Phani Srikanth

Apologies for responding late on this PR @Nun-z. I am unable to locate the free credit for new users. Could you please point me where exactly is this information disclosed?...

Thanks for this @ashtonsix. The PR says "Salamander gives 82.5$ by redeeming AWS coupons" while students themselves need to have the coupon. Hence, maybe we should change the "Salamander gives..."...

I'm interested in contributing them. Could you point out a few references.

For large datasets, we typically tend to use online learning and FTRL is known to do well because of the adaptive learning rate scheme and it's ability to activate a...

Hi @psinger - let me share some background context. I'm trying to finetune the falcon 7b model using oasst dataset with 8 V100GPUs. if I'm able to do this, I...

I am using 16GB V100 for now. Do you have a working hyperparameters set which is used for h2ogpt training?

Great, thanks! What was the hardware used for finetuning?

This helps, thanks @psinger. Let me try the following experiments and get back to this thread. 1. `int4` on V100 (32GB). 2. Both `int8` and `int4` on A10. Appreciate the...

As I mentioned in #185 , `int4` finetuning works and I'm able to finetune a 20B model on 8 32GB V100s. I just noticed falcon 7b and 40b are not...

That works. Thank you @psinger . I'm finetuning the falcon models and it looks like I'm hitting the error tracked in #170 . I'll follow that issue. Closing this for...