Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-19 17:25:07 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge commit '66001722aa0c79ef12505488eb2d15400c3e8569' into concedo_experimental

Browse Source 
# Conflicts:
#	README.md
#	docs/ops.md
#	docs/ops/SYCL.csv
#	examples/sycl/start-svr.sh
#	ggml/src/ggml-hexagon/ggml-hexagon.cpp
#	ggml/src/ggml-hexagon/htp/CMakeLists.txt
#	ggml/src/ggml-hexagon/htp/htp-ctx.h
#	ggml/src/ggml-hexagon/htp/htp-ops.h
#	ggml/src/ggml-hexagon/htp/main.c
#	ggml/src/ggml-hexagon/htp/unary-ops.c
#	ggml/src/ggml-opencl/CMakeLists.txt
#	ggml/src/ggml-opencl/ggml-opencl.cpp
#	ggml/src/ggml-opencl/kernels/cvt.cl
#	ggml/src/ggml-sycl/gated_delta_net.hpp
#	ggml/src/ggml-sycl/ggml-sycl.cpp
#	ggml/src/ggml-sycl/pad.cpp
#	ggml/src/ggml-sycl/ssm_conv.cpp
#	tests/test-backend-ops.cpp
#	tests/test-reasoning-budget.cpp
#	tools/server/README.md
#	tools/server/webui/src/lib/constants/settings-config.ts
This commit is contained in:
Concedo
2026-05-11 15:40:10 +08:00
parent bfaddd7a3b 66001722aa
commit 9b0b36b5ef
97 changed files with 8192 additions and 4853 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-context.cpp
+52 -2
View File
@@ -2461,7 +2461,30 @@ public:
for (auto & [buft, mbuf] : mbufs_new) {
auto & mbuf_cur = mbufs[buft];
if (!mbuf_cur.buf || mbuf_cur.org.size() != mbuf.org.size() || mbuf_cur.total_size != mbuf.total_size) {
bool need_alloc = false;
need_alloc = need_alloc || (!mbuf_cur.buf);
need_alloc = need_alloc || (mbuf_cur.org.size() != mbuf.org.size());
need_alloc = need_alloc || (mbuf_cur.total_size != mbuf.total_size);
if (!need_alloc) {
for (size_t i = 0; i < mbuf_cur.org.size(); ++i) {
auto * org0 = mbuf_cur.org[i];
auto * org1 = mbuf.org[i];
if (!ggml_are_same_shape(org0, org1)) {
need_alloc = true;
break;
}
if (org0->view_src != org1->view_src || org0->view_offs != org1->view_offs) {
need_alloc = true;
break;
}
}
}
if (need_alloc) {
mbuf_cur = std::move(mbuf);
mbuf_cur.buf.reset(ggml_backend_alloc_ctx_tensors_from_buft(mbuf_cur.ctx.get(), buft));
@@ -2525,6 +2548,31 @@ public:
mbufs_new[buft].total_size += rinfo.size;
}
for (auto & [buft, mbuf] : mbufs_new) {
ggml_init_params params = {
/*.mem_size =*/ mbuf.n_tensors*ggml_tensor_overhead(),
/*.mem_buffer =*/ NULL,
/*.no_alloc =*/ true,
};
mbuf.ctx.reset(ggml_init(params));
mbuf.org.reserve(mbuf.n_tensors);
}
for (const auto & rinfo : rinfos) {
auto * buft = ggml_backend_buffer_get_type(rinfo.tensor->buffer);
const int64_t n = rinfo.size/ggml_element_size(rinfo.tensor);
auto & mbuf = mbufs_new[buft];
mbuf.org.push_back(ggml_view_1d(mbuf.ctx.get(), rinfo.tensor, n, rinfo.offset));
auto & view = mbuf.org.back();
view->buffer = rinfo.tensor->buffer;
}
for (auto & [buft, mbuf] : mbufs_new) {
const auto & mbuf_cur = mbufs.at(buft);
@@ -2533,9 +2581,11 @@ public:
}
for (size_t i = 0; i < mbuf_cur.org.size(); ++i) {
ggml_backend_tensor_copy(mbuf_cur.cpy[i], mbuf_cur.org[i]);
ggml_backend_tensor_copy(mbuf_cur.cpy[i], mbuf.org[i]);
}
}
GGML_ASSERT(buf_size == 0);
}
void read(void * dst, size_t size) override {