Samhita/flyte-slack-data-new · Datasets at Hugging Face

input stringlengths 1 12.6k	output stringlengths 1 12k
I'm using Pod tasks and am not seeing it! It looks like they inherit from PythonFunctionTask though so should be there?	it’s not accessible with FlyteRemote for now. Would you like to help with it?
it’s not accessible with FlyteRemote for now. Would you like to help with it?	possibly!
possibly!	Thanks!
Hi team, I create two issues and prs inside for review: 1. <https://github.com/flyteorg/flyte/issues/3567\|[Core feature] Support horovod in mpi task> cc: <@USU6W5ATA> <@U029U35LRDJ> 2. <https://github.com/flyteorg/flyte/issues/3566\|[Core feature] Add user id to security context> cc <@UNR3C6Y4T> needs your pointer about the middleware Thanks <@USU6W5ATA> for the review. Left a question <https://github.com/flyteorg/flytekit/pull/1575/files#r1156378225> <@UNR3C6Y4T> can you review the 2nd one when you get some time? thanks in advance <@USU6W5ATA> addressed the comment <https://github.com/flyteorg/flyteidl/pull/388> <https://github.com/flyteorg/flyteadmin/pull/549/files> Addressed the comment and added e2e test for the user identifier <@UNR3C6Y4T> <@U031V5ZTDE0>	thanks Byron for the context here! So if i understand this correctly: From UI, or flytekit you'll be able to add `user_identifier` as part of execution Then in admin, it will pass the identifier to workflow. Is this all that you need from <http://union.ai\|union.ai> to achieve your intended feature for linkedin? (Would there be a change in console to show the user identity as well?)
thanks Byron for the context here! So if i understand this correctly: From UI, or flytekit you'll be able to add `user_identifier` as part of execution Then in admin, it will pass the identifier to workflow. Is this all that you need from <http://union.ai\|union.ai> to achieve your intended feature for linkedin? (Would there be a change in console to show the user identity as well?)	No no Quick chat? User_id will be parsed from oauth access token
Hey y'all. Would love feedback on <https://github.com/flyteorg/flytekit/pull/1565>.	Sorry for the delay - cc <@UNR3C6Y4T> / <@USU6W5ATA>
Sorry for the delay - cc <@UNR3C6Y4T> / <@USU6W5ATA>	thanks, will take a look today cc <@U04664Z7H37> who was looking into integrating pydantic as well.
thanks, will take a look today cc <@U04664Z7H37> who was looking into integrating pydantic as well.	<@U04664Z7H37> I think that's a great idea. Let me know if you want help with that.
<@U04664Z7H37> I think that's a great idea. Let me know if you want help with that.	will take a look today
Thanks everyone, that was a fantastic meeting. This is quickly becoming my favorite meeting of the week! Recording will be posted in the YT channel. There were a couple of housekeeping items that we didn't have the chance to discuss there: 1. Branches vs forks. For contributing to Flyte, do we want to require PRs to come from forks exclusively? or keep the current approach of creating branches in the repo. <@U0265RTUJ5B> <@UNZB4NW3S>. The decision could have implications on the privileges and requirements for new Contributors 2. For the CODEOWNERS idea from <@U04664Z7H37>. Is it the goal of it to auto assign TSC members (at least) as reviewers for new RFCs?	1. I think right now we allow Flyte maintainers (or contributors?) to open branches in the Flyte repos. The general idea is that users must fork for the first few contributions and then they'll be added to the requisite permissions to allow branching directly. Not very opinionated on this moving forward, but just some context.
1. I think right now we allow Flyte maintainers (or contributors?) to open branches in the Flyte repos. The general idea is that users must fork for the first few contributions and then they'll be added to the requisite permissions to allow branching directly. Not very opinionated on this moving forward, but just some context.	1. +1 to what Dan said. Maintainers have permissions to open branches in the repo. For all other contributions we follow the usual OSS process of forking the repo. 2. This is reasonable. This will solve the problem of awareness for all new RFCs.
1. +1 to what Dan said. Maintainers have permissions to open branches in the repo. For all other contributions we follow the usual OSS process of forking the repo. 2. This is reasonable. This will solve the problem of awareness for all new RFCs.	1. I’m personally also just fine with having a fork and creating PRs from there, this is what I’ve done so far. However you prefer. 2. Yes, my idea was automatically becoming aware of RFCs to review
1. I’m personally also just fine with having a fork and creating PRs from there, this is what I’ve done so far. However you prefer. 2. Yes, my idea was automatically becoming aware of RFCs to review	thanks everyone so regarding #1, and to comply with the RFC process, I'll submit a short RFC to change the current privileges of the Contributor role (which include branching) to make it forks only. Branching would remain a privilege for maintainers. There are concerns, though, that this movement could elevate the barriers for someone to become a contributor. I'll let elaborate <@UNR3C6Y4T> better on this as a comment in the PR to come
<@U04H6UUE78B> I would also love if <@U02B12QHY9J> can join the contributor discussions and Also <@UP4D9EY6T> (and hopefully other folks from spotify)	Sorry, not able to join. It was 3am in my timezone :sweat_smile:
<@U019PBV483E> targeted jupyter group please !!!!!!!!! cc <@USU6W5ATA> and I have been looking into making tasks fully pickleable	I'm interested in never having to leave my Jupyter notebook so like never have to look at console tbh (even though console is very nice) This is current UX, but I want that syncing execution log to be an ipywidget that shows graph view (this runs and returns output to user) But task development inside notebook would also be super dope!
I'm interested in never having to leave my Jupyter notebook so like never have to look at console tbh (even though console is very nice) This is current UX, but I want that syncing execution log to be an ipywidget that shows graph view (this runs and returns output to user) But task development inside notebook would also be super dope!	add your idea here :slightly_smiling_face: : <https://github.com/flyteorg/flyte/discussions/3515>
This message was deleted.	cc <@U0265RTUJ5B> and <@U04H6UUE78B> do you think this is something we should discuss in the normal contributor sync (one happening tomorrow I believe) or should we schedule something separate just for this
cc <@U0265RTUJ5B> and <@U04H6UUE78B> do you think this is something we should discuss in the normal contributor sync (one happening tomorrow I believe) or should we schedule something separate just for this	no mostly linkedin specific thing imo it takes 1 hour
no mostly linkedin specific thing imo it takes 1 hour	will let <@U0265RTUJ5B> coordinate… early next week would be good for me
will let <@U0265RTUJ5B> coordinate… early next week would be good for me	thank you <@UNR3C6Y4T> <@U042Z2S8268> feel free to bring to the meetup tomorrow (or post here) the initiatives that could be shared in public. I think the project and the community benefits from public discussions
thank you <@UNR3C6Y4T> <@U042Z2S8268> feel free to bring to the meetup tomorrow (or post here) the initiatives that could be shared in public. I think the project and the community benefits from public discussions	oh wait i just realize i put in the wrong channel…
<@UNZB4NW3S> What's the status of this proposal? <https://github.com/flyteorg/flyte/pull/3320> Would you be able to discuss it at tomorrow's Contributors meetup? We can do async too	oops this reminds me about the RFC. I will submit my config override RFC today
oops this reminds me about the RFC. I will submit my config override RFC today	I can wait actually tomorrow is hard for me
I can wait actually tomorrow is hard for me	np, do you think is ready for TSC reviews?
np, do you think is ready for TSC reviews?	I think so i will join in for the first 15-20 minutes if it is ok, to share the RFC, or i will find a replacement
I think so i will join in for the first 15-20 minutes if it is ok, to share the RFC, or i will find a replacement	Great! Thanks
<@U04H6UUE78B> Can you explain the process of rfc review again?	David has a <https://github.com/flyteorg/flyte/pull/3460\|pr> for rfc review. Feel free to leave any comment.
hi <@U01DYLVUNJE> have anyone picked on this task? or any plan regarding this issue <https://github.com/flyteorg/flyte/issues/3094>	It’s done. the pr has been merged. <https://github.com/flyteorg/flytectl/releases/tag/v0.6.31>
It’s done. the pr has been merged. <https://github.com/flyteorg/flytectl/releases/tag/v0.6.31>	oh awesome, thanks
oh awesome, thanks	hey <@U045124RRFX> thank you for your interest on contributing to Flyte! Would you mind taking a look at the `help wanted` issues to see if there's something you could help with? <https://github.com/flyteorg/flyte/labels/help%20wanted>
hey <@U045124RRFX> thank you for your interest on contributing to Flyte! Would you mind taking a look at the `help wanted` issues to see if there's something you could help with? <https://github.com/flyteorg/flyte/labels/help%20wanted>	Hi David, I have this two issues that I am working on, I will take a look once I am done with those <https://github.com/flyteorg/flyte/issues/3533> <https://github.com/flyteorg/flyte/issues/3308>
Hi David, I have this two issues that I am working on, I will take a look once I am done with those <https://github.com/flyteorg/flyte/issues/3533> <https://github.com/flyteorg/flyte/issues/3308>	<@U045124RRFX> mind if I assign those two issues to you?
<@U045124RRFX> mind if I assign those two issues to you?	NP, please go ahead <@USU6W5ATA> one question, how do we change the logging level for flytectl? Trying to debug an issue with the flytectl itself
NP, please go ahead <@USU6W5ATA> one question, how do we change the logging level for flytectl? Trying to debug an issue with the flytectl itself	flytectl [command] --logger.level 5
flytectl [command] --logger.level 5	thank you <@USU6W5ATA> I see the gap here. I think only structure dataset is supported in your PR. Not the union type. I am having two errors because of this: 1. `flytectl get execution` error out with `{"json":{"src":"main.go:13"},"level":"error","msg":"unsupported literal scalar type *core.Scalar_Union","ts":"2023-03-27T14:20:51-07:00"}` 2. on the UI, a json with union type showed up as `This type is not yet supported` Do you want me to create a new issue regarding this one? <https://github.com/flyteorg/flyteidl/blob/master/clients/go/coreutils/extract_literal.go#L63> here is the code related
thank you <@USU6W5ATA> I see the gap here. I think only structure dataset is supported in your PR. Not the union type. I am having two errors because of this: 1. `flytectl get execution` error out with `{"json":{"src":"main.go:13"},"level":"error","msg":"unsupported literal scalar type *core.Scalar_Union","ts":"2023-03-27T14:20:51-07:00"}` 2. on the UI, a json with union type showed up as `This type is not yet supported` Do you want me to create a new issue regarding this one? <https://github.com/flyteorg/flyteidl/blob/master/clients/go/coreutils/extract_literal.go#L63> here is the code related	ohh, I see. we don’t have pr for union type. would you like to work in it?
ohh, I see. we don’t have pr for union type. would you like to work in it?	I can work on it, but I might need some mentoring on this if I hit error can I ask you for help in that case?
I can work on it, but I might need some mentoring on this if I hit error can I ask you for help in that case?	of course Thanks for the help.
Hi everyone <@U04664Z7H37> <@U03C1MJQ892> <@U01DYLVUNJE> can I have your reviews on the updates to the RFC process? <https://github.com/flyteorg/flyte/pull/3460> Thanks!	I will review tomorrow <@U04H6UUE78B> :slightly_smiling_face:
I will review tomorrow <@U04H6UUE78B> :slightly_smiling_face:	Just reviewed :+1:
OK to move the Contributor's meeting 2 hours earlier? (Every other Thursday, 11AM PT)	<@U04H6UUE78B> Will tomorrow’s meeting be two hours earlier?
<@U04H6UUE78B> Will tomorrow’s meeting be two hours earlier?	Yes <@U03C1MJQ892>! Was about to set up the reminder!
Yes <@U03C1MJQ892>! Was about to set up the reminder!	Thank you :bow:
<@U01DYLVUNJE> i think we can open tickets to improve flytepropeller and flyteadmin’s error messages! Sometimes we find them not very verbose	please do! I just created a new `improve-error-message` tag, so you can either create issues with this tag directly, or add a <https://github.com/flyteorg/flyte/discussions/3502\|comment to the discussion> depending on your preference.
please do! I just created a new `improve-error-message` tag, so you can either create issues with this tag directly, or add a <https://github.com/flyteorg/flyte/discussions/3502\|comment to the discussion> depending on your preference.	cc <@U042TPX3G06>
Thanks everyone for joining, that was a fantastic meeting! Action items: • <@U042Z2S8268> to push the <https://docs.google.com/document/d/1NS93TghOzwKamihQMDATd_jdYJtrtWQ1sViSTSDduAA/edit#heading=h.rcvd21a5zyol\|config overrides proposal> as PR for comments/discussion • Start a survey in this channel to move the meeting 2 hours earlier (David) • Create an issue to discuss potential SIGs (David) • Create Github teams and CODEOWNERS to reflect the structure of TSC and Maintainer members (I don't seem to have permissions on the repo for this. <@U01DYLVUNJE> / <@U0265RTUJ5B> is this something we could review?) <https://hackmd.io/@davidmirror/rkqCpbK1n\|Notes> Recording will be posted soon	will do!

