finetune : add --n-gpu-layers flag info to --help (#4128)

diff --git a/examples/finetune/finetune.cpp b/examples/finetune/finetune.cpp
index af46e44a6e216fc910d716ec033ba5f681f68066..e991e37ef39f689e2cc3973651b6a20b66c9e747 100644 (file)
--- a/examples/finetune/finetune.cpp
+++ b/examples/finetune/finetune.cpp
@@ -1288,6 +1288,7 @@ static void train_print_usage(int argc, char ** argv, const struct train_params
     fprintf(stderr, "  --model-base FNAME         model path from which to load base model (default '%s')\n", params->fn_model_base);
     fprintf(stderr, "  --lora-out FNAME           path to save llama lora (default '%s')\n", params->fn_lora_out);
     fprintf(stderr, "  --only-write-lora          only save llama lora, don't do any training.  use this if you only want to convert a checkpoint to a lora adapter.\n");
+    fprintf(stderr, "  --n-gpu-layers N           Number of model layers to offload to GPU (default 0).\n");
     fprintf(stderr, "  --norm-rms-eps F           RMS-Norm epsilon value (default %f)\n", params->f_norm_rms_eps);
     fprintf(stderr, "  --rope-freq-base F         Frequency base for ROPE (default %f)\n", params->rope_freq_base);
     fprintf(stderr, "  --rope-freq-scale F        Frequency scale for ROPE (default %f)\n", params->rope_freq_scale);