Skip to content

Commit ac1d799

Browse files
committed
Update tensorflow wheels documentation
1 parent a160884 commit ac1d799

File tree

2 files changed

+21
-11
lines changed

2 files changed

+21
-11
lines changed

README.md

+14-10
Original file line numberDiff line numberDiff line change
@@ -22,12 +22,12 @@ If you the first step above doesn't work for your use case, [open an issue](http
2222
## Opening a pull request
2323

2424
1. Update the *Dockerfile*
25-
1. For changes specific to the GPU image, update the [gpu.Dockerfile](gpu.Dockerfile).
26-
2. Otherwise, update the [Dockerfile](Dockerfile).
27-
2. Follow the instructions below to build a new image.
28-
3. Add tests for your new package. See this [example](https://github.com/Kaggle/docker-python/blob/master/tests/test_fastai.py).
29-
4. Follow the instructions below to test the new image.
30-
5. Open a PR on this repo and you are all set!
25+
1. For changes specific to the GPU image, update the [gpu.Dockerfile](gpu.Dockerfile).
26+
1. Otherwise, update the [Dockerfile](Dockerfile).
27+
1. Follow the instructions below to build a new image.
28+
1. Add tests for your new package. See this [example](https://github.com/Kaggle/docker-python/blob/master/tests/test_fastai.py).
29+
1. Follow the instructions below to test the new image.
30+
1. Open a PR on this repo and you are all set!
3131

3232
## Building a new image
3333

@@ -52,11 +52,15 @@ Flags:
5252

5353
* `--gpu` to test the GPU image.
5454

55-
## Tensorflow custom pre-built wheels
55+
## Tensorflow custom pre-built wheel
5656

57-
We are building Tensorflow from sources mainly for:
57+
A Tensorflow custom pre-built wheel is used mainly for:
5858

59-
* Better performance. When building from sources, we can leverage CPU specific optimizations
60-
* Tensorflow with GPU support must be built from sources
59+
* Faster build time: Building tensorflow from sources takes ~1h. Keeping this process outside the main build allows faster iterations when working on our Dockerfiles.
60+
61+
Building Tensorflow from sources:
62+
63+
* Increase performance: When building from sources, we can leverage CPU specific optimizations
64+
* Is required: Tensorflow with GPU support must be built from sources
6165

6266
The [Dockerfile](tensorflow-whl/Dockerfile) and the [instructions](tensorflow-whl/README.md) can be found in the [tensorflow-whl folder/](tensorflow/).

tensorflow-whl/README.md

+7-1
Original file line numberDiff line numberDiff line change
@@ -4,7 +4,7 @@
44
./build
55
```
66

7-
# Push the new wheels
7+
# Push the new wheels (Kaggle Engineers only)
88

99
1. Add an entry in the [CHANGELOG](CHANGELOG.md) with an appropriate `LABEL`.
1010
2. Push the new image using the `LABEL` you picked above.
@@ -17,6 +17,12 @@
1717
1818
Update the line below in the [CPU Dockerfile](../Dockerfile) and the [GPU Dockerfile](../gpu.Dockerfile) to use the new `LABEL`.
1919
20+
To use wheels built locally:
21+
```
22+
FROM kaggle/python-tensorflow-whl as tensorflow_whl
23+
```
24+
25+
To use our pre-built wheels:
2026
```
2127
FROM gcr.io/kaggle-images/python-tensorflow-whl:<LABEL> as tensorflow_whl
2228
```

0 commit comments

Comments
 (0)