vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-11 01:24:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'master' into concedo_experimental

Browse source 
# Conflicts:
#	CMakeLists.txt
#	Makefile
#	README.md
#	flake.lock
#	llama.cpp
This commit is contained in:
Concedo 2024-02-07 22:21:32 +08:00
commit ec2dbd99a3
21 changed files with 2614 additions and 1863 deletions
Download patch file Download diff file Expand all files Collapse all files

23
ggml.c
View file

@ -2343,7 +2343,7 @@ struct ggml_context * ggml_init(struct ggml_init_params params) {
#elif defined(GGML_USE_CLBLAST)
ggml_cl_init();
#elif defined(GGML_USE_VULKAN)
ggml_vk_init();
ggml_vk_init_cpu_assist();
#elif defined(GGML_USE_SYCL)
ggml_init_sycl();
#endif
@ -2470,7 +2470,8 @@ size_t ggml_get_max_tensor_size(const struct ggml_context * ctx) {
size_t max_size = 0;
for (struct ggml_tensor * tensor = ggml_get_first_tensor(ctx); tensor != NULL; tensor = ggml_get_next_tensor(ctx, tensor)) {
max_size = MAX(max_size, ggml_nbytes(tensor));
size_t bytes = ggml_nbytes(tensor);
max_size = MAX(max_size, bytes);
}
return max_size;
@ -11887,8 +11888,10 @@ GGML_CALL void ggml_rope_yarn_corr_dims(
int n_dims, int n_orig_ctx, float freq_base, float beta_fast, float beta_slow, float dims[2]
) {
// start and end correction dims
dims[0] = MAX(0, floorf(ggml_rope_yarn_corr_dim(n_dims, n_orig_ctx, beta_fast, freq_base)));
dims[1] = MIN(n_dims - 1, ceilf(ggml_rope_yarn_corr_dim(n_dims, n_orig_ctx, beta_slow, freq_base)));
float start = floorf(ggml_rope_yarn_corr_dim(n_dims, n_orig_ctx, beta_fast, freq_base));
float end = ceilf(ggml_rope_yarn_corr_dim(n_dims, n_orig_ctx, beta_slow, freq_base));
dims[0] = MAX(0, start);
dims[1] = MIN(n_dims - 1, end);
}
static void ggml_compute_forward_rope_f32(
@ -14847,10 +14850,10 @@ static void ggml_compute_forward(struct ggml_compute_params * params, struct ggm
GGML_ASSERT(tensor->src[0] == NULL || tensor->src[0]->backend == GGML_BACKEND_CPU);
GGML_ASSERT(tensor->src[1] == NULL || tensor->src[1]->backend == GGML_BACKEND_CPU);
#elif defined(GGML_USE_VULKAN)
const bool skip_cpu = ggml_vk_compute_forward(params, tensor);
const bool skip_cpu = ggml_vk_compute_forward_cpu_assist(params, tensor);
#ifdef GGML_VULKAN_CHECK_RESULTS
if (skip_cpu) {
ggml_vk_check_results_1(params, tensor);
ggml_vk_check_results_1_cpu_assist(params, tensor);
}
#endif
if (skip_cpu) {
@ -17266,12 +17269,12 @@ int ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan) {
#ifdef GGML_USE_VULKAN
for (int i = 0; i < cgraph->n_nodes; i++) {
ggml_vk_preallocate_buffers_graph(cgraph->nodes[i]);
ggml_vk_preallocate_buffers_graph_cpu_assist(cgraph->nodes[i]);
}
ggml_vk_preallocate_buffers();
ggml_vk_preallocate_buffers_cpu_assist();
for (int i = 0; i < cgraph->n_nodes; i++) {
ggml_vk_build_graph(cgraph->nodes[i], i == cgraph->n_nodes - 1);
ggml_vk_build_graph_cpu_assist(cgraph->nodes[i], i == cgraph->n_nodes - 1);
}
#endif
@ -17327,7 +17330,7 @@ int ggml_graph_compute(struct ggml_cgraph * cgraph, struct ggml_cplan * cplan) {
}
#ifdef GGML_USE_VULKAN
ggml_vk_graph_cleanup();
ggml_vk_graph_cleanup_cpu_assist();
#endif
// performance stats (graph)