Where is the link for todays meeting	Cc <@U04H6UUE78B>
Cc <@U04H6UUE78B>	hi <@U042Z2S8268> thanks for your interest! First meeting will take place next Thursday March 16 <https://flyte-org.slack.com/archives/CNMKCU6FR/p1678228825460909?thread_ts=1677596817.155979&cid=CNMKCU6FR\|as announced here> In that way, it would give us a bit more time to start inking a renewed RFC process to discuss it there
hi <@U042Z2S8268> thanks for your interest! First meeting will take place next Thursday March 16 <https://flyte-org.slack.com/archives/CNMKCU6FR/p1678228825460909?thread_ts=1677596817.155979&cid=CNMKCU6FR\|as announced here> In that way, it would give us a bit more time to start inking a renewed RFC process to discuss it there	oh. my mail said today
oh. my mail said today	oh that's a bummer. I'm sorry. The way Doodle works is that it's focused on a one-time meeting so it makes it look that way in this case the idea is to agree on a day/hour to have a recurring meeting. Does the current schedule work for you <@U042Z2S8268>?
oh that's a bummer. I'm sorry. The way Doodle works is that it's focused on a one-time meeting so it makes it look that way in this case the idea is to agree on a day/hour to have a recurring meeting. Does the current schedule work for you <@U042Z2S8268>?	so it will be a weekly meeting?
so it will be a weekly meeting?	bi-weekly
bi-weekly	it works for me
it works for me	I have a feeling that we'll need to eventually have them weekly because there are several proposals to discuss and also to explore an alternating schedule where we could also meet at a EU-friendly time every 2 weeks (also thinking on TSC members like <@U04664Z7H37> and <@U03C1MJQ892>) We'll see how it progress :slightly_smiling_face:
<@UNZB4NW3S> FYI, i was looking at the device flow. It seems like fosite doesn't support it yet: <https://github.com/ory/fosite/pull/695> (seems that it was closed?) This is the higher level ticket for feature: <https://github.com/ory/hydra/issues/2416>	What is fosite
What is fosite	used by flyteadmin for auth server: <https://github.com/flyteorg/flyteadmin/blob/master/auth/authzserver/provider.go#L25>
used by flyteadmin for auth server: <https://github.com/flyteorg/flyteadmin/blob/master/auth/authzserver/provider.go#L25>	Do you use FlyteAdmins auth server Do you not have a full oauth2 implementation
Do you use FlyteAdmins auth server Do you not have a full oauth2 implementation	for auth server yes, we use FlyteAdmin's auth server we use Azure sso for authentication.. which doesn't provide auth expected by flyte?
for auth server yes, we use FlyteAdmin's auth server we use Azure sso for authentication.. which doesn't provide auth expected by flyte?	Can we talk
Can we talk	sure
FYI, I contributed two PRs to resolve tensorflow/podtemplate issues we uncovered during internal testing <https://github.com/flyteorg/flyteplugins/pull/327> <https://github.com/flyteorg/flyteplugins/pull/326>	thanks for these byron! they've both been merged and will be included in the 1.4 release!
I had a chat with <@UNR3C6Y4T> last week, and we talked about sync hour agendas to see if we can talk this week. <@UTU7ZNA56> when will you be available to talk? Let's try to do at a time that Yee can participate as well	Would 8PM UTC+2:00 work for everyone?
Would 8PM UTC+2:00 work for everyone?	that’s 11 for me here in seattle. works for me.
that’s 11 for me here in seattle. works for me.	That's 3 pm for me. Depending on the day, works for me as well
That's 3 pm for me. Depending on the day, works for me as well	could you make an invite for a day that works <@U04HLNJMXJA> ?
could you make an invite for a day that works <@U04HLNJMXJA> ?	8.30 would also be fine, if it helps
8.30 would also be fine, if it helps	Yes, I can :slightly_smiling_face: Just gonna look my schedule, and I send an invite.
Thanks for finishing the Pr and merging :rocket:	Flyteplugins too
Flyteplugins too	Yes, saw it :slightly_smiling_face: I will amend the flytesnacks docs PR with the min_replicas change and ping for review there as well.
Yes, saw it :slightly_smiling_face: I will amend the flytesnacks docs PR with the min_replicas change and ping for review there as well.	amazing work!
amazing work!	<https://github.com/flyteorg/flytesnacks/pull/987> Can you pls take a look?
<https://github.com/flyteorg/flytesnacks/pull/987> Can you pls take a look?	The changes look good! but there’s an unrelated issue with the `sphinxcontrib-yt` package in our docs :disappointed: needa take a look
Also, <@U04664Z7H37> do you folks use - <https://github.com/libffcv/ffcv>? created this - <https://github.com/flyteorg/flyte/issues/3615>	Mh no, not using it but looks interesting. <https://arxiv.org/abs/2209.13705\|This> paper compares ffcv to other libraries including squirrel which my previous company built. The authors didn’t use many features of squirrel though, otherwise would be faster. When I look at the code snippets on ffcv’s website+github, I’d say that this all should live in user code though, just needs an image with dependencies installed. Which job should Flyte take in your opinion? > For example, we could not run FFCV with a dataset hosted in an S3 bucket to perform our remote experiments. (From the comparison paper) This is a downside for a data loading library tbh. Squirrel from my previous company uses fsspec and was designed for remote loading.
Mh no, not using it but looks interesting. <https://arxiv.org/abs/2209.13705\|This> paper compares ffcv to other libraries including squirrel which my previous company built. The authors didn’t use many features of squirrel though, otherwise would be faster. When I look at the code snippets on ffcv’s website+github, I’d say that this all should live in user code though, just needs an image with dependencies installed. Which job should Flyte take in your opinion? > For example, we could not run FFCV with a dataset hosted in an S3 bucket to perform our remote experiments. (From the comparison paper) This is a downside for a data loading library tbh. Squirrel from my previous company uses fsspec and was designed for remote loading.	Interesting that it would only work with local files
<@U042Z2S8268> we are enabling torch-elastic in flytekit now	thanks this is amazing! will inform my team
• <https://github.com/flyteorg/flyte/issues/3614> • <https://github.com/flyteorg/flytekit/pull/1603> • <https://github.com/flyteorg/flyteidl/pull/394> • <https://github.com/flyteorg/flyteplugins/pull/343> • <https://github.com/flyteorg/flytesnacks/pull/987>	we have to merge starting flyteidl flytekit will be the last to merge this allows us to change things if needed cc <@U017K8AJBAN>
One other thing about which I’m interested in your opinion: `torchrun` allows the user to set `--nnodes` which could e.g. be `2` but also be `"1:2"` which means min 1 max 2. Currently this is what iour new `task_config=Elastic()` exposes as well. The kubeflow PytorchJob allows setting `minReplicas`, `maxReplicas` (which by default are both None), and `replicas` (see <https://github.com/kubeflow/training-operator/blob/master/examples/pytorch/elastic/echo/echo.yaml\|here>). In theory you could say min 2, max 4, replicas 3 (without going into how much sense this makes). If a user specifies `2:3` we currently set min to 2 and max and replicas to 3. To summarize: Should we expose `nnodes` like torchrun or `min_replicas`, `max_replicas`, and `replicas` like the pytorchjob to the user?	ohh is that a question? i like min and max isnt it the same? but more explicit?
