpu/ggml-cpu-hbm.h +8 -0
  60. data/ext/ggml/src/ggml-cpu/ggml-cpu-impl.h +386 -0
  61. data/ext/ggml/src/ggml-cpu/ggml-cpu-quants.c +10835 -0
  62. data/ext/ggml/src/ggml-cpu/ggml-cpu-quants.h +63 -0
  63. data/ext/ggml/src/ggml-cpu/ggml-cpu-traits.cpp +36 -0
  64. data/ext/ggml/src/ggml-cpu/ggml-cpu-traits.h +38 -0
  65. data/ext/ggml/src/ggml-cpu/ggml-cpu.c +14123 -0
  66. data/ext/ggml/src/ggml-cpu/ggml-cpu.cpp +622 -0
  67. data/ext/ggml/src/ggml-cpu/llamafile/sgemm.cpp +1884 -0
  68. data/ext/ggml/src/ggml-cpu/llamafile/sgemm.h +14 -0
  69. data/ext/ggml/src/ggml-cuda/vendors/cuda.h +14 -0
  70. data/ext/ggml/src/ggml-cuda/vendors/hip.h +186 -0
  71. data/ext/ggml/src/ggml-cuda/vendors/musa.h +134 -0
  72. data/ext/ggml/src/ggml-impl.h +556 -0
  73. data/ext/ggml/src/ggml-kompute/ggml-kompute.cpp +2251 -0
  74. data/ext/ggml/src/ggml-metal/ggml-metal-impl.h +288 -0
  75. data/ext/ggml/src/ggml-metal/ggml-metal.m +4884 -0
  76. data/ext/ggml/src/ggml-metal/ggml-metal.metal +6732 -0
  77. data/ext/ggml/src/ggml-opt.cpp +854 -0
  78. data/ext/ggml/src/ggml-quants.c +5238 -0
  79. data/ext/ggml/src/ggml-quants.h +100 -0
  80. data/ext/ggml/src/ggml-rpc/ggml-rpc.cpp +1406 -0
  81. data/ext/ggml/src/ggml-sycl/common.cpp +95 -0
  82. data/ext/ggml/src/ggml-sycl/concat.cpp +196 -0
  83. data/ext/ggml/src/ggml-sycl/conv.cpp +99 -0
  84. data/ext/ggml/src/ggml-sycl/convert.cpp +547 -0
  85. data/ext/ggml/src/ggml-sycl/dmmv.cpp +1023 -0
  86. data/ext/ggml/src/ggml-sycl/element_wise.cpp +1030 -0
  87. data/ext/ggml/src/ggml-sycl/ggml-sycl.cpp +4729 -0
  88. data/ext/ggml/src/ggml-sycl/im2col.cpp +126 -0
  89. data/ext/ggml/src/ggml-sycl/mmq.cpp +3031 -0
  90. data/ext/ggml/src/ggml-sycl/mmvq.cpp +1015 -0
  91. data/ext/ggml/src/ggml-sycl/norm.cpp +378 -0
  92. data/ext/ggml/src/ggml-sycl/outprod.cpp +56 -0
  93. data/ext/ggml/src/ggml-sycl/rope.cpp +276 -0
  94. data/ext/ggml/src/ggml-sycl/softmax.cpp +251 -0
  95. data/ext/ggml/src/ggml-sycl/tsembd.cpp +72 -0
  96. data/ext/ggml/src/ggml-sycl/wkv6.cpp +141 -0
  97. data/ext/ggml/src/ggml-threading.cpp +12 -0
  98. data/ext/ggml/src/ggml-threading.h +14 -0
  99. data/ext/ggml/src/ggml-vulkan/ggml-vulkan.cpp +8657 -0
  100. data/ext/ggml/src/ggml-vulkan/vulkan-shaders/vulkan-shaders-gen.cpp +593 -0
  101. data/ext/ggml/src/ggml.c +7694 -0
  102. data/ext/include/whisper.h +672 -0
  103. data/ext/metal-embed.mk +17 -0
  104. data/ext/metal.mk +6 -0
  105. data/ext/ruby_whisper.cpp +1608 -159
  106. data/ext/ruby_whisper.h +10 -0
  107. data/ext/scripts/get-flags.mk +38 -0
  108. data/ext/src/coreml/whisper-decoder-impl.h +146 -0
  109. data/ext/src/coreml/whisper-decoder-impl.m +201 -0
  110. data/ext/src/coreml/whisper-encoder-impl.h +142 -0
  111. data/ext/src/coreml/whisper-encoder-impl.m +197 -0
  112. data/ext/src/coreml/whisper-encoder.h +26 -0
  113. data/ext/src/openvino/whisper-openvino-encoder.cpp +108 -0
  114. data/ext/src/openvino/whisper-openvino-encoder.h +31 -0
  115. data/ext/src/whisper.cpp +7393 -0
  116. data/extsources.rb +6 -0
  117. data/lib/whisper/model/uri.rb +157 -0
  118. data/lib/whisper.rb +2 -0
  119. data/tests/helper.rb +7 -0
  120. data/tests/jfk_reader/.gitignore +5 -0
  121. data/tests/jfk_reader/extconf.rb +3 -0
  122. data/tests/jfk_reader/jfk_reader.c +68 -0
