Ban `tensorboard==2.5.0` and `deepspeed==0.3.15` (#7159)

* ban TB 2.5

* note

* push

* Ban tb==2.5.0 and deepspeed==0.3.15

* Fix pip command

* pull

* up

* up

Co-authored-by: Carlos Mocholi <carlossmocholi@gmail.com>

azure-pipelines.yml

@@ -36,7 +36,7 @@ jobs:
     #container: "pytorchlightning/pytorch_lightning:base-cuda-py$[ variables['python.version'] ]-torch1.6"
     container:
       # base ML image: mcr.microsoft.com/azureml/openmpi3.1.2-cuda10.2-cudnn8-ubuntu18.04
-      image: "pytorchlightning/pytorch_lightning:base-cuda-py3.7-torch1.6"
+      image: "pytorchlightning/pytorch_lightning:base-cuda-py3.8-torch1.6"
       #endpoint: azureContainerRegistryConnection
       options: "--runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=all"
 

dockers/base-cuda/Dockerfile

@@ -114,7 +114,8 @@ RUN \
 
 RUN \
     # install DeepSpeed
-    pip install deepspeed>=0.3.14
+    # TODO(@SeanNaren): 0.3.15 is broken - skipping to unblock
+    pip install 'deepspeed>=0.3.14,!=0.3.15'
 
 RUN \
     # Show what we have

requirements.txt

@@ -7,6 +7,6 @@ future>=0.17.1  # required for builtins in setup.py
 PyYAML>=5.1, !=5.4.*  # OmegaConf requirement >=5.1
 tqdm>=4.41.0
 fsspec[http]>=0.8.1
-tensorboard>=2.2.0
+tensorboard>=2.2.0, !=2.5.0  # 2.5.0 GPU CI error: 'Couldn't build proto file into descriptor pool!'
 torchmetrics==0.2.0
 pyDeprecate==0.2.0