ohh is that a question? i like min and max isnt it the same? but more explicit?	Currently we make the assumption that when user specifies `3:5`, we set `maxReplicas` but also `Replicas` to 5. In theory this doesn’t have to be the case in the pytorchjob manifest. I’ll change it to the more explicit version :+1:
Currently we make the assumption that when user specifies `3:5`, we set `maxReplicas` but also `Replicas` to 5. In theory this doesn’t have to be the case in the pytorchjob manifest. I’ll change it to the more explicit version :+1:	Aah got it
Got it, first single node multi GPU :+1: I can implement this on Friday or Saturday and then ping you for review. Or do you need it before that?	Hmm that should be ok. What I want to do is train alpaca on Flyte and have that as a demo Actually if you open a PR directly on flytekit I can hack too Else I can copy paste hack and open PR Let me do it for single machine and you can make it work for distributed
Hmm that should be ok. What I want to do is train alpaca on Flyte and have that as a demo Actually if you open a PR directly on flytekit I can hack too Else I can copy paste hack and open PR Let me do it for single machine and you can make it work for distributed	Can you give me permissions to open a PR in flytekit please? Then I’ll push there Or feel free to just copy, whichever is easier for you :slightly_smiling_face:
Can you give me permissions to open a PR in flytekit please? Then I’ll push there Or feel free to just copy, whichever is easier for you :slightly_smiling_face:	Ohh you don’t have perms Can I give you I can send it Wait you should get it in 2 minutes
Ohh you don’t have perms Can I give you I can send it Wait you should get it in 2 minutes	Ok, branch is ready to push :slightly_smiling_face: thx
Ok, branch is ready to push :slightly_smiling_face: thx	Ok you should have it
Ok you should have it	<https://github.com/flyteorg/flytekit/pull/1583> Cool thanks :slightly_smiling_face: I closed the other PR from the fork Feel free to also hack/commit on this branch
<https://github.com/flyteorg/flytekit/pull/1583> Cool thanks :slightly_smiling_face: I closed the other PR from the fork Feel free to also hack/commit on this branch	Perfect
Perfect	This is going to be awesome We currently beat ignite into launching the local process group instead of torchrun. Looking very forward to throw that logic out <@UNZB4NW3S> I pushed a few commits to the wip branch. Cleanup + docstrings. Also working on making it work in a distributed way now.
This is going to be awesome We currently beat ignite into launching the local process group instead of torchrun. Looking very forward to throw that logic out <@UNZB4NW3S> I pushed a few commits to the wip branch. Cleanup + docstrings. Also working on making it work in a distributed way now.	Yup, I pushed some Commits too If you had seen This is looking great
Yup, I pushed some Commits too If you had seen This is looking great	Saw them :+1:
Saw them :+1:	I will try to get alpaca working on it too Then we can test On a side note I also got tasks working from a jupyter notebook - That way you can train large models directly from an interactive environment
I will try to get alpaca working on it too Then we can test On a side note I also got tasks working from a jupyter notebook - That way you can train large models directly from an interactive environment	You mean write task in notebook and then just run task from there?
You mean write task in notebook and then just run task from there?	Yup No need to have it in a phythojnscript Finally you will have to copy
Yup No need to have it in a phythojnscript Finally you will have to copy	I’m not much of a notebook user ^^ But I guess for many data scientists this is a killer feature
I’m not much of a notebook user ^^ But I guess for many data scientists this is a killer feature	Ya that’s my hope Mee too
Ya that’s my hope Mee too	I have a question about how to select the plugin for the task type. I have this: ```class MultiNodePytorchElasticFunctionTask(PythonFunctionTask[Elastic]): _ELASTIC_TASK_TYPE = "torch-elastic" def __init__(self, task_config: Elastic, task_function: Callable, kwargs): super(MultiNodePytorchElasticFunctionTask, self).__init__( task_type=self._ELASTIC_TASK_TYPE, kwargs, def get_custom(...): ...``` I also added this to helm values: ``` enabled_plugins: tasks: task-plugins: enabled-plugins: - ... - pytorch default-for-task-types: - ... pytorch: pytorch torch-elastic: pytorch``` Propeller says: ```{"json":{"exec_id":"f1c678faf0fd74fad828","node":"n0","ns":"flytesnacks-development","res_ver":"23058","routine":"worker-2","tasktype":"torch-elastic","wf":"flytesnacks:development:<http://wf.wf\|wf.wf>"},"level":"warning","msg":"No plugin found for Handler-type [torch-elastic], defaulting to [container]","ts":"2023-04-08T22:26:30Z"}``` Do I need to configure this somewhere else as well? The existing pytorch plugin in flyteplugins just needs an additional if else whether to configure an <https://github.com/kubeflow/training-operator/blob/b2ee1cb380b94004798b44ca32a14de3bddc675f/pkg/apis/kubeflow.org/v1/pytorch_types.go#L90\|ElasticPolicy>.
I have a question about how to select the plugin for the task type. I have this: ```class MultiNodePytorchElasticFunctionTask(PythonFunctionTask[Elastic]): _ELASTIC_TASK_TYPE = "torch-elastic" def __init__(self, task_config: Elastic, task_function: Callable, kwargs): super(MultiNodePytorchElasticFunctionTask, self).__init__( task_type=self._ELASTIC_TASK_TYPE, kwargs, def get_custom(...): ...``` I also added this to helm values: ``` enabled_plugins: tasks: task-plugins: enabled-plugins: - ... - pytorch default-for-task-types: - ... pytorch: pytorch torch-elastic: pytorch``` Propeller says: ```{"json":{"exec_id":"f1c678faf0fd74fad828","node":"n0","ns":"flytesnacks-development","res_ver":"23058","routine":"worker-2","tasktype":"torch-elastic","wf":"flytesnacks:development:<http://wf.wf\|wf.wf>"},"level":"warning","msg":"No plugin found for Handler-type [torch-elastic], defaulting to [container]","ts":"2023-04-08T22:26:30Z"}``` Do I need to configure this somewhere else as well? The existing pytorch plugin in flyteplugins just needs an additional if else whether to configure an <https://github.com/kubeflow/training-operator/blob/b2ee1cb380b94004798b44ca32a14de3bddc675f/pkg/apis/kubeflow.org/v1/pytorch_types.go#L90\|ElasticPolicy>.	AFK i think your config looks right <@U04664Z7H37> when you get a chance check the first few log lines if you start flytepropeller i think this config looks ok <@U04664Z7H37> quick question, we should not need `standalone` / single node pytorch operator right? we should automatically adapt? what if, we add a check in TorchElasticConstructor and change the task-type `if num replicas is 1` then the plugin type is `torch-elastic-standalone` else it is `torch-elastic` and the backend config is set of `torch-elastic` to use `pytorch-operator`?