  123. data/tests/test_callback.rb +160 -0
  124. data/tests/test_error.rb +20 -0
  125. data/tests/test_model.rb +71 -0
  126. data/tests/test_package.rb +31 -0
  127. data/tests/test_params.rb +160 -0
  128. data/tests/test_segment.rb +83 -0
  129. data/tests/test_whisper.rb +211 -123
  130. data/whispercpp.gemspec +36 -0
  131. metadata +137 -11
  132. data/ext/ggml.c +0 -8616
  133. data/ext/ggml.h +0 -748
  134. data/ext/whisper.cpp +0 -4829
  135. data/ext/whisper.h +0 -402
data/ext/ggml/src/ggml-cpu/ggml-cpu-quants.h ADDED
@@ -0,0 +1,63 @@
1
+ #pragma once
2
+
3
+ #define GGML_COMMON_DECL_C
4
+ #include "ggml-common.h"
5
+
6
+ #include "ggml.h"
7
+
8
+ // GGML CPU internal header
9
+
10
+ #ifdef __cplusplus
11
+ extern "C" {
12
+ #endif
13
+
14
+ // Quantization
15
+ void quantize_row_q4_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
16
+ void quantize_row_q4_1(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
17
+ void quantize_row_q5_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
18
+ void quantize_row_q5_1(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
19
+ void quantize_row_q8_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
20
+ void quantize_row_q8_1(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
21
+
22
+ void quantize_row_q2_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
23
+ void quantize_row_q3_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
24
+ void quantize_row_q4_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
25
+ void quantize_row_q5_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
26
+ void quantize_row_q6_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
27
+ void quantize_row_q8_K(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
28
+
29
+ void quantize_row_tq1_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
30
+ void quantize_row_tq2_0(const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
31
+
32
+ void quantize_row_iq4_nl (const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
33
+ void quantize_row_iq4_xs (const float * GGML_RESTRICT x, void * GGML_RESTRICT y, int64_t k);
34
+
35
+ // Dot product
36
+ void ggml_vec_dot_q4_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
37
+ void ggml_vec_dot_q4_1_q8_1(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
38
+ void ggml_vec_dot_q5_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
39
+ void ggml_vec_dot_q5_1_q8_1(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
40
+ void ggml_vec_dot_q8_0_q8_0(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
41
+
42
+ void ggml_vec_dot_q2_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
43
+ void ggml_vec_dot_q3_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
44
+ void ggml_vec_dot_q4_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
45
+ void ggml_vec_dot_q5_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
46
+ void ggml_vec_dot_q6_K_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
47
+
48
+ void ggml_vec_dot_tq1_0_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
49
+ void ggml_vec_dot_tq2_0_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
50
+
51
+ void ggml_vec_dot_iq2_xxs_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
52
+ void ggml_vec_dot_iq2_xs_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
53
