16 сағат бұрын
Ай бұрын
3 ай бұрын
4 ай бұрын
4 ай бұрын
4 ай бұрын
4 ай бұрын
4 ай бұрын
Пікірлер
@vivekkarumudi 5 күн бұрын
i was super impressed with supernova medius.. it was my personal best under 14b best llm open source hands down, now its time to check the Virtuoso small.Thanks a ton for making this available to us🤩🤩😍
@juliensimonfr 5 күн бұрын
Couldn't agree more!
@nolanchen1876 5 күн бұрын
Thanks, Julien. You make deploying AI seem so easy!
@juliensimonfr 5 күн бұрын
Thanks Nolan ;)
@MatthewSajedi-l4j 6 күн бұрын
Great stuff
@nolanchen1876 6 күн бұрын
Another great video Julien! Very easy to follow.
@billykotsos4642 7 күн бұрын
HYYYYYPEEEEEEE Hopefully OpenAI tanks sooner rather than later, theyve had such a distractive impact on the AI space..
@rpitsnogle 8 күн бұрын
Nice Iron Maiden Poster...
@MonteRei-sb1co 13 күн бұрын
thank you sir
@juliensimonfr 12 күн бұрын
Most welcome
@coredog64 13 күн бұрын
Incredibly approachable video explaining not only Spectrum but why it’s useful. Looking forward to giving it a go myself.
@juliensimonfr 13 күн бұрын
Glad it was helpful!
@I_like_this_sports_tv 15 күн бұрын
This is too good!
@ramsever5087 16 күн бұрын
Thank you for sharing the video! I believe an important point that may have been overlooked in the context of model merging is the necessity for the models to remain within the same optimization basin. For instance, if the models are fine-tuned for too long and diverge significantly, weight averaging could result in a collapse in performance instead of an improvement.
@Tejramkamble-e9g 17 күн бұрын
😂😢😮
@vishvagonaduwa1668 18 күн бұрын
Would this be able to recognize letters like old English letters which are mostly used by University Certificates?
@juliensimonfr 16 күн бұрын
Textract can recognize handwriting, but your mileage will certainly vary. You have to try it out.
@MahdiSajedi1996 20 күн бұрын
Great stuff
@juliensimonfr 16 күн бұрын
Thank you!
@arnabsinha2339 21 күн бұрын
Thanks Julien another good video explaining model merging strategies. It just blew my mind when I heard Maxime Labonne talk about it at a conference. I am guessing the hyperscalars and NVDA are not hyping up this technique as there is no need for accelerated compute. :) Is this still research? Have you seen practical implementation of this? Why is SLM more hyped than merging LLMs? Thank you for responding.
@juliensimonfr 16 күн бұрын
Merging is still an active research field, but great production models are built with it, like Google Gemma2 and of course the Arcee models. Merging and SLMs are a great fit because we have so many models to choose from. LLMs are much much more expensive to build....
@FalahgsGate 21 күн бұрын
thank you so much but what is your opinion on using serverless API or Dedicated API?
@juliensimonfr 16 күн бұрын
Serverless (as in 'pay per token') is a good option if you want to keep things simple and don't have a ton of traffic. Pay per token gets expensive very quickly, which is why server-based endpoints are a better option at scale, as they give you more control over sizing, scaling, and cost optimization.
@dines1983 23 күн бұрын
This is a brilliant chat, totally enjoying it and the analogies. 🤩
@juliensimonfr 22 күн бұрын
Awesome, thank you!
@philipghuPride 24 күн бұрын
Thanks for the great video! What is the difference between mark_step used in Inferentia kzbin.info/www/bejne/haCTaJl8l5KUfpI and PyTorch 2's compile?
@juliensimonfr 16 күн бұрын
torch.compile() is the newer API. Hopefully everyone will converge to it.
@AmritPattnaik-p2c 28 күн бұрын
This is incredible Thanks a lot Julien for such informative session
@juliensimonfr 27 күн бұрын
Glad you enjoyed it!
@westwardquest 29 күн бұрын
Thanks for making this video. What do you use for this set up instead of Pylance? I'm really missing the ability to ctrl-click or mouseover a variable and go to the object definition.
@growthhackingsaas Ай бұрын
Hi, great video. Could I pay you for some advice on how I can produce an app that makes it easy for me to download the SEC filings each quarter and produce a set of financial metrics for about 20 different SaaS companies
@cuzonGames Ай бұрын
Nice 👌
@juliensimonfr Ай бұрын
Thanks 😊
@kalyanadepu5666 Ай бұрын
Amazing series
@kalyanadepu5666 Ай бұрын
can you provide the githublink
@kalyanadepu5666 Ай бұрын
great session. please provide the github link
@kalyanadepu5666 Ай бұрын
Great series, please continue
@kalyanadepu5666 Ай бұрын
can i expect end to end project in aws? i wasted 200 dollars in learning these things.
@renewed_radiance Ай бұрын
hey Is the deployment paid ?
@juliensimonfr Ай бұрын
Yes, you pay for the underlying instance. Models are free.
@EkShunya Ай бұрын
5star ⭐
@juliensimonfr Ай бұрын
Thank you!
@melikanobakhtian6018 Ай бұрын
Love these merging videos!
@juliensimonfr Ай бұрын
Thank you!
@AI_by_AI_007 Ай бұрын
Congratulations -- as a founding member of the AI community your contributions are too numerous to mention! EuroPython 2017 - Talk - 2017-07-13 - Anfiteatro 1 - Rimini, Italy :)
@juliensimonfr Ай бұрын
Yes, I remember this one. That's a while ago :)
@Create-The-Imaginable Ай бұрын
Julien! God Bless! ♥
@juliensimonfr Ай бұрын
Thank you!
@alexisgouraud2011 Ай бұрын
Bravo ! Well deserved
@juliensimonfr Ай бұрын
Thank you!
@kunalbatra4871 Ай бұрын
Congratulations !
@juliensimonfr Ай бұрын
Thank you!!
@TheDarco78 Ай бұрын
Good going man! Thank you right back
@kalyanadepu5666 Ай бұрын
Love the content
@juliensimonfr Ай бұрын
Thank you
@kalyanadepu5666 Ай бұрын
Great session
@juliensimonfr Ай бұрын
Thank you
@kalyanadepu5666 Ай бұрын
Great series
@juliensimonfr Ай бұрын
Thank you
@kalyanadepu5666 Ай бұрын
@@juliensimonfr plsease continue this series
@satishbanka-q7q Ай бұрын
That is a brilliant content
@juliensimonfr Ай бұрын
Very kind, thank you!
@shyamknayuduthokala1113 Ай бұрын
J
@Mdrashid-01 Ай бұрын
S
@reza2kn Ай бұрын
Awesome 🎉🎉🎉 Go Arcee!🔥
@juliensimonfr Ай бұрын
Thank you!
@lyubka6 Ай бұрын
awesome, thank you! Exactly what we are experimenting with
@juliensimonfr Ай бұрын
Wonderful! :)
@SudaisAlam-w1k Ай бұрын
Really Nice Julien, You can even start a comedy code channel :D
@juliensimonfr Ай бұрын
Maybe one day! :)
@vivekkarumudi Ай бұрын
i was testing the 14b supernova model , i am more than impressed , it aced almost every question i threw at it on reasoning, math , logic
@juliensimonfr Ай бұрын
Great feedback, thank you!
@QorQar Ай бұрын
Thank you for the video, but is there a complete mod to use the ideas in the video, especially for beginners?
@juliensimonfr Ай бұрын
All these techniques are implemented in open-source inference servers and models. I'd recommend reading the relevant papers, then exploring the implementation in TGI or vLLM.
