-
Notifications
You must be signed in to change notification settings - Fork 92
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
jax initialization issues #488
Comments
@karanv99 can you leave more information here so we can address this? |
Yup sure, On MAC M1For this diagnosis I am running Pymbar4.0.1 in a M1 mac air 2020(macOS Monterey). I will attach a screenshot of successful installation of pymbar 4.0 for your diagnosis. This is the error that I get when trying to import pymbar 4.0 I tried reading about it and went through your source code trying to figure out what is wrong? The most I could come up with by comparing pymbar3.0 and pymbar4.0 source code was that you in 4.0 onwards you started using jax.numpy instead of numpy arrays(probably because of efficiency or smoothnes). And I found jax has issues on apples silicon chips because of the different architecture and is not well configured. On Inter i7 PCPrompt of successful installation of pymbar4.0 This is where initializing mbar using pymbar 4.0 gets stuck on the i7 chip. I tried and waited for more than an hour, hopping it would get initialized but it was unsuccessful. I compared mbar_solvers.py between 3.0 and 4.0, in pymbar3.0 adaptive is the default solver while it is hybr in 4.0. Even though the prompt does claim to try out next method, the code gets stuck as mentioned in the image 5. DisclaimerI am new to this whole thing and I am trying to figure things out myself, so I might be wrong on many levels which I have mentioned above. If i am doing something silly, I really do apologise. Thank you!! :) |
So, there is no installation issue on Intel, which is good. "No GPU/TPU found, falling back to CPU" is exactly what you want to happen. For pymbar 4.0 on intel, try using the solver 'robust', and setting the output to 'verbose' (see some of the examples), see what is happening, and let me know what is going on. If there is poor overlap in the collected data, then convergence can take some playing around. For the problems installing in M1, conda installation is the suggested route (on all applications). It could be that the pip install is getting confused. Can you try to install via conda instead? Install anaconda or miniconda, set the channels as described in the documentation, and conda install pymbar there. |
I don't have an M1 to test against directly, but from what I am reading of other people having the "partially initialized" is their The other main issue I am seeing is Homebrew |
Hello, everyone. I have an update: I am trying to figure things out how pymbar 4.0 works(like its syntax etc etc) on the intel PC. I will get back once I figure out with it. As for the M1, as @mrshirts suggested I think it was the issue with the pip install. I installed pymbar4.0 using conda and it installed the required jax libraries and pymbar is getting imported. I am still figuring out how generate_fes() works on pymbar. Once I do that I will get back. Thank you all for your help!! Note!! Could you please check out the output repository for parallel tempering? I think it has not been updated with the new parallel tempering code. It has the same output from the previous repo. |
Split from #487: Pymbar 4.0 uses jax.numpy and jax is not well configured for newer macs(m1) and I tