aparler/c-llm
1
0
Fork 0
mirror of https://github.com/karpathy/llm.c.git synced 2026-07-20 23:05:08 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

Hardcode shuffle info in all train scripts

Browse source
This commit is contained in:
Aleksa Gordic 2024-06-13 21:27:42 +02:00
parent b7933ef62c
commit 1dafa60ad9
2 changed files with 4 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

4
train_gpt2.cu
View file

@ -1632,8 +1632,8 @@ int main(int argc, char *argv[]) {
// build DataLoaders for both train and val
DataLoader train_loader, val_loader;
dataloader_init(&train_loader, train_data_pattern, B, T, multi_gpu_config.process_rank, multi_gpu_config.num_processes, 1);
dataloader_init(&val_loader, val_data_pattern, B, T, multi_gpu_config.process_rank, multi_gpu_config.num_processes, 0);
dataloader_init(&train_loader, train_data_pattern, B, T, multi_gpu_config.process_rank, multi_gpu_config.num_processes, true);
dataloader_init(&val_loader, val_data_pattern, B, T, multi_gpu_config.process_rank, multi_gpu_config.num_processes, false);
// figure out the number of training steps we will run for
int train_num_batches = max_steps; // passed in from command line
if (train_num_batches == -1) {

4
train_gpt2_fp32.cu
View file

@ -1634,8 +1634,8 @@ int main(int argc, char *argv[]) {
// build DataLoaders for both train and val
DataLoader train_loader, val_loader;
dataloader_init(&train_loader, train_data_pattern, B, T, 0, 1);
dataloader_init(&val_loader, val_data_pattern, B, T, 0, 1);
dataloader_init(&train_loader, train_data_pattern, B, T, 0, 1, 1);
dataloader_init(&val_loader, val_data_pattern, B, T, 0, 1, 0);
int train_num_batches = train_loader.num_tokens / (B*T); // let's do 1 epoch by default for now
int val_num_batches = val_loader.num_tokens / (B*T);
if (val_num_batches > val_max_steps) { val_num_batches = val_max_steps; }