update t5small pytorch version & doc

FlagOpen · Feb 3, 2024 · dd008fe · dd008fe
1 parent 86f4cb3
commit dd008fe
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 3 deletions.
diff --git a/training/nvidia/t5_small-pytorch/README.md b/training/nvidia/t5_small-pytorch/README.md
@@ -15,7 +15,7 @@
    - OS kernel版本: 5.4.0-113-generic
    - 加速卡驱动版本：470.129.06
    - Docker 版本：20.10.16
-   - 训练框架版本：pytorch-1.8.0a0+52ea372
+   - 训练框架版本：pytorch-1.12.0a0+bd13bc6
    - 依赖软件版本：
      - cuda: 11.4
 
@@ -46,7 +46,7 @@
 | 配置               | precision | fix_hp | e2e_time | p_whole | p_train | p_core | rouge1 | rouge2 | rougeL | rougeLsum | mem        |
 | ------------------ | --------- | ------ | -------- | ------- | ------- | ------ | ------ | ------ | ------ | --------- | ---------- |
 | A100单机8卡（1x1） | fp32      | /      | /        |         |         |        | /      | /      | /      | /         |            |
-| A100单机8卡（1x8） | fp32      | /      | 996.11   | 338     | 398     | 400    | 41.12  | 18.84  | 29.15  | 38.32     | 35.3 /40.0 |
+| A100单机8卡（1x8） | fp32      | /      | 549.5    | 659.6   | 871.8   | 875.1  | 41.09  | 18.83  | 29.14  | 38.29     | 32.5 /40.0 |
 | A100单机8卡（2x8） | fp32      | /      | /        |         |         |        | /      | /      | /      | /         |            |
 
 注意: T5模型MFU数值较低, 为11.8%

diff --git a/training/run_benchmarks/config/test_conf.py b/training/run_benchmarks/config/test_conf.py
@@ -85,7 +85,7 @@
     # "transformer:pytorch_1.13:A100:1:8:1": "/raid/dataset/transformer/wmt14_en_de_joined_dict",
     # "swin_transformer:pytorch_1.8:A100:1:8:1": "/raid/dataset/ImageNet_1k_2012/",
     # "transformer_xl:pytorch_1.8:A100:1:8:1": "/raid/dataset/transformer_xl/",
-    # "t5_small:pytorch_1.8:A100:1:8:1": "/home/datasets_ckpt/t5_small_train",
+    "t5_small:pytorch_1.12:A100:1:8:1": "/raid/dataset/t5_small_train",
     # "gpt2:pytorch_1.12:A100:1:8:1": "/raid/dataset/gpt2",
 
     # "bert_hf:pytorch_1.13:A100:1:8:1": "/raid/dataset/bert_hf_train",