Skip to content

Commit e0d81fc

Browse files
zhtmikezhtmike
authored
[README] PixArt Sigma to MS 2.5.0 (mindspore-lab#927)
* update README * update README * update README
1 parent 67be511 commit e0d81fc

File tree

1 file changed

+27
-5
lines changed

1 file changed

+27
-5
lines changed

‎examples/pixart_sigma/README.md‎

Lines changed: 27 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -24,6 +24,7 @@ This repo contains Mindspore model definitions, pre-trained weights and inferenc
2424
| mindspore | ascend driver | firmware | cann toolkit/kernel |
2525
|:-----------:|:---------------:|:-------------:|:---------------------:|
2626
| 2.3.1 | 24.1.RC2 | 7.3.0.1.231 | 8.0.RC2.beta1 |
27+
| 2.5.0 | 24.1.0 | 7.5.0.3.220 | 8.0.0.beta1 |
2728

2829

2930
Python: 3.9 or later.
@@ -247,27 +248,48 @@ Generated image from the code.
247248

248249
### Training Performance
249250

250-
Experiments are tested on ascend 910* with mindspore 2.3.1 graph mode
251+
- Mindspore 2.3.1
251252

252253
| model name | cards | batch size | resolution | recompute | sink | jit level |graph compile | s/step | img/s | config |
253254
|:------------:|:-----:|:----------:|:------------:|:----------:|:---------:|:---------:|:---------:|:---------:|:-------------:|:--------------------------------------------------------------------|
254-
| PixArt-Sigma | 4 | 64 | 256x256 | ON | OFF | O1 |3~5 mins | 2.907s | 88.1 | [pixart-sigma-256x256.yaml](configs/train/pixart-sigma-256x256.yaml) |
255-
| PixArt-Sigma | 4 | 32 | 512 (multi) | ON | OFF | O1 |3~5 mins | 3.817s | 33.5 | [pixart-sigma-512-MS.yaml](configs/train/pixart-sigma-512-MS.yaml) |
256-
| PixArt-Sigma | 4 | 12 | 1024 (multi) | ON | OFF | O1 |3~5 mins | 7.042s | 6.8 | [pixart-sigma-1024-MS.yaml](configs/train/pixart-sigma-1024-MS.yaml) |
257-
| PixArt-Sigma | 4 | 1 | 2048 (multi) | ON | OFF | O1 |3~5 mins | 8.772s | 0.5 | [pixart-sigma-2K-MS.yaml](configs/train/pixart-sigma-2K-MS.yaml) |
255+
| PixArt-Sigma | 4 | 64 | 256x256 | ON | OFF | O1 |3~5 mins | 2.907 | 88.1 | [pixart-sigma-256x256.yaml](configs/train/pixart-sigma-256x256.yaml) |
256+
| PixArt-Sigma | 4 | 32 | 512 (multi) | ON | OFF | O1 |3~5 mins | 3.817 | 33.5 | [pixart-sigma-512-MS.yaml](configs/train/pixart-sigma-512-MS.yaml) |
257+
| PixArt-Sigma | 4 | 12 | 1024 (multi) | ON | OFF | O1 |3~5 mins | 7.042 | 6.8 | [pixart-sigma-1024-MS.yaml](configs/train/pixart-sigma-1024-MS.yaml) |
258+
| PixArt-Sigma | 4 | 1 | 2048 (multi) | ON | OFF | O1 |3~5 mins | 8.772 | 0.5 | [pixart-sigma-2K-MS.yaml](configs/train/pixart-sigma-2K-MS.yaml) |
259+
260+
- Mindspore 2.5.0
261+
262+
| model name | cards | batch size | resolution | recompute | sink | jit level |graph compile | s/step | img/s | config |
263+
|:------------:|:-----:|:----------:|:------------:|:----------:|:---------:|:---------:|:---------:|:---------:|:-------------:|:--------------------------------------------------------------------|
264+
| PixArt-Sigma | 4 | 64 | 256x256 | ON | OFF | O1 |3~5 mins | 2.652 | 96.5 | [pixart-sigma-256x256.yaml](configs/train/pixart-sigma-256x256.yaml) |
265+
| PixArt-Sigma | 4 | 32 | 512 (multi) | ON | OFF | O1 |3~5 mins | 3.223 | 39.7 | [pixart-sigma-512-MS.yaml](configs/train/pixart-sigma-512-MS.yaml) |
266+
| PixArt-Sigma | 4 | 12 | 1024 (multi) | ON | OFF | O1 |3~5 mins | 5.141 | 9.3 | [pixart-sigma-1024-MS.yaml](configs/train/pixart-sigma-1024-MS.yaml) |
267+
| PixArt-Sigma | 4 | 1 | 2048 (multi) | ON | OFF | O1 |3~5 mins | 5.702 | 0.7 | [pixart-sigma-2K-MS.yaml](configs/train/pixart-sigma-2K-MS.yaml) |
258268

259269
> s/step: training time measured in the number of seconds for each training step.\
260270
> imgs/s: images per second during training. imgs/s = cards * batch_size / step time
261271
262272
### Inference Performance
263273

274+
275+
- Mindspore 2.3.1
276+
264277
| model name | cards | batch size | resolution | jit level | graph compile | s/step | recipe |
265278
|:------------:|:-----:|:-----------:|:------------:|:----------:|:---------:|:---------:|:------------------------------------------------------------------------|
266279
| PixArt-Sigma | 1 | 1 | 256 x 256 | O1 | < 3 mins | 0.055 | [yaml](configs/inference/pixart-sigma-256x256.yaml) |
267280
| PixArt-Sigma | 1 | 1 | 512 x 512 | O1 | < 3 mins | 0.063 | [yaml](configs/inference/pixart-sigma-512-MS.yaml) |
268281
| PixArt-Sigma | 1 | 1 | 1024 x 1024 | O1 | < 3 mins | 0.202 | [yaml](configs/inference/pixart-sigma-1024-MS.yaml) |
269282
| PixArt-Sigma | 1 | 1 | 2048 x 2048 | O1 | < 3 mins | 1.754 | [yaml](configs/inference/pixart-sigma-2K-MS.yaml) |
270283

284+
- Mindspore 2.5.0
285+
286+
| model name | cards | batch size | resolution | jit level | graph compile | s/step | recipe |
287+
|:------------:|:-----:|:-----------:|:------------:|:----------:|:---------:|:---------:|:------------------------------------------------------------------------|
288+
| PixArt-Sigma | 1 | 1 | 256 x 256 | O1 | < 3 mins | 0.073 | [yaml](configs/inference/pixart-sigma-256x256.yaml) |
289+
| PixArt-Sigma | 1 | 1 | 512 x 512 | O1 | < 3 mins | 0.076 | [yaml](configs/inference/pixart-sigma-512-MS.yaml) |
290+
| PixArt-Sigma | 1 | 1 | 1024 x 1024 | O1 | < 3 mins | 0.167 | [yaml](configs/inference/pixart-sigma-1024-MS.yaml) |
291+
| PixArt-Sigma | 1 | 1 | 2048 x 2048 | O1 | < 3 mins | 1.130 | [yaml](configs/inference/pixart-sigma-2K-MS.yaml) |
292+
271293

272294
## LoRA Finetuning
273295
We provide the train and inference of Pixart-Alpha/Pixart-Sigma in `pixart/lora-finetuning`

0 commit comments

Comments
 (0)