+ void ggml_vec_dot_iq2_s_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
54
+ void ggml_vec_dot_iq3_xxs_q8_K(int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
55
+ void ggml_vec_dot_iq1_s_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
56
+ void ggml_vec_dot_iq1_m_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
57
+ void ggml_vec_dot_iq4_nl_q8_0 (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
58
+ void ggml_vec_dot_iq4_xs_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
59
+ void ggml_vec_dot_iq3_s_q8_K (int n, float * GGML_RESTRICT s, size_t bs, const void * GGML_RESTRICT vx, size_t bx, const void * GGML_RESTRICT vy, size_t by, int nrc);
60
+
61
+ #ifdef __cplusplus
62
+ }
63
+ #endif
data/ext/ggml/src/ggml-cpu/ggml-cpu-traits.cpp ADDED
@@ -0,0 +1,36 @@
1
+ #include "ggml-cpu-traits.h"
2
+
3
+ #include "ggml-backend-impl.h"
4
+ #include "ggml-backend.h"
5
+
6
+ namespace ggml::cpu {
7
+ tensor_traits::~tensor_traits() {}
8
+
9
+ extra_buffer_type::~extra_buffer_type() {}
10
+ } // namespace ggml::cpu
11
+
12
+ bool ggml_cpu_extra_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * op) {
13
+ for (auto extra : ggml_backend_cpu_get_extra_buffers_type()) {
14
+ if (extra && extra->context) {
15
+ auto buf_extra = (ggml::cpu::extra_buffer_type *) extra->context;
16
+ auto tensor_traits = buf_extra->get_tensor_traits(op);
17
+ if (tensor_traits && tensor_traits->compute_forward(params, op)) {
18
+ return true;
19
+ }
20
+ }
21
+ }
22
+ return false;
23
+ }
24
+
25
+ bool ggml_cpu_extra_work_size(int n_threads, const struct ggml_tensor * op, size_t * size) {
26
+ for (auto extra : ggml_backend_cpu_get_extra_buffers_type()) {
27
+ if (extra && extra->context) {
28
+ auto buf_extra = (ggml::cpu::extra_buffer_type *) extra->context;
29
+ auto tensor_traits = buf_extra->get_tensor_traits(op);
30
+ if (tensor_traits && tensor_traits->work_size(n_threads, op, *size)) {
31
+ return true;
32
+ }
33
+ }
34
+ }
35
+ return false;
36
+ }
data/ext/ggml/src/ggml-cpu/ggml-cpu-traits.h ADDED
@@ -0,0 +1,38 @@
1
+ #pragma once
2
+ #include "ggml-backend-impl.h"
3
+ #include "ggml-cpu-impl.h"
4
+ #include "ggml.h"
5
+
6
+ #ifdef __cplusplus
7
+ # include <vector>
8
+ extern "C" {
9
+ #endif
10
+
11
+ // return true if op part of extra "accelerator"
12
+ bool ggml_cpu_extra_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * op);
13
+ bool ggml_cpu_extra_work_size(int n_threads, const struct ggml_tensor * op, size_t * size);
14
+
15
+ #ifdef __cplusplus
16
+ }
17
+
18
+ namespace ggml::cpu {
19
+ // register in tensor->extra
20
+ class tensor_traits {
21
+ public:
22
+ virtual ~tensor_traits();
23
+ virtual bool work_size(int n_threads, const struct ggml_tensor * op, size_t & size) = 0;
24
+ virtual bool compute_forward(struct ggml_compute_params * params, struct ggml_tensor * op) = 0;
25
+ };
26
+
27
+ class extra_buffer_type {
28
+ public:
29
+ virtual ~extra_buffer_type();
30
+ virtual bool supports_op(ggml_backend_dev_t dev, const struct ggml_tensor * op) = 0;
31
+ virtual tensor_traits * get_tensor_traits(const struct ggml_tensor * op) = 0;
32
+ };
33
+ } // namespace ggml::cpu
34
+
35
+ // implemented in ggml-cpu.cpp.
36
+ std::vector<ggml_backend_buffer_type_t> & ggml_backend_cpu_get_extra_buffers_type();
37
+
38
+ #endif