Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix issue #4791 alloc causes compute_size to be calculated incorrectly in train-text-from-scratch, end result core dump #5033

Merged
merged 2 commits into from
Jan 19, 2024
Merged

Fix issue #4791 alloc causes compute_size to be calculated incorrectly in train-text-from-scratch, end result core dump #5033

Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
f783c59
Fix issue with alloc causing max_compute_size to be calculated
bzuzo Jan 19, 2024
38f2e6e
remove ggml_allocr_free as suggested in issue #4791
bzuzo Jan 19, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Prev Previous commit
remove ggml_allocr_free as suggested in issue #4791
  • Loading branch information
@bzuzo
bzuzo committed Jan 19, 2024
commit 38f2e6e7c3792b587e8b5ec0d6ca3eb084000aff
14 changes: 6 additions & 8 deletions examples/train-text-from-scratch/train-text-from-scratch.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -263,7 +263,6 @@ static void init_model(struct my_llama_model * model) {
model->data.resize(size + tensor_alignment);
alloc = ggml_allocr_new(model->data.data(), model->data.size(), tensor_alignment);
alloc_model(alloc, model);
ggml_allocr_free(alloc);
}

static void randomize_model(struct my_llama_model * model, int seed, float mean, float std, float min, float max) {
Expand Down Expand Up @@ -1077,6 +1076,7 @@ int main(int argc, char ** argv) {
std::vector<uint8_t> mem_input_data;
std::vector<uint8_t> mem_compute_data;

ggml_allocr * alloc = NULL;

// context for input tensors without their data
struct ggml_init_params ctx_input_params = {
Expand All @@ -1098,9 +1098,9 @@ int main(int argc, char ** argv) {

// allocate input tensors
mem_input_data.resize(max_input_size);
ggml_allocr_t alloc_inps = ggml_allocr_new(mem_input_data.data(), mem_input_data.size(), tensor_alignment);
ggml_allocr_alloc(alloc_inps, tokens_input);
ggml_allocr_alloc(alloc_inps, target_probs);
alloc = ggml_allocr_new(mem_input_data.data(), mem_input_data.size(), tensor_alignment);
ggml_allocr_alloc(alloc, tokens_input);
ggml_allocr_alloc(alloc, target_probs);

// context for compute tensors without their data
const size_t estimated_compute_size_wo_data = (
Expand All @@ -1127,7 +1127,7 @@ int main(int argc, char ** argv) {
// find best evaluation order
for (unsigned order = 0; order < (unsigned) GGML_CGRAPH_EVAL_ORDER_COUNT; ++order) {
ctx_compute = ggml_init(ctx_compute_params);
ggml_allocr_t alloc = ggml_allocr_new_measure(tensor_alignment);
alloc = ggml_allocr_new_measure(tensor_alignment);
gf = ggml_new_graph_custom(ctx_compute, LLAMA_TRAIN_MAX_NODES, true);
gf->order = (enum ggml_cgraph_eval_order) order;
gb = ggml_new_graph_custom(ctx_compute, LLAMA_TRAIN_MAX_NODES, true);
Expand All @@ -1147,7 +1147,6 @@ int main(int argc, char ** argv) {
best_compute_size = max_compute_size;
best_order = gf->order;
}
ggml_allocr_free(alloc);
ggml_free(ctx_compute);
}
size_t max_compute_size = best_compute_size;
Expand All @@ -1160,7 +1159,7 @@ int main(int argc, char ** argv) {
// allocate compute tensors
mem_compute_data.resize(max_compute_size);
ctx_compute = ggml_init(ctx_compute_params);
ggml_allocr_t alloc = ggml_allocr_new(mem_compute_data.data(), mem_compute_data.size(), tensor_alignment);
alloc = ggml_allocr_new(mem_compute_data.data(), mem_compute_data.size(), tensor_alignment);
gf = ggml_new_graph_custom(ctx_compute, LLAMA_TRAIN_MAX_NODES, true);
gf->order = best_order;
gb = ggml_new_graph_custom(ctx_compute, LLAMA_TRAIN_MAX_NODES, true);
Expand All @@ -1175,7 +1174,6 @@ int main(int argc, char ** argv) {
params.common.use_flash,
params.common.use_checkpointing
);
ggml_allocr_free(alloc);

std::vector<llama_token> train_tokens;
std::vector<size_t> train_samples_begin;
Expand Down