Prasad Chalasani

icon indicating a website link https://github.com/langroid/langroid icon indicating an email [email protected]

icon company MLResearch icon location Pittsburgh, PA icon location LLM/ML research/practice. Building langroid, the easiest way to develop LLM-agent-oriented systems. https://github.com/langroid/langroid

Results 131 comments of Prasad Chalasani

Fix issue #275: training=False needed for eval envs

> thanks for the PR, overall LGTM =) i will do some quick tests (check that it doesn't affect results) when I'm back from holidays next week ;) (you can...

Fix issue #275: training=False needed for eval envs

Great, I took a crack at updating the changelog, not sure what the right convention is (and unsure about release numbering and such)

Fix issue #275: training=False needed for eval envs

> @pchalasani did you check "allow edit from maintainers"? I cannot push to your branch.... yes it is checked

Fix issue #275: training=False needed for eval envs

sure, just added you

Support for Stoch Wt Avg (SWA) closes #321

I realized we need to do `opt.swap_swa_sgd()` at the *end* of training, and some further thought is needed, to see how this impacts computation of validation metrics by `EvalCallback` etc

Support for Stoch Wt Avg (SWA) closes #321

I added `opt.swap_swa_sgd()` after `model.learn`. We also need to do this before and after each `evaluate_policy()` call in `EvalCallback` (which is in the original sb3 repo), so that validation metrics...

[Feature Request] Support Stochastic Weight Averaging (SWA) for improved stability

Thanks, I did not know SWA is in main PyTorch. I will look into it. As for empirical evidence, I'll continue experimenting and report back.

what exactly is the input data format expected by Metronome?

Thanks for the clarifications. I was just trying to figure out how I can (say) use Metronome to deploy deep-learning on Hadoop for one of our data-sets. Eventually, I'll probably...

Run git-deps transparently/automatically

I agree, it would be great if I simply put the appropriate `:git-dependencies` and `:plugins` into my `project.clj`, and then when I run `lein uberjar`, it should: (a) git clone/pull...

Update to Pydantic 2.0

> blocked on this by agent-search. I'm surprised this is the only blocker (if so, that's a minor one, since agent-search is just an extra and we can work around...