llama_cpp 0.2.0 → 0.2.1
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.
- checksums.yaml +4 -4
- data/CHANGELOG.md +7 -0
- data/examples/README.md +60 -0
- data/examples/chat.rb +195 -0
- data/ext/llama_cpp/llama_cpp.cpp +52 -0
- data/ext/llama_cpp/src/ggml-cuda.cu +697 -130
- data/ext/llama_cpp/src/ggml-cuda.h +4 -1
- data/ext/llama_cpp/src/ggml-metal.h +1 -0
- data/ext/llama_cpp/src/ggml-metal.m +548 -497
- data/ext/llama_cpp/src/ggml-metal.metal +425 -122
- data/ext/llama_cpp/src/ggml-opencl.cpp +3 -32
- data/ext/llama_cpp/src/ggml-opencl.h +1 -2
- data/ext/llama_cpp/src/ggml.c +1904 -303
- data/ext/llama_cpp/src/ggml.h +126 -2
- data/ext/llama_cpp/src/llama.cpp +212 -108
- data/ext/llama_cpp/src/llama.h +12 -3
- data/lib/llama_cpp/version.rb +2 -2
- data/sig/llama_cpp.rbs +3 -0
- metadata +4 -2
|
@@ -24,11 +24,14 @@ void ggml_cuda_mul_mat(const struct ggml_tensor * src0, const struct ggml_tens
|
|
|
24
24
|
void * ggml_cuda_host_malloc(size_t size);
|
|
25
25
|
void ggml_cuda_host_free(void * ptr);
|
|
26
26
|
|
|
27
|
-
void
|
|
27
|
+
void ggml_cuda_transform_tensor(void * data, struct ggml_tensor * tensor);
|
|
28
|
+
|
|
28
29
|
void ggml_cuda_free_data(struct ggml_tensor * tensor);
|
|
29
30
|
void ggml_cuda_assign_buffers(struct ggml_tensor * tensor);
|
|
31
|
+
void ggml_cuda_assign_buffers_no_scratch(struct ggml_tensor * tensor);
|
|
30
32
|
void ggml_cuda_set_main_device(int main_device);
|
|
31
33
|
void ggml_cuda_set_scratch_size(size_t scratch_size);
|
|
34
|
+
void ggml_cuda_free_scratch(void);
|
|
32
35
|
bool ggml_cuda_compute_forward(struct ggml_compute_params * params, struct ggml_tensor * tensor);
|
|
33
36
|
|
|
34
37
|
#ifdef __cplusplus
|
|
@@ -55,6 +55,7 @@ void ggml_metal_set_tensor(struct ggml_metal_context * ctx, struct ggml_tensor *
|
|
|
55
55
|
void ggml_metal_get_tensor(struct ggml_metal_context * ctx, struct ggml_tensor * t);
|
|
56
56
|
|
|
57
57
|
// same as ggml_graph_compute but uses Metal
|
|
58
|
+
// creates gf->n_threads command buffers in parallel
|
|
58
59
|
void ggml_metal_graph_compute(struct ggml_metal_context * ctx, struct ggml_cgraph * gf);
|
|
59
60
|
|
|
60
61
|
#ifdef __cplusplus
|