Federico Belotti
Federico Belotti
I've run an experiment on 4 A10G on a [Lightning Studio](https://lightning.ai/) with the following command: ```bash python sheeprl.py exp=dreamer_v3_100k_ms_pacman fabric.devices=4 fabric.precision=32 fabric.accelerator=gpu ``` where I've manually changed the following: *...
Hi @geranim0, you need to update PyTorch to the nightly build (`torch>=2.3`) where they have fixed the `one_hot`. Could you update the pytorch version and tell us your results? Thank...
Hi @zichunxx, I will have a look in the next few days after some deadlines. Thank you
Have you tried with another buffer, like the standard `ReplayBuffer` or the `SequentialReplayBuffer`? Does it give you the same error?
Hi @zichunxx, I tried yesterday on my machine and reached more than 200k steps without errors: how many steps can you print before the error is raised? PS I had...
That would awesome. Have you considered to open a PR with an initial draft?
Hi @defrag-bambino, that was something that we have always had in mind to implement but never did. Yes, we should definitely try to add it to SheepRL. As far as...
Hi @defrag-bambino. Yes, Dreamer-V3 needs sequential data, so you should modify the SequentialReplayBuffer
This is awesome! I'll try it out on lightning.ai with crafter in the next few days! Thank you @defrag-bambino
Hi @defrag-bambino, sorry for the laaate response. Have you managed to implement this in the buffers.py? I would create a new buffer class extending from the SequentialReplayBuffer by changing the...