cui-llama.rn 1.2.0 → 1.2.2
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.
- package/README.md +2 -0
- package/android/src/main/CMakeLists.txt +2 -2
- package/android/src/main/java/com/rnllama/LlamaContext.java +31 -9
- package/android/src/main/java/com/rnllama/RNLlama.java +39 -0
- package/android/src/newarch/java/com/rnllama/RNLlamaModule.java +5 -0
- package/android/src/oldarch/java/com/rnllama/RNLlamaModule.java +5 -0
- package/cpp/common.cpp +36 -1
- package/cpp/common.h +5 -1
- package/cpp/ggml-aarch64.c +2 -11
- package/cpp/ggml-alloc.h +1 -1
- package/cpp/ggml-backend-impl.h +151 -78
- package/cpp/{ggml-backend.c → ggml-backend.cpp} +565 -269
- package/cpp/ggml-backend.h +147 -62
- package/cpp/ggml-impl.h +15 -0
- package/cpp/ggml-metal.h +8 -9
- package/cpp/ggml-metal.m +2428 -2111
- package/cpp/ggml-quants.c +2 -2
- package/cpp/ggml-quants.h +0 -4
- package/cpp/ggml.c +799 -1121
- package/cpp/ggml.h +79 -72
- package/cpp/llama-vocab.cpp +189 -106
- package/cpp/llama-vocab.h +18 -9
- package/cpp/llama.cpp +736 -341
- package/cpp/llama.h +9 -4
- package/cpp/unicode-data.cpp +6 -4
- package/cpp/unicode-data.h +4 -4
- package/cpp/unicode.cpp +14 -7
- package/lib/commonjs/NativeRNLlama.js.map +1 -1
- package/lib/commonjs/index.js +4 -0
- package/lib/commonjs/index.js.map +1 -1
- package/lib/module/NativeRNLlama.js.map +1 -1
- package/lib/module/index.js +3 -0
- package/lib/module/index.js.map +1 -1
- package/lib/typescript/NativeRNLlama.d.ts +6 -0
- package/lib/typescript/NativeRNLlama.d.ts.map +1 -1
- package/lib/typescript/index.d.ts +2 -1
- package/lib/typescript/index.d.ts.map +1 -1
- package/package.json +1 -1
- package/src/NativeRNLlama.ts +7 -0
- package/src/index.ts +5 -0
package/cpp/ggml-quants.c
CHANGED
|
@@ -4013,7 +4013,7 @@ void lm_ggml_vec_dot_q4_0_q8_0(int n, float * restrict s, size_t bs, const void
|
|
|
4013
4013
|
svfloat32_t sumv0 = svdup_n_f32(0.0f);
|
|
4014
4014
|
svfloat32_t sumv1 = svdup_n_f32(0.0f);
|
|
4015
4015
|
|
|
4016
|
-
const int vector_length =
|
|
4016
|
+
const int vector_length = lm_ggml_cpu_get_sve_cnt()*8;
|
|
4017
4017
|
|
|
4018
4018
|
// VLA Implementation using switch case
|
|
4019
4019
|
switch (vector_length) {
|
|
@@ -5597,7 +5597,7 @@ void lm_ggml_vec_dot_q8_0_q8_0(int n, float * restrict s, size_t bs, const void
|
|
|
5597
5597
|
svfloat32_t sumv0 = svdup_n_f32(0.0f);
|
|
5598
5598
|
svfloat32_t sumv1 = svdup_n_f32(0.0f);
|
|
5599
5599
|
|
|
5600
|
-
const int vector_length =
|
|
5600
|
+
const int vector_length = lm_ggml_cpu_get_sve_cnt()*8;
|
|
5601
5601
|
|
|
5602
5602
|
//VLA Implemenation for SVE
|
|
5603
5603
|
switch (vector_length) {
|
package/cpp/ggml-quants.h
CHANGED
|
@@ -142,10 +142,6 @@ void iq2xs_free_impl(enum lm_ggml_type type);
|
|
|
142
142
|
void iq3xs_init_impl(int grid_size);
|
|
143
143
|
void iq3xs_free_impl(int grid_size);
|
|
144
144
|
|
|
145
|
-
#if defined(__ARM_FEATURE_SVE)
|
|
146
|
-
extern int lm_ggml_sve_cnt_b;
|
|
147
|
-
#endif
|
|
148
|
-
|
|
149
145
|
#ifdef __cplusplus
|
|
150
146
|
}
|
|
151
147
|
#endif
|