It seems fixed. I've tested some Math ops that were previously failing and now are consistent on Chrome 150.0.7871.128 for linux and 150.0.7871.129 for macOS.
It was a sad surprise last week when we tried to upgrade the workspace AI plan for some of our team members to Ultra and it was gone. We're moving to Claude/Codex.
I’ve been using DinoV2 for some months now. I’ve tried the models with 4 register tokens along with CLS + patch tokens. I’ve several embeddings (tokens) from previous model (no registers) which are part of my solution, so I didn’t adopt the newer “register” models because the CLS tokens are not aligned between 0 registers and 4 registers models. It would be nice if the CLS and patch tokens were somehow aligned between those models.
It's much worse than that. The leak contains dozens of datasets (relatives, addresses, jobs (+ linkedin), schools, vehicles, income, debts, pictures of faces, companies).
This leak is much more harmful. If the data really comes from Serasa Experian, they have more accurate and structured data from people/companies/assets in Brazil than anyone else.
Thanks, @rasmi. I have a feedback for you guys. The pricing for predictions inference in GCP is not very fair. If I deploy a small model (like a SqueezeNet or Mobilenet) I pay almost the same price of someone deploying large models (like Resnet or VGG). That’s why I’m deploying my models on serverless environments and paying about 5 dollars for 1 million inferences.
The pricing of GCP is: $0.10 per thousand predictions, plus $0.40 per hour. That’s more than 100 dollars for 1 million inferences.
Thanks for the answer. Currently I’m using AWS Lambda to deploy my TensorFlow models. But it’s pretty hard and hacky. I need to remove a considerable portion of the code base that is not needed for inference only routines. I do that so the code loads faster and to fit the deployment package size limit.
If TensorFlow Lite is already a compressed code, then it may be much easier to deploy it to a serverless environment.
I’ll be trying it in my next deployments.
Automating searches on a government website that decided to use reCAPTCHA just because it wants to look modern. There are dozens of them in Brazil for example.
Nice work. It also publishes a file mfcc.py which uses Mel spectrogram to solve the audio offline. With enough data, a model based on MFCC should work much better than any cloud service (general speech recognizer).
Segmentation is not needed for a modern deep learning system, since this is learned by the neural network. It’s a solved problem for many handwritting recognition problems.
Would be nice if it was possible to change the proxy settings for each request or for each session. Last time I checked it was only possible with the C API.