cui-llama.rn 1.1.2 → 1.1.5

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 (31) hide show
  1. package/android/src/main/CMakeLists.txt +1 -2
  2. package/android/src/main/jni.cpp +26 -21
  3. package/cpp/common.cpp +181 -1584
  4. package/cpp/common.h +131 -52
  5. package/cpp/ggml-aarch64.c +612 -0
  6. package/cpp/ggml-alloc.h +2 -2
  7. package/cpp/ggml-backend.c +33 -6
  8. package/cpp/ggml-backend.h +2 -0
  9. package/cpp/ggml-common.h +20 -0
  10. package/cpp/ggml-impl.h +36 -7
  11. package/cpp/ggml-metal.m +68 -8
  12. package/cpp/ggml-quants.c +932 -50
  13. package/cpp/ggml-quants.h +15 -0
  14. package/cpp/ggml.c +1712 -325
  15. package/cpp/ggml.h +169 -100
  16. package/cpp/llama-grammar.cpp +721 -122
  17. package/cpp/llama-grammar.h +120 -15
  18. package/cpp/llama-impl.h +132 -1
  19. package/cpp/llama-sampling.cpp +1483 -354
  20. package/cpp/llama-sampling.h +20 -48
  21. package/cpp/llama-vocab.cpp +140 -7
  22. package/cpp/llama-vocab.h +3 -2
  23. package/cpp/llama.cpp +824 -327
  24. package/cpp/llama.h +235 -256
  25. package/cpp/rn-llama.hpp +18 -14
  26. package/cpp/sampling.cpp +353 -354
  27. package/cpp/sampling.h +62 -143
  28. package/cpp/sgemm.cpp +153 -0
  29. package/package.json +1 -1
  30. package/cpp/grammar-parser.cpp +0 -539
  31. package/cpp/grammar-parser.h +0 -29
package/cpp/ggml-quants.h CHANGED
@@ -26,6 +26,9 @@ void quantize_row_q5_K_ref(const float * LM_GGML_RESTRICT x, block_q5_K * LM_GGM
26
26
  void quantize_row_q6_K_ref(const float * LM_GGML_RESTRICT x, block_q6_K * LM_GGML_RESTRICT y, int64_t k);
27
27
  void quantize_row_q8_K_ref(const float * LM_GGML_RESTRICT x, block_q8_K * LM_GGML_RESTRICT y, int64_t k);
28
28
 
29
+ void quantize_row_tq1_0_ref(const float * LM_GGML_RESTRICT x, block_tq1_0 * LM_GGML_RESTRICT y, int64_t k);
30
+ void quantize_row_tq2_0_ref(const float * LM_GGML_RESTRICT x, block_tq2_0 * LM_GGML_RESTRICT y, int64_t k);
31
+
29
32
  void quantize_row_iq3_xxs_ref(const float * LM_GGML_RESTRICT x, block_iq3_xxs * LM_GGML_RESTRICT y, int64_t k);
30
33
  void quantize_row_iq4_nl_ref (const float * LM_GGML_RESTRICT x, block_iq4_nl * LM_GGML_RESTRICT y, int64_t k);
31
34
  void quantize_row_iq4_xs_ref (const float * LM_GGML_RESTRICT x, block_iq4_xs * LM_GGML_RESTRICT y, int64_t k);
@@ -46,6 +49,9 @@ void quantize_row_q5_K(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT
46
49
  void quantize_row_q6_K(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
47
50
  void quantize_row_q8_K(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
48
51
 
52
+ void quantize_row_tq1_0(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
53
+ void quantize_row_tq2_0(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
54
+
49
55
  void quantize_row_iq3_xxs(const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
50
56
  void quantize_row_iq4_nl (const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
51
57
  void quantize_row_iq4_xs (const float * LM_GGML_RESTRICT x, void * LM_GGML_RESTRICT y, int64_t k);
@@ -67,6 +73,9 @@ void dequantize_row_q5_K(const block_q5_K * LM_GGML_RESTRICT x, float * LM_GGML_
67
73
  void dequantize_row_q6_K(const block_q6_K * LM_GGML_RESTRICT x, float * LM_GGML_RESTRICT y, int64_t k);
68
74
  void dequantize_row_q8_K(const block_q8_K * LM_GGML_RESTRICT x, float * LM_GGML_RESTRICT y, int64_t k);
69
75
 
76
+ void dequantize_row_tq1_0(const block_tq1_0 * LM_GGML_RESTRICT x, float * LM_GGML_RESTRICT y, int64_t k);
77
+ void dequantize_row_tq2_0(const block_tq2_0 * LM_GGML_RESTRICT x, float * LM_GGML_RESTRICT y, int64_t k);
78
+
70
79
  void dequantize_row_iq2_xxs(const block_iq2_xxs * LM_GGML_RESTRICT x, float * LM_GGML_RESTRICT y, int64_t k);
71
80
  void dequantize_row_iq2_xs (const block_iq2_xs * LM_GGML_RESTRICT x, float * LM_GGML_RESTRICT y, int64_t k);
72
81
  void dequantize_row_iq2_s (const block_iq2_s * LM_GGML_RESTRICT x, float * LM_GGML_RESTRICT y, int64_t k);
@@ -90,6 +99,9 @@ void lm_ggml_vec_dot_q4_K_q8_K(int n, float * LM_GGML_RESTRICT s, size_t bs, con
90
99
  void lm_ggml_vec_dot_q5_K_q8_K(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
91
100
  void lm_ggml_vec_dot_q6_K_q8_K(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
92
101
 
102
+ void lm_ggml_vec_dot_tq1_0_q8_K(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
103
+ void lm_ggml_vec_dot_tq2_0_q8_K(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
104
+
93
105
  void lm_ggml_vec_dot_iq2_xxs_q8_K(int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
94
106
  void lm_ggml_vec_dot_iq2_xs_q8_K (int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
95
107
  void lm_ggml_vec_dot_iq2_s_q8_K (int n, float * LM_GGML_RESTRICT s, size_t bs, const void * LM_GGML_RESTRICT vx, size_t bx, const void * LM_GGML_RESTRICT vy, size_t by, int nrc);
@@ -111,6 +123,9 @@ size_t quantize_iq4_nl (const float * LM_GGML_RESTRICT src, void * LM_GGML_RESTR
111
123
  size_t quantize_iq4_xs (const float * LM_GGML_RESTRICT src, void * LM_GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
112
124
  size_t quantize_iq3_s (const float * LM_GGML_RESTRICT src, void * LM_GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
113
125
 
126
+ size_t quantize_tq1_0(const float * LM_GGML_RESTRICT src, void * LM_GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
127
+ size_t quantize_tq2_0(const float * LM_GGML_RESTRICT src, void * LM_GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
128
+
114
129
  size_t quantize_q2_K(const float * LM_GGML_RESTRICT src, void * LM_GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
115
130
  size_t quantize_q3_K(const float * LM_GGML_RESTRICT src, void * LM_GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);
116
131
  size_t quantize_q4_K(const float * LM_GGML_RESTRICT src, void * LM_GGML_RESTRICT dst, int64_t nrows, int64_t n_per_row, const float * imatrix);