adjusted error thresholds

This commit is contained in:
Erik Schultheis 2025-04-13 10:18:17 +02:00
parent 166f5e6c60
commit 42aa89184f

View file

@ -113,8 +113,6 @@ int main(int argc, char *argv[]) {
size_t V = model.config.vocab_size;
size_t Vp = model.config.padded_vocab_size;
size_t maxT = model.config.max_seq_len;
size_t L = model.config.num_layers;
size_t C = model.config.channels;
for (int i = 1; i < argc; i+=2) {
if (i + 1 >= argc) { exit(EXIT_FAILURE); } // must have arg after flag
@ -268,10 +266,10 @@ int main(int argc, char *argv[]) {
// actual errors can be hardware specific.
float grad_thresholds[NUM_PARAMETER_TENSORS] = {
5e-1f, 4e-3f, 5e-4f, 8e-3f,
1e-1f, 3.5e-2f, 2e-2f, 3e-2f,
1.5e-3f,2.5e-3f, 5e-2f, 5e-2f,
5e-2f, 1.5e-2f,1e-1f,2e-2f};
5e-1f, 4e-3f, 1e-1f, 4e-2f,
5e-2f, 3.5e-2f, 2e-2f, 3e-2f,
5e-2f, 3e-2f, 3e-2f, 3e-2f,
2e-2f, 1e-2f,1e-1f,2e-2f};
#if defined(ENABLE_FP32)
for (int i = 0; i < NUM_PARAMETER_TENSORS; i++) {