AFK i think your config looks right <@U04664Z7H37> when you get a chance check the first few log lines if you start flytepropeller i think this config looks ok <@U04664Z7H37> quick question, we should not need `standalone` / single node pytorch operator right? we should automatically adapt? what if, we add a check in TorchElasticConstructor and change the task-type `if num replicas is 1` then the plugin type is `torch-elastic-standalone` else it is `torch-elastic` and the backend config is set of `torch-elastic` to use `pytorch-operator`?	I was thinking exactly the same. I feel like this should go into the existing pytorch plugin, not new ones, since also for the kubeflow training operator vanilla torch distributed training and torch elastic training only differs by the elastic config in the pytorchjob manifest. Same k8s kind though. This stays the same for backwards compatibility of course `pip install flytekitplugins-kfpytorch` ```from flytekitplugins.kfpytorch import Pytorch @task( task_config=Pytorch(...) )``` But people could to `pip install flytekitplugins-kfpytorch[elastic]` (for the torch dependency) and then: ```from flytekitplugins.kfpytorch import ElasticPytorch @task( task_config=ElasticPytorch(nnodes=1) # single pod, no operator ) @task( task_config=ElasticPytorch(nnodes=2) # pytorch operator )``` And in flyteplugins all the pytorch code can be reused as well, just an if whether we need to set elastic config in pytorchjob. Already works: ```class PytorchElasticFunctionTask(PythonFunctionTask[Elastic]): _ELASTIC_TASK_TYPE = "pytorch" _ELASTIC_TASK_TYPE_STANDALONE = "container" def __init__(self, task_config: Elastic, task_function: Callable, **kwargs): task_type = self._ELASTIC_TASK_TYPE_STANDALONE if task_config.nnodes == 1 else self._ELASTIC_TASK_TYPE super(PytorchElasticFunctionTask, self).__init__( task_config=task_config, task_type=task_type, ... def get_custom(self, settings: SerializationSettings) -> Optional[Dict[str, Any]]: if self.task_config.nnodes == 1: """ Torch elastic distributed training is executed in a normal k8s pod so that this works without the kubeflow train operator. """ return super().get_custom(settings) else: from flytekitplugins.kfpytorch.models import PyTorchJob job = PyTorchJob(``` ```Every 2.0s: kubectl get pods -n flytesnacks-development Fabios-MacBook-Pro.local: Sun Apr 9 23:04:03 2023 NAME READY STATUS RESTARTS AGE f91014ed8990b4c79b32-n0-0-master-0 1/1 Running 0 23s f91014ed8990b4c79b32-n0-0-worker-0 1/1 Running 0 22s f91014ed8990b4c79b32-n0-0-worker-1 1/1 Running 0 16s f7e922a78842044aba46-n0-0 1/1 Running 0 7s``` Only diff between the two is `nnodes` being 1 or not. Can you pls give me perms to make a PR in idl and plugins next week? Or shell I do from fork there? I will work on the changes in plugins tomorrow. Free day in Germany :slightly_smiling_face:
I was thinking exactly the same. I feel like this should go into the existing pytorch plugin, not new ones, since also for the kubeflow training operator vanilla torch distributed training and torch elastic training only differs by the elastic config in the pytorchjob manifest. Same k8s kind though. This stays the same for backwards compatibility of course `pip install flytekitplugins-kfpytorch` ```from flytekitplugins.kfpytorch import Pytorch @task( task_config=Pytorch(...) )``` But people could to `pip install flytekitplugins-kfpytorch[elastic]` (for the torch dependency) and then: ```from flytekitplugins.kfpytorch import ElasticPytorch @task( task_config=ElasticPytorch(nnodes=1) # single pod, no operator ) @task( task_config=ElasticPytorch(nnodes=2) # pytorch operator )``` And in flyteplugins all the pytorch code can be reused as well, just an if whether we need to set elastic config in pytorchjob. Already works: ```class PytorchElasticFunctionTask(PythonFunctionTask[Elastic]): _ELASTIC_TASK_TYPE = "pytorch" _ELASTIC_TASK_TYPE_STANDALONE = "container" def __init__(self, task_config: Elastic, task_function: Callable, **kwargs): task_type = self._ELASTIC_TASK_TYPE_STANDALONE if task_config.nnodes == 1 else self._ELASTIC_TASK_TYPE super(PytorchElasticFunctionTask, self).__init__( task_config=task_config, task_type=task_type, ... def get_custom(self, settings: SerializationSettings) -> Optional[Dict[str, Any]]: if self.task_config.nnodes == 1: """ Torch elastic distributed training is executed in a normal k8s pod so that this works without the kubeflow train operator. """ return super().get_custom(settings) else: from flytekitplugins.kfpytorch.models import PyTorchJob job = PyTorchJob(``` ```Every 2.0s: kubectl get pods -n flytesnacks-development Fabios-MacBook-Pro.local: Sun Apr 9 23:04:03 2023 NAME READY STATUS RESTARTS AGE f91014ed8990b4c79b32-n0-0-master-0 1/1 Running 0 23s f91014ed8990b4c79b32-n0-0-worker-0 1/1 Running 0 22s f91014ed8990b4c79b32-n0-0-worker-1 1/1 Running 0 16s f7e922a78842044aba46-n0-0 1/1 Running 0 7s``` Only diff between the two is `nnodes` being 1 or not. Can you pls give me perms to make a PR in idl and plugins next week? Or shell I do from fork there? I will work on the changes in plugins tomorrow. Free day in Germany :slightly_smiling_face:	I can give your perms i like the idea idl and plugins permissions added also we can simply add the same plugin for different config types <@U04664Z7H37> also thought some more of ```pip install flytekitplugins-kfpytorch[elastic]``` Maybe we can simply add an import gate. if modulenot found, raise an error that torch should be installed Also <@U04664Z7H37> I have this repo created = <https://github.com/unionai-oss/stanford_alpaca/pull/1> check it out
I can give your perms i like the idea idl and plugins permissions added also we can simply add the same plugin for different config types <@U04664Z7H37> also thought some more of ```pip install flytekitplugins-kfpytorch[elastic]``` Maybe we can simply add an import gate. if modulenot found, raise an error that torch should be installed Also <@U04664Z7H37> I have this repo created = <https://github.com/unionai-oss/stanford_alpaca/pull/1> check it out	I saw you simplified the models, didn’t know this was possible, nice :+1: Update from my side: I opened a draft <https://github.com/flyteorg/flyteidl/pull/394\|PR in idl> and in <https://github.com/flyteorg/flyteplugins/pull/343\|plugins>. Built a propeller image, creating a distributed pytorchjob with elastic config works. What is not working reliably yet is the rendevouz when initiating the process group. We definitely need something similar to <https://github.com/flyteorg/flytekit/pull/1583/commits/333e6008c2b3bac2fc77a379fd2220288a2a4519\|what I added here>: ``` rdzv_endpoint=os.environ.get("PET_RDZV_ENDPOINT", f"localhost:0"),``` Here, `localhost:0` means torchrun picks a free port (see <https://github.com/pytorch/pytorch/blob/537c346117967da690c9fe719e27d08ce9d43424/torch/distributed/run.py#L100\|docs>). I’m currently working on making <https://github.com/kubeflow/training-operator/tree/master/examples/pytorch/elastic/echo\|this minimal elastic example> from the kubeflow training operator repo work: ```import os import logging from flytekit import task, workflow from flytekitplugins.kfpytorch import PyTorch, Elastic logging.basicConfig(level=<http://logging.INFO\|logging.INFO>) # To see torchrun trying to establish the rendevouz @task( task_config=Elastic( nnodes=2, nproc_per_node=2, start_method="fork", ) #task_config=PyTorch(num_workers=2) ) def train() -> str: import io import os import pprint import sys import time import torch.distributed as dist env_dict = { k: os.environ[k] for k in ( "LOCAL_RANK", "RANK", "GROUP_RANK", "WORLD_SIZE", "MASTER_ADDR", "MASTER_PORT", "TORCHELASTIC_RESTART_COUNT", "TORCHELASTIC_MAX_RESTARTS", ) } with io.StringIO() as buff: print("======================================================", file=buff) print( f"Environment variables set by the agent on PID {os.getpid()}:", file=buff ) pprint.pprint(env_dict, stream=buff) print("======================================================", file=buff) print(buff.getvalue()) sys.stdout.flush() dist.init_process_group(backend="gloo") dist.barrier() rank = dist.get_rank() print( ( f"On PID {os.getpid()}, after init process group, " f"rank={dist.get_rank()}, world_size = {dist.get_world_size()}\n" ) ) dist.destroy_process_group() return f"foo-{rank}" @workflow def wf(): train() if __name__ == "__main__": print(f"Parent {os.getpid()}") print(wf())``` Rendevouz sometimes fails, sometimes works, currently debugging why. Just as fyi where I’m at…

I'm using Pod tasks and am not seeing it! It looks like they inherit from PythonFunctionTask though so should be there?

it’s not accessible with FlyteRemote for now. Would you like to help with it?

possibly!

Thanks!

Hi team, I create two issues and prs inside for review: 1. <https://github.com/flyteorg/flyte/issues/3567|[Core feature] Support horovod in mpi task> cc: <@USU6W5ATA> <@U029U35LRDJ> 2. <https://github.com/flyteorg/flyte/issues/3566|[Core feature] Add user id to security context> cc <@UNR3C6Y4T> needs your pointer about the middleware Thanks <@USU6W5ATA> for the review. Left a question <https://github.com/flyteorg/flytekit/pull/1575/files#r1156378225> <@UNR3C6Y4T> can you review the 2nd one when you get some time? thanks in advance <@USU6W5ATA> addressed the comment <https://github.com/flyteorg/flyteidl/pull/388> <https://github.com/flyteorg/flyteadmin/pull/549/files> Addressed the comment and added e2e test for the user identifier <@UNR3C6Y4T> <@U031V5ZTDE0>

thanks Byron for the context here! So if i understand this correctly: From UI, or flytekit you'll be able to add `user_identifier` as part of execution Then in admin, it will pass the identifier to workflow. Is this all that you need from <http://union.ai|union.ai> to achieve your intended feature for linkedin? (Would there be a change in console to show the user identity as well?)

No no Quick chat? User_id will be parsed from oauth access token

Hey y'all. Would love feedback on <https://github.com/flyteorg/flytekit/pull/1565>.

Sorry for the delay - cc <@UNR3C6Y4T> / <@USU6W5ATA>

thanks, will take a look today cc <@U04664Z7H37> who was looking into integrating pydantic as well.

<@U04664Z7H37> I think that's a great idea. Let me know if you want help with that.

will take a look today

Thanks everyone, that was a fantastic meeting. This is quickly becoming my favorite meeting of the week! Recording will be posted in the YT channel. There were a couple of housekeeping items that we didn't have the chance to discuss there: 1. Branches vs forks. For contributing to Flyte, do we want to require PRs to come from forks exclusively? or keep the current approach of creating branches in the repo. <@U0265RTUJ5B> <@UNZB4NW3S>. The decision could have implications on the privileges and requirements for new Contributors 2. For the CODEOWNERS idea from <@U04664Z7H37>. Is it the goal of it to auto assign TSC members (at least) as reviewers for new RFCs?

1. I think right now we allow Flyte maintainers (or contributors?) to open branches in the Flyte repos. The general idea is that users must fork for the first few contributions and then they'll be added to the requisite permissions to allow branching directly. Not very opinionated on this moving forward, but just some context.

1. +1 to what Dan said. Maintainers have permissions to open branches in the repo. For all other contributions we follow the usual OSS process of forking the repo. 2. This is reasonable. This will solve the problem of awareness for all new RFCs.

1. I’m personally also just fine with having a fork and creating PRs from there, this is what I’ve done so far. However you prefer. 2. Yes, my idea was automatically becoming aware of RFCs to review

thanks everyone so regarding #1, and to comply with the RFC process, I'll submit a short RFC to change the current privileges of the Contributor role (which include branching) to make it forks only. Branching would remain a privilege for maintainers. There are concerns, though, that this movement could elevate the barriers for someone to become a contributor. I'll let elaborate <@UNR3C6Y4T> better on this as a comment in the PR to come

<@U04H6UUE78B> I would also love if <@U02B12QHY9J> can join the contributor discussions and Also <@UP4D9EY6T> (and hopefully other folks from spotify)

Sorry, not able to join. It was 3am in my timezone :sweat_smile:

<@U019PBV483E> targeted jupyter group please !!!!!!!!! cc <@USU6W5ATA> and I have been looking into making tasks fully pickleable

I'm interested in never having to leave my Jupyter notebook so like never have to look at console tbh (even though console is very nice) This is current UX, but I want that syncing execution log to be an ipywidget that shows graph view (this runs and returns output to user) But task development inside notebook would also be super dope!

add your idea here :slightly_smiling_face: : <https://github.com/flyteorg/flyte/discussions/3515>

This message was deleted.

cc <@U0265RTUJ5B> and <@U04H6UUE78B> do you think this is something we should discuss in the normal contributor sync (one happening tomorrow I believe) or should we schedule something separate just for this

no mostly linkedin specific thing imo it takes 1 hour

will let <@U0265RTUJ5B> coordinate… early next week would be good for me

thank you <@UNR3C6Y4T> <@U042Z2S8268> feel free to bring to the meetup tomorrow (or post here) the initiatives that could be shared in public. I think the project and the community benefits from public discussions

oh wait i just realize i put in the wrong channel…

<@UNZB4NW3S> What's the status of this proposal? <https://github.com/flyteorg/flyte/pull/3320> Would you be able to discuss it at tomorrow's Contributors meetup? We can do async too

oops this reminds me about the RFC. I will submit my config override RFC today

I can wait actually tomorrow is hard for me

np, do you think is ready for TSC reviews?

I think so i will join in for the first 15-20 minutes if it is ok, to share the RFC, or i will find a replacement

Great! Thanks

<@U04H6UUE78B> Can you explain the process of rfc review again?

David has a <https://github.com/flyteorg/flyte/pull/3460|pr> for rfc review. Feel free to leave any comment.

hi <@U01DYLVUNJE> have anyone picked on this task? or any plan regarding this issue <https://github.com/flyteorg/flyte/issues/3094>

It’s done. the pr has been merged. <https://github.com/flyteorg/flytectl/releases/tag/v0.6.31>

oh awesome, thanks

hey <@U045124RRFX> thank you for your interest on contributing to Flyte! Would you mind taking a look at the `help wanted` issues to see if there's something you could help with? <https://github.com/flyteorg/flyte/labels/help%20wanted>

Hi David, I have this two issues that I am working on, I will take a look once I am done with those <https://github.com/flyteorg/flyte/issues/3533> <https://github.com/flyteorg/flyte/issues/3308>

<@U045124RRFX> mind if I assign those two issues to you?

NP, please go ahead <@USU6W5ATA> one question, how do we change the logging level for flytectl? Trying to debug an issue with the flytectl itself

flytectl [command] --logger.level 5

thank you <@USU6W5ATA> I see the gap here. I think only structure dataset is supported in your PR. Not the union type. I am having two errors because of this: 1. `flytectl get execution` error out with `{"json":{"src":"main.go:13"},"level":"error","msg":"unsupported literal scalar type *core.Scalar_Union","ts":"2023-03-27T14:20:51-07:00"}` 2. on the UI, a json with union type showed up as `This type is not yet supported` Do you want me to create a new issue regarding this one? <https://github.com/flyteorg/flyteidl/blob/master/clients/go/coreutils/extract_literal.go#L63> here is the code related

ohh, I see. we don’t have pr for union type. would you like to work in it?

I can work on it, but I might need some mentoring on this if I hit error can I ask you for help in that case?

of course Thanks for the help.

Hi everyone <@U04664Z7H37> <@U03C1MJQ892> <@U01DYLVUNJE> can I have your reviews on the updates to the RFC process? <https://github.com/flyteorg/flyte/pull/3460> Thanks!

I will review tomorrow <@U04H6UUE78B> :slightly_smiling_face:

Just reviewed :+1:

OK to move the Contributor's meeting 2 hours earlier? (Every other Thursday, 11AM PT)

<@U04H6UUE78B> Will tomorrow’s meeting be two hours earlier?

Yes <@U03C1MJQ892>! Was about to set up the reminder!

Thank you :bow:

<@U01DYLVUNJE> i think we can open tickets to improve flytepropeller and flyteadmin’s error messages! Sometimes we find them not very verbose

please do! I just created a new `improve-error-message` tag, so you can either create issues with this tag directly, or add a <https://github.com/flyteorg/flyte/discussions/3502|comment to the discussion> depending on your preference.

cc <@U042TPX3G06>

Thanks everyone for joining, that was a fantastic meeting! *Action items:* • *<@U042Z2S8268>* to push the <https://docs.google.com/document/d/1NS93TghOzwKamihQMDATd_jdYJtrtWQ1sViSTSDduAA/edit#heading=h.rcvd21a5zyol|config overrides proposal> as PR for comments/discussion • Start a survey in this channel to move the meeting 2 hours earlier (David) • Create an issue to discuss potential SIGs (David) • Create Github teams and CODEOWNERS to reflect the structure of TSC and Maintainer members (I don't seem to have permissions on the repo for this. <@U01DYLVUNJE> / <@U0265RTUJ5B> is this something we could review?) <https://hackmd.io/@davidmirror/rkqCpbK1n|Notes> Recording will be posted soon

will do!

