From 307d51aa8c9f10490544dbae5f29372a1fef11b5 Mon Sep 17 00:00:00 2001 From: chenyingshu Date: Thu, 6 Feb 2025 11:17:51 +0800 Subject: [PATCH] add ms ver and mode for inference and training performance eval --- examples/openlrm/README.md | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/examples/openlrm/README.md b/examples/openlrm/README.md index 79f673b57d..a303d0ca8a 100644 --- a/examples/openlrm/README.md +++ b/examples/openlrm/README.md @@ -159,10 +159,9 @@ LRM is the first Large Reconstruction Model that predicts the 3D model of an obj ``` ## Performace -Experiments are tested on 1 ascend 910* with mindSpore 2.4.1 pynative mode. - ### Inference Performance +Experiments are tested on ascend 910* with mindspore 2.4.1 pynative mode. - Input a single image, here reports speed of image-to-3D and image rendering. | model name| cards| batch size | resolution | precision | flash attn| (image to 3D) s/step | (render) img/s| recipe| weight| @@ -181,6 +180,7 @@ Experiments are tested on 1 ascend 910* with mindSpore 2.4.1 pynative mode. ### Training Performance +Experiments are tested on ascend 910* with mindspore 2.4.1 pynative mode. - Train with objaverse data only: |model name | cards | batch size | resolution | precision| recompute | s/step | batch/s|