nikola/llama_cpp
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity

mtmd: fix mtmd_get_memory_usage (#24867)

Browse Source
This commit is contained in:
Xuan-Son Nguyen
2026-06-21 14:12:15 +02:00
committed by GitHub
parent bf533823cd
commit 0d135df48c
3 changed files with 37 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/mtmd/clip.cpp
+33 -29
View File
@@ -2796,7 +2796,7 @@ struct clip_model_loader {
}
// load data
if (!ctx_clip.no_alloc) {
{
std::vector<uint8_t> read_buf;
// start loading event
@@ -2814,38 +2814,42 @@ struct clip_model_loader {
ggml_backend_buffer_type_t buft = ggml_backend_get_default_buffer_type(ctx_clip.backend);
ctx_clip.buf.reset(ggml_backend_alloc_ctx_tensors_from_buft(ctx_clip.ctx_data.get(), buft));
ggml_backend_buffer_set_usage(ctx_clip.buf.get(), GGML_BACKEND_BUFFER_USAGE_WEIGHTS);
size_t data_loaded = 0;
for (auto & t : tensors_to_load) {
ggml_tensor * cur = ggml_get_tensor(ctx_clip.ctx_data.get(), t->name);
GGML_ASSERT(cur && "tensor not found in ctx_data");
auto it_off = tensor_offset.find(t->name);
GGML_ASSERT(it_off != tensor_offset.end() && "no offset for tensor");
const size_t offset = it_off->second;
fin.seekg(offset, std::ios::beg);
if (!fin) {
throw std::runtime_error(string_format("%s: failed to seek for tensor %s\n", __func__, t->name));
}
size_t num_bytes = ggml_nbytes(cur);
if (ggml_backend_buft_is_host(buft)) {
// for the CPU and Metal backend, we can read directly into the tensor
fin.read(reinterpret_cast<char *>(cur->data), num_bytes);
} else {
// read into a temporary buffer first, then copy to device memory
read_buf.resize(num_bytes);
fin.read(reinterpret_cast<char *>(read_buf.data()), num_bytes);
ggml_backend_tensor_set(cur, read_buf.data(), 0, num_bytes);
}
data_loaded += num_bytes;
if (progress_callback && total_data_size > 0) {
const float progress = (float)data_loaded / (float)total_data_size;
if (!progress_callback(progress, progress_callback_user_data)) {
throw std::runtime_error(string_format("%s: model loading cancelled by progress_callback\n", __func__));
// read the weight from file
if (!ctx_clip.no_alloc) {
size_t data_loaded = 0;
for (auto & t : tensors_to_load) {
ggml_tensor * cur = ggml_get_tensor(ctx_clip.ctx_data.get(), t->name);
GGML_ASSERT(cur && "tensor not found in ctx_data");
auto it_off = tensor_offset.find(t->name);
GGML_ASSERT(it_off != tensor_offset.end() && "no offset for tensor");
const size_t offset = it_off->second;
fin.seekg(offset, std::ios::beg);
if (!fin) {
throw std::runtime_error(string_format("%s: failed to seek for tensor %s\n", __func__, t->name));
}
size_t num_bytes = ggml_nbytes(cur);
if (ggml_backend_buft_is_host(buft)) {
// for the CPU and Metal backend, we can read directly into the tensor
fin.read(reinterpret_cast<char *>(cur->data), num_bytes);
} else {
// read into a temporary buffer first, then copy to device memory
read_buf.resize(num_bytes);
fin.read(reinterpret_cast<char *>(read_buf.data()), num_bytes);
ggml_backend_tensor_set(cur, read_buf.data(), 0, num_bytes);
}
data_loaded += num_bytes;
if (progress_callback && total_data_size > 0) {
const float progress = (float)data_loaded / (float)total_data_size;
if (!progress_callback(progress, progress_callback_user_data)) {
throw std::runtime_error(string_format("%s: model loading cancelled by progress_callback\n", __func__));
}
}
}
LOG_DBG("%s: loaded %zu tensors from %s\n", __func__, tensors_to_load.size(), fname.c_str());
} else {
LOG_DBG("%s: no_alloc is set, skipping tensor data loading (%zu tensors)\n", __func__, tensors_to_load.size());
}
fin.close();
LOG_DBG("%s: loaded %zu tensors from %s\n", __func__, tensors_to_load.size(), fname.c_str());
}
}