Where is the link for todays meeting

Cc <@U04H6UUE78B>

hi <@U042Z2S8268> thanks for your interest! First meeting will take place next Thursday March 16 <https://flyte-org.slack.com/archives/CNMKCU6FR/p1678228825460909?thread_ts=1677596817.155979&cid=CNMKCU6FR|as announced here> In that way, it would give us a bit more time to start inking a renewed RFC process to discuss it there

oh. my mail said today

oh that's a bummer. I'm sorry. The way Doodle works is that it's focused on a one-time meeting so it makes it look that way in this case the idea is to agree on a day/hour to have a recurring meeting. Does the current schedule work for you <@U042Z2S8268>?

so it will be a weekly meeting?

bi-weekly

it works for me

I have a feeling that we'll need to eventually have them weekly because there are several proposals to discuss and also to explore an alternating schedule where we could also meet at a EU-friendly time every 2 weeks (also thinking on TSC members like <@U04664Z7H37> and <@U03C1MJQ892>) We'll see how it progress :slightly_smiling_face:

<@UNZB4NW3S> FYI, i was looking at the device flow. It seems like fosite doesn't support it yet: <https://github.com/ory/fosite/pull/695> (seems that it was closed?) This is the higher level ticket for feature: <https://github.com/ory/hydra/issues/2416>

What is fosite

used by flyteadmin for auth server: <https://github.com/flyteorg/flyteadmin/blob/master/auth/authzserver/provider.go#L25>

Do you use FlyteAdmins auth server Do you not have a full oauth2 implementation

for auth server yes, we use FlyteAdmin's auth server we use Azure sso for authentication.. which doesn't provide auth expected by flyte?

Can we talk

sure

FYI, I contributed two PRs to resolve tensorflow/podtemplate issues we uncovered during internal testing <https://github.com/flyteorg/flyteplugins/pull/327> <https://github.com/flyteorg/flyteplugins/pull/326>

thanks for these byron! they've both been merged and will be included in the 1.4 release!

I had a chat with <@UNR3C6Y4T> last week, and we talked about sync hour agendas to see if we can talk this week. <@UTU7ZNA56> when will you be available to talk? Let's try to do at a time that Yee can participate as well

Would 8PM UTC+2:00 work for everyone?

that’s 11 for me here in seattle. works for me.

That's 3 pm for me. Depending on the day, works for me as well

could you make an invite for a day that works <@U04HLNJMXJA> ?

8.30 would also be fine, if it helps

Yes, I can :slightly_smiling_face: Just gonna look my schedule, and I send an invite.

Thanks for finishing the Pr and merging :rocket:

Flyteplugins too

Yes, saw it :slightly_smiling_face: I will amend the flytesnacks docs PR with the min_replicas change and ping for review there as well.

amazing work!

<https://github.com/flyteorg/flytesnacks/pull/987> Can you pls take a look?

The changes look good! but there’s an unrelated issue with the `sphinxcontrib-yt` package in our docs :disappointed: needa take a look

Also, <@U04664Z7H37> do you folks use - <https://github.com/libffcv/ffcv>? created this - <https://github.com/flyteorg/flyte/issues/3615>

Mh no, not using it but looks interesting. <https://arxiv.org/abs/2209.13705|This> paper compares ffcv to other libraries including squirrel which my previous company built. The authors didn’t use many features of squirrel though, otherwise would be faster. When I look at the code snippets on ffcv’s website+github, I’d say that this all should live in user code though, just needs an image with dependencies installed. Which job should Flyte take in your opinion? > For example, we could not run FFCV with a dataset hosted in an S3 bucket to perform our remote experiments. (From the comparison paper) This is a downside for a data loading library tbh. Squirrel from my previous company uses fsspec and was designed for remote loading.

Interesting that it would only work with local files

<@U042Z2S8268> we are enabling torch-elastic in flytekit now

thanks this is amazing! will inform my team

• <https://github.com/flyteorg/flyte/issues/3614> • <https://github.com/flyteorg/flytekit/pull/1603> • <https://github.com/flyteorg/flyteidl/pull/394> • <https://github.com/flyteorg/flyteplugins/pull/343> • <https://github.com/flyteorg/flytesnacks/pull/987>

we have to merge starting flyteidl flytekit will be the last to merge this allows us to change things if needed cc <@U017K8AJBAN>

One other thing about which I’m interested in your opinion: `torchrun` allows the user to set `--nnodes` which could e.g. be `2` but also be `"1:2"` which means min 1 max 2. Currently this is what iour new `task_config=Elastic()` exposes as well. The kubeflow PytorchJob allows setting `minReplicas`, `maxReplicas` (which by default are both None), and `replicas` (see <https://github.com/kubeflow/training-operator/blob/master/examples/pytorch/elastic/echo/echo.yaml|here>). In theory you could say min 2, max 4, replicas 3 (without going into how much sense this makes). If a user specifies `2:3` we currently set min to 2 and max and replicas to 3. To summarize: Should we expose `nnodes` like torchrun or `min_replicas`, `max_replicas`, and `replicas` like the pytorchjob to the user?

ohh is that a question? i like min and max isnt it the same? but more explicit?

Currently we make the assumption that when user specifies `3:5`, we set `maxReplicas` but also `Replicas` to 5. In theory this doesn’t have to be the case in the pytorchjob manifest. I’ll change it to the more explicit version :+1:

Aah got it

Got it, first single node multi GPU :+1: I can implement this on Friday or Saturday and then ping you for review. Or do you need it before that?

Hmm that should be ok. What I want to do is train alpaca on Flyte and have that as a demo Actually if you open a PR directly on flytekit I can hack too Else I can copy paste hack and open PR Let me do it for single machine and you can make it work for distributed

Can you give me permissions to open a PR in flytekit please? Then I’ll push there Or feel free to just copy, whichever is easier for you :slightly_smiling_face:

Ohh you don’t have perms Can I give you I can send it Wait you should get it in 2 minutes

Ok, branch is ready to push :slightly_smiling_face: thx

Ok you should have it

<https://github.com/flyteorg/flytekit/pull/1583> Cool thanks :slightly_smiling_face: I closed the other PR from the fork Feel free to also hack/commit on this branch

Perfect

This is going to be awesome We currently beat ignite into launching the local process group instead of torchrun. Looking very forward to throw that logic out <@UNZB4NW3S> I pushed a few commits to the wip branch. Cleanup + docstrings. Also working on making it work in a distributed way now.

Yup, I pushed some Commits too If you had seen This is looking great

Saw them :+1:

I will try to get alpaca working on it too Then we can test On a side note I also got tasks working from a jupyter notebook - That way you can train large models directly from an interactive environment

You mean write task in notebook and then just run task from there?