@RusiruHasaranga 2 ай бұрын
i'm getting below error when creating the project in sagemaker studio.Can someone help me on this please? I added the necessary permission groups for the IAM role as well but still same error comes Your project couldn't be created Client error: Provisioning failed with error: Errors from CloudFormation: [{LogicalResourceId : SC-455545491947-pp-sqgkgh4kibchq, ResourceType : AWS::CloudFormation::Stack, StatusReason : The following resource(s) failed to create: [ModelDeployTestProject, ModelDeployCodeCommitRepository, MlOpsArtifactsBucket, ModelBuildCodeCommitRepository]. Rollback requested by user.}, {LogicalResourceId : MlOpsArtifactsBucket, ResourceType : AWS::S3::Bucket, StatusReason : Resource creation cancelled}, {LogicalResourceId : ModelDeployTestProject, ResourceType : AWS::CodeBuild::Project, StatusReason : Resource creation cancelled}, {LogicalResourceId : ModelDeployTestProject, ResourceType : AWS::CodeBuild::Project, StatusReason : Resource creation Initiated}, {LogicalResourceId : ModelBuildCodeCommitRepository, ResourceType : AWS::CodeCommit::Repository, StatusReason : CreateRepository request is not allowed because there is no existing repository in this AWS account or AWS Organization (Service: AWSCodeCommit; Status Code: 400; Error Code: OperationNotAllowedException; Request ID: 3da2f3df-50d0-4356-8ff3-518c763c1211; Proxy: null)}, {LogicalResourceId : ModelDeployCodeCommitRepository, ResourceType : AWS::CodeCommit::Repository, StatusReason : CreateRepository request is not allowed because there is no existing repository in this AWS account or AWS Organization (Service: AWSCodeCommit; Status Code: 400; Error Code: OperationNotAllowedException; Request ID: 4858c7c8-4881-4bf9-924b-edc706ff5eab; Proxy: null)}, {LogicalResourceId : SC-455545491947-pp-sqgkgh4kibchq, ResourceType : AWS::CloudFormation::Stack, StatusReason : User Initiated}].
@cathyli1167 2 ай бұрын
Hi Julie, I got the error which said "ClientError: An error occurred (ValidationException) when calling the CreateModel operation: Caller is not subscribed to the marketplace offering." Do you know how to fix it? Thanks!
@juliensimonfr 2 ай бұрын
Looks like you need to deploy a model listed on the AWS marketplace. You need to suscribe to the model first on the AWS marketplace, and then deploy it with the notebook.
@thomashuynh6263 2 ай бұрын
In my account the "INF2" option is disable. How can I enable this option.
@juliensimonfr 2 ай бұрын
What do you mean? Your inf2 quota is zero? You need to open a support ticket in the console and ask them to increase your limit for ml.inf2.xlarge on SageMaker endpoints. See docs.aws.amazon.com/general/latest/gr/aws_service_limits.html
@LiDeng-v9r 2 ай бұрын
Thank you Julien! I have always enjoyed and learned a ton from your videos! Seems that for this one, there is no slides attached as you typically do for the other videos :) ?
@juliensimonfr 2 ай бұрын
Thank you. I shared the slides at fr.slideshare.net/slideshow/deep-dive-compiling-deep-learning-models/271892112
@ushiferreyra 2 ай бұрын
It wasn't clear to me if these methodologies first do some kind of sorting by weight and connectivity similarity accross layers. I can imagine that when merging models that were fine tuned from the same base checkpoint, that we can proceed without sorting. But if we trained two models from different random initializations, we would need to sort them by similarity previously. In any case, has there been any research into this?
@juliensimonfr 2 ай бұрын
Not sure what you mean by sorting. Most methods require that merged models share the same base architecture. Frankenmerging is different and you need to pick which layers come from which model.
@ushiferreyra 2 ай бұрын
@@juliensimonfr Sorry, I meant that not only should they share the same architecture, but they should also share the same initial pretraining and weights. Two models with the same architecture but trained from scratch with different initial randomized weights would not merge very well. Unless that is, and assuming that the patterns found are the same or very similar, some analysis is done on the layers of the two models to find similar weight and connectivity distributions scattered in different parts in each, then somehow ordering for similarity before merging.
@ushiferreyra 2 ай бұрын
@@juliensimonfr The question came up because I'm currently training two models from scratch, with different initial randomized weights, with the same X data but different Y data each and I'm curious about any research done in merging these two into a single model with both Ys as multi-headed output.
@ndz7372 2 ай бұрын
Very detailed, thanks Julien!
@juliensimonfr 2 ай бұрын
Glad it was helpful!