Accelerator docs #2076
Accelerator docs #2076
Conversation
Signed-off-by: Ketan Umare <kumare3@users.noreply.github.com>
kumare3
requested review from
wild-endeavor,
eapolinario,
pingsutw and
cosmicBboy
as code owners
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## master #2076 +/- ##
==========================================
- Coverage 85.73% 85.52% -0.22%
==========================================
Files 313 302 -11
Lines 23435 23084 -351
Branches 3512 3512
==========================================
- Hits 20093 19743 -350
+ Misses 2734 2733 -1
Partials 608 608 ☔ View full report in Codecov by Sentry. |
Signed-off-by: Ketan Umare <kumare3@users.noreply.github.com>
Signed-off-by: Ketan Umare <kumare3@users.noreply.github.com>
Signed-off-by: Ketan Umare <kumare3@users.noreply.github.com>
| @@ -80,6 +80,7 @@ Expected output: | |||
| testing | |||
| extend | |||
| deck | |||
| extras.accelerators | |||
There was a problem hiding this comment.
It seems like this section should just be extras, for consistency with the other sections, and then "Accelerators" should be a subsection in the docs. I'm not sure why the other extras aren't showing up in the API docs (pytorch, sklearn, sqlite, tasks, tensorflow), but if they should be there, those should also be subsections under extras.
There was a problem hiding this comment.
this is because some of them are under types.extend and some under tasks.extend.
While accelerators is under extras, but it is not a type or a task extension. it is just a helper to help configure accelerators
flytekit/extras/accelerators.py
Outdated
|
|
||
| .. tags:: MachineLearning, Advanced, Hardware | ||
|
|
||
| Flyte tasks are very powerful and allow you to select `gpu` resources for your task. However, there are some cases where |
There was a problem hiding this comment.
Suggested edit for this section below. (The suggestion tool won't let me leave a suggestion for more than one line, otherwise I'd use that.)
Flyte allows you to specify `gpu` resources for a given task. However, in some cases, you may want to use a different accelerator type, such as TPU, specific variations of GPUs, or fractional GPUs. You can configure the Flyte backend to use your preferred accelerators, and those who write workflow code can import the `flytekit.extras.accelerators` module to specify an accelerator in the task decorator.
The `flytekit.extras.accelerators` module provides some constants for known accelerators, listed below, but this is not a complete list. If you know the name of the accelerator, you can pass the string name to the task.
| @@ -0,0 +1,4 @@ | |||
| .. automodule:: flytekit.extras.accelerators | |||
There was a problem hiding this comment.
Should this just be flytekit.extras?
There was a problem hiding this comment.
no accelerators is a specific module under extras
Signed-off-by: Ketan Umare <kumare3@users.noreply.github.com>
|
we can ignore the monodocs build ci error for now, I'm working on the monodocs transition and will add this once that's done. |
Signed-off-by: Ketan Umare <kumare3@users.noreply.github.com>
Tracking issue
Adds basic documentation for accelerators in flytekit
Generated docs at:
@neverett / @samhita-alla can you please help with copy-edit and layout
@jeevb / @zeryx can you please help improve content - like partition information
Nvidia docs here - https://docs.nvidia.com/datacenter/tesla/mig-user-guide/index.html#abstract