Yup No need to have it in a phythojnscript Finally you will have to copy

I’m not much of a notebook user ^^ But I guess for many data scientists this is a killer feature

Ya that’s my hope Mee too

I have a question about how to select the plugin for the task type. I have this: ```class MultiNodePytorchElasticFunctionTask(PythonFunctionTask[Elastic]): _ELASTIC_TASK_TYPE = "torch-elastic" def __init__(self, task_config: Elastic, task_function: Callable, **kwargs): super(MultiNodePytorchElasticFunctionTask, self).__init__( task_type=self._ELASTIC_TASK_TYPE, **kwargs, def get_custom(...): ...``` I also added this to helm values: ``` enabled_plugins: tasks: task-plugins: enabled-plugins: - ... - pytorch default-for-task-types: - ... pytorch: pytorch torch-elastic: pytorch``` Propeller says: ```{"json":{"exec_id":"f1c678faf0fd74fad828","node":"n0","ns":"flytesnacks-development","res_ver":"23058","routine":"worker-2","tasktype":"torch-elastic","wf":"flytesnacks:development:<http://wf.wf|wf.wf>"},"level":"warning","msg":"No plugin found for Handler-type [torch-elastic], defaulting to [container]","ts":"2023-04-08T22:26:30Z"}``` Do I need to configure this somewhere else as well? The existing pytorch plugin in flyteplugins just needs an additional if else whether to configure an <https://github.com/kubeflow/training-operator/blob/b2ee1cb380b94004798b44ca32a14de3bddc675f/pkg/apis/kubeflow.org/v1/pytorch_types.go#L90|ElasticPolicy>.

AFK i think your config looks right <@U04664Z7H37> when you get a chance check the first few log lines if you start flytepropeller i think this config looks ok <@U04664Z7H37> quick question, we should not need `standalone` / single node pytorch operator right? we should automatically adapt? what if, we add a check in TorchElasticConstructor and change the task-type `if num replicas is 1` then the plugin type is `torch-elastic-standalone` else it is `torch-elastic` and the backend config is set of `torch-elastic` to use `pytorch-operator`?

I was thinking exactly the same. I feel like this should go into the existing pytorch plugin, not new ones, since also for the kubeflow training operator vanilla torch distributed training and torch elastic training only differs by the elastic config in the pytorchjob manifest. Same k8s kind though. This stays the same for backwards compatibility of course `pip install flytekitplugins-kfpytorch` ```from flytekitplugins.kfpytorch import Pytorch @task( task_config=Pytorch(...) )``` But people could to `pip install flytekitplugins-kfpytorch[elastic]` (for the torch dependency) and then: ```from flytekitplugins.kfpytorch import ElasticPytorch @task( task_config=ElasticPytorch(nnodes=1) # single pod, no operator ) @task( task_config=ElasticPytorch(nnodes=2) # pytorch operator )``` And in flyteplugins all the pytorch code can be reused as well, just an if whether we need to set elastic config in pytorchjob. Already works: ```class PytorchElasticFunctionTask(PythonFunctionTask[Elastic]): _ELASTIC_TASK_TYPE = "pytorch" _ELASTIC_TASK_TYPE_STANDALONE = "container" def __init__(self, task_config: Elastic, task_function: Callable, **kwargs): task_type = self._ELASTIC_TASK_TYPE_STANDALONE if task_config.nnodes == 1 else self._ELASTIC_TASK_TYPE super(PytorchElasticFunctionTask, self).__init__( task_config=task_config, task_type=task_type, ... def get_custom(self, settings: SerializationSettings) -> Optional[Dict[str, Any]]: if self.task_config.nnodes == 1: """ Torch elastic distributed training is executed in a normal k8s pod so that this works without the kubeflow train operator. """ return super().get_custom(settings) else: from flytekitplugins.kfpytorch.models import PyTorchJob job = PyTorchJob(``` ```Every 2.0s: kubectl get pods -n flytesnacks-development Fabios-MacBook-Pro.local: Sun Apr 9 23:04:03 2023 NAME READY STATUS RESTARTS AGE f91014ed8990b4c79b32-n0-0-master-0 1/1 Running 0 23s f91014ed8990b4c79b32-n0-0-worker-0 1/1 Running 0 22s f91014ed8990b4c79b32-n0-0-worker-1 1/1 Running 0 16s f7e922a78842044aba46-n0-0 1/1 Running 0 7s``` Only diff between the two is `nnodes` being 1 or not. Can you pls give me perms to make a PR in idl and plugins next week? Or shell I do from fork there? I will work on the changes in plugins tomorrow. Free day in Germany :slightly_smiling_face:

I can give your perms i like the idea idl and plugins permissions added also we can simply add the same plugin for different config types <@U04664Z7H37> also thought some more of ```pip install flytekitplugins-kfpytorch[elastic]``` Maybe we can simply add an import gate. if modulenot found, raise an error that torch should be installed Also <@U04664Z7H37> I have this repo created = <https://github.com/unionai-oss/stanford_alpaca/pull/1> check it out

I saw you simplified the models, didn’t know this was possible, nice :+1: Update from my side: I opened a draft <https://github.com/flyteorg/flyteidl/pull/394|PR in idl> and in <https://github.com/flyteorg/flyteplugins/pull/343|plugins>. Built a propeller image, creating a distributed pytorchjob with elastic config works. What is not working reliably yet is the rendevouz when initiating the process group. We definitely need something similar to <https://github.com/flyteorg/flytekit/pull/1583/commits/333e6008c2b3bac2fc77a379fd2220288a2a4519|what I added here>: ``` rdzv_endpoint=os.environ.get("PET_RDZV_ENDPOINT", f"localhost:0"),``` Here, `localhost:0` means torchrun picks a free port (see <https://github.com/pytorch/pytorch/blob/537c346117967da690c9fe719e27d08ce9d43424/torch/distributed/run.py#L100|docs>). I’m currently working on making <https://github.com/kubeflow/training-operator/tree/master/examples/pytorch/elastic/echo|this minimal elastic example> from the kubeflow training operator repo work: ```import os import logging from flytekit import task, workflow from flytekitplugins.kfpytorch import PyTorch, Elastic logging.basicConfig(level=<http://logging.INFO|logging.INFO>) # To see torchrun trying to establish the rendevouz @task( task_config=Elastic( nnodes=2, nproc_per_node=2, start_method="fork", ) #task_config=PyTorch(num_workers=2) ) def train() -> str: import io import os import pprint import sys import time import torch.distributed as dist env_dict = { k: os.environ[k] for k in ( "LOCAL_RANK", "RANK", "GROUP_RANK", "WORLD_SIZE", "MASTER_ADDR", "MASTER_PORT", "TORCHELASTIC_RESTART_COUNT", "TORCHELASTIC_MAX_RESTARTS", ) } with io.StringIO() as buff: print("======================================================", file=buff) print( f"Environment variables set by the agent on PID {os.getpid()}:", file=buff ) pprint.pprint(env_dict, stream=buff) print("======================================================", file=buff) print(buff.getvalue()) sys.stdout.flush() dist.init_process_group(backend="gloo") dist.barrier() rank = dist.get_rank() print( ( f"On PID {os.getpid()}, after init process group, " f"rank={dist.get_rank()}, world_size = {dist.get_world_size()}\n" ) ) dist.destroy_process_group() return f"foo-{rank}" @workflow def wf(): train() if __name__ == "__main__": print(f"Parent {os.getpid()}") print(wf())``` Rendevouz sometimes fails, sometimes works, currently debugging why. Just as fyi where I’m at…