aparler/c-llm
1
0
Fork 0
mirror of https://github.com/karpathy/llm.c.git synced 2026-07-27 20:25:09 -04:00
Code Issues Projects Releases Packages Wiki Activity Actions

fix bugs, must use size_t instead of int, this was bad. and also init cpu_losses to NULL if we want to free it later

Browse source
This commit is contained in:
Andrej Karpathy 2024-04-15 22:20:44 +00:00
parent 3c6dafeb53
commit 27ccf0b595
2 changed files with 5 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

4
train_gpt2.c
View file

@ -560,7 +560,7 @@ typedef struct {
ParameterTensors params;
size_t param_sizes[NUM_PARAMETER_TENSORS];
float* params_memory;
int num_parameters;
size_t num_parameters;
// gradients of the weights
ParameterTensors grads;
float* grads_memory;
@ -571,7 +571,7 @@ typedef struct {
ActivationTensors acts;
size_t act_sizes[NUM_ACTIVATION_TENSORS];
float* acts_memory;
int num_activations;
size_t num_activations;
// gradients of the activations
ActivationTensors grads_acts;
float* grads_acts_memory;

5
train_gpt2.cu
View file

@ -740,7 +740,7 @@ typedef struct {
ParameterTensors params;
size_t param_sizes[NUM_PARAMETER_TENSORS];
float* params_memory;
int num_parameters;
size_t num_parameters;
// gradients of the weights
ParameterTensors grads;
float* grads_memory;
@ -751,7 +751,7 @@ typedef struct {
ActivationTensors acts;
size_t act_sizes[NUM_ACTIVATION_TENSORS];
float* acts_memory;
int num_activations;
size_t num_activations;
// gradients of the activations
ActivationTensors grads_acts;
float* grads_acts_memory;
@ -833,6 +833,7 @@ void gpt2_build_from_checkpoint(GPT2 *model, const char* checkpoint_path) {
model->grads_acts_memory = NULL;
model->inputs = NULL;
model->targets = NULL;
model->cpu_losses = NULL;
model->batch_size = 0;
model->seq_len = 0;
model->mean_loss = -1.0f; // -1.0f will designate no loss