diff options
Diffstat (limited to 'examples')
| -rw-r--r-- | examples/baby-llama/baby-llama.cpp | 6 | ||||
| -rw-r--r-- | examples/common.h | 2 | ||||
| -rw-r--r-- | examples/train-text-from-scratch/train-text-from-scratch.cpp | 2 |
3 files changed, 7 insertions, 3 deletions
diff --git a/examples/baby-llama/baby-llama.cpp b/examples/baby-llama/baby-llama.cpp index f9dc0aa..6fa55b3 100644 --- a/examples/baby-llama/baby-llama.cpp +++ b/examples/baby-llama/baby-llama.cpp @@ -8,7 +8,11 @@ #pragma warning(disable: 4244 4267) // possible loss of data #endif -static const float rms_norm_eps = 1e-6f; +#ifdef LLAMA_DEFAULT_RMS_EPS +static const float rms_norm_eps = LLAMA_DEFAULT_RMS_EPS; +#else +static const float rms_norm_eps = 5e-6f; +#endif float frand() { return (float)rand()/(float)RAND_MAX; diff --git a/examples/common.h b/examples/common.h index 2d87c92..672dcf7 100644 --- a/examples/common.h +++ b/examples/common.h @@ -34,7 +34,7 @@ struct gpt_params { int32_t main_gpu = 0; // the GPU that is used for scratch and small tensors float tensor_split[LLAMA_MAX_DEVICES] = {0}; // how split tensors should be distributed across GPUs int32_t n_probs = 0; // if greater than 0, output the probabilities of top n_probs tokens. - float rms_norm_eps = 1e-6; // rms norm epsilon + float rms_norm_eps = LLAMA_DEFAULT_RMS_EPS; // rms norm epsilon float rope_freq_base = 10000.0f; // RoPE base frequency float rope_freq_scale = 1.0f; // RoPE frequency scaling factor diff --git a/examples/train-text-from-scratch/train-text-from-scratch.cpp b/examples/train-text-from-scratch/train-text-from-scratch.cpp index 4bbf6b7..54dc2be 100644 --- a/examples/train-text-from-scratch/train-text-from-scratch.cpp +++ b/examples/train-text-from-scratch/train-text-from-scratch.cpp @@ -16,7 +16,7 @@ #pragma warning(disable: 4244 4267) // possible loss of data #endif -static const float rms_norm_eps = 1e-6f; +static const float rms_norm_eps = LLAMA_DEFAULT_RMS_EPS; struct random_normal_distribution { std::mt19937 gen; |