Rohan Mehta

icon indicating a website link https://www.rohanmehta.com

icon company @OpenAI icon location New York, NY icon location I'm a software engineer based in New York. I currently work at OpenAI, and previously founded a YC backed startup and worked at Meta for almost 9 years.

Results 5 issues of Rohan Mehta

Possible to continue pretraining?

10 comment

I know the general recommendation is to leave the backbone frozen and train task-specific heads. However I'm interested in continuing pre-training to better fit the backbone features to my dataset....

FLAN_T5_XXL_GENERATION model is inaccessible

## 🐛 Bug **Describe the bug** The FLAN T5 XXL model is not downloadable: it throws a 403 error. Likely either needs to be uploaded to the bucket, or just...

Show dates from previous/next month?

6 comment

Hi, quick question - is there a way to configure the view to show dates from the previous/next month? Here’s a random example of what I mean, from a different...

enhancement

How many samples were used to train Flan-T5?

Hey all, possibly silly question. I see that the huggingface collection has many millions of samples, and the google blog post suggests that the collection has 15M samples: https://ai.googleblog.com/2023/02/the-flan-collection-advancing-open.html On...

Multi-GPU training crashes with "Cannot copy out of meta tensor; no data"

I cloned the repo, and ran the provided training command from [here](https://github.com/mlfoundations/open_flamingo/blob/main/open_flamingo/train/README.md#example-training-command) on 1 node, 2 GPUs, and it failed with the stack trace below. I've made no changes to...

bug