avin_regmi
avin_regmi on Hacker News
Posted by
avin_regmi on Monday May 12, 2025 @01:21AM
from the show-hn dept.
https://www.shutterly.co/
Hi HN, I recently built a web app that lets users generate AI images using models we trained and fine-tuned on our own custom datasets. One thing that sets it apart is that we run generation with more iterations than most tools. This makes our operational costs a bit higher, but it allows us to produce noticeably better quality images.The app is paid, just enough to cover infrastructure and GPU...
Posted by
avin_regmi on Wednesday October 14, 2020 @11:05PM
from the show-hn dept.
What do you guys think about www.labelur.com. Here is a blog that talks more about it https://avinregmi.medium.com/multi-domain-text-classification-via-labelur-bb4dafbf9580
That's really good to know. Why do SEOs use this? Would you be able to connect them with me?
This requires you to upload a labeled dataset. My idea was more of like you enter instances of text without label and ML model will label those texts for you.
Posted by
avin_regmi on Tuesday September 01, 2020 @09:56AM
from the ask-hn dept.
I'm working on ML model to classify NLP dataset. I'm thinking about online web app where a user would upload CSV file and possible labels. Back end would use ML to classify NLP texts into possible labels. Would you use service like this? Would you pay for it? If yes, how much?
Posted by
avin_regmi on Wednesday October 23, 2019 @10:51AM
from the cnn.com dept.
This looks really good. I would like to speak with you. Do you have an email I can contact you?
what about playing with different hyperparameters? I always found that time consuming? What do you guys think?
Thanks for the clarification.
This looks interesting but who are your targeted customers? Are you building this for developers? Also what kind of models do you support?
Posted by
avin_regmi on Wednesday April 10, 2019 @05:11PM
from the show-hn dept.
Posted by
avin_regmi on Thursday March 21, 2019 @09:49PM
from the ask-hn dept.
Uber just open sourced Ludwig. Ludwig allows you to train and test deep learning models without the need to write code. Do you think people will find this useful? Who do you think will use it?
I'll take your feedback into consideration :)
Our platform is free for the beta users to try it with limit of 2GB per model. We are just starting and we haven't decided on our business model yet.
If a user downloads panini to their private server and use it that will always be free since there is not infrastructure cost for us. If you're deploying it in our website we will be charging you to pay for the infrastracture cost.
Our main goal currently is to find out if people find this product useful and if it's worth for us to spend more time working on it. Thanks for watching the YouTube tutorial and if you have further questions, please contact us. Thanks
It really depends on the application. Such as content recommendation, prediction of popular items are requested frequently. We maintain prediction cache so we can serve the frequent cache without passing into the model. We also use cache for selecting a model. To do this we join the original prediction with the feedback it receives. Feedbacks are received soon after the prediction, even unique query can benefit from a cache. Most of the prediction models are not Deep learning these days. Most companies are using classical machine learning. In our case, we trained SVM in SciKit learn feedback throughput of 1.8x. We have a simple LRU eviction for cache and use normal cache eviction algorithm.
What are you currently using to server ML models?
Optimized TF serving would perform similarly to Panini however, it's really hard to find good documentation on optimizing TF serving compilation parameters. Panini automatically finds the right batch size to maximize the throughput and it adaptively changes. We also have a technique to reduce bound tail latency. I would love it for you to try it and provide me some feedback. Thanks
I would love it if you try and provide me some feedback.
Sorry, I should've been more clear. Both predictions for TF serving and panini serving was done in a single thread in the same specification machine. We used a simple model for image classification of CIFAR dataset. Roughly, 500 predictions were made for panini and 200 predictions for TF serving. The graph on the website is for throughput. I'm planning to write a medium post soon regarding the benchmark test.
1. Caching the input will save lots of time. Inputs are not unique each time. In a production environment, lots of inputs are the same. Many platforms in fact will do caching such as Algorithmia, TF Serving, and Sagemaker. If a time to do a search in Redis database is faster than forward pass, caching will reduce time dramatically. Watch my youtube video where I give an example.
2. It's up to you if you want to use it in GPU or CPU. Benchmark was done in a CPU but you're free to download panini via Helm and use GPU in your private kubernetes.
3. For now, during beta testing, we're offering free inference and there is a limit of model size cannot exceed over 2GB.
Hope this was helpful.
Sorry, I should've been more clear. Both predictions for TF serving and panini serving was done in a single thread in the same specification machine. We used a simple model for image classification of CIFAR dataset. Roughly, 500 predictions were made for panini and 200 predictions for TF serving. The graph on the website is for throughput. I'm planning to write a medium post soon regarding the benchmark test. There are many other projects getting higher throughput compare to TF serving. I've heard TF Serving could be optimized to make it more efficient but making it more optimized is not documented properly. We're planning to make it open source if there is enough interest from the community!
Hey, both prediction for TF serving and panini serving was done in a single thread in the same specification machine. We used a simple model for image classification of CIFAR dataset. Roughly, 500 predictions were made for panini and 200 predictions for TF serving.
You can always download the entire panini in your own private server and not pay anything. Ie. used Helm to install in your own kubernetes or DockerHub. For now, We're making it free for models under 2GB. Our main goal is to make it usable and we don't want cost to be a factor.
Hey, you don't have to deploy in GKE and it's not GKE that makes it faster. We also give you option to deploy in your own private Kubernetes via Helm or private server via DockerHub. GKE may not be the right option for you depending on your application. Your feedback would be very valuable to us. Please tell me why you think its fishy? We're always tryiing to make it better.
Posted by
avin_regmi on Monday March 11, 2019 @07:01PM
from the show-hn dept.
How big is low latency issue for you? What happens if it's more than 100ms? Also we do offer our software to be deployed in your kubernetes couster via helm.
Posted by
avin_regmi on Wednesday February 27, 2019 @12:58PM
from the ask-hn dept.
https://panini.ai/ is the easiest and fastest way to serve ML/DL models at low latency and makes the model deployment to Kubernetes in a few minutes. It also handles load balancing, caching and batching of user inputs. What are you guys using to serve ML models in low latency?
Inferring the model with Flask is slow and requires custom code for caching and batching. Scaling in multiple machines using Flask also causes many complications. To address these issues, we have developed Panini.
https://www.panini.ai
What do you guys think?
Posted by
avin_regmi on Wednesday February 27, 2019 @12:03AM
from the medium.com dept.
Posted by
avin_regmi on Sunday February 24, 2019 @10:08PM
from the medium.com dept.
Posted by
avin_regmi on Sunday February 24, 2019 @06:58PM
from the self-post dept.
I've always had a hard time deploying ML models into production. The traditional approach is to use Flask, Gunicorn with Nginx. This requires a lot of setup time, and it's slow!https://www.panini.ai/ is a platform to serve ML/DL models at low latency and makes ML model deployment to production from few days to a few minutes. Once, deployed in the cloud it will provide you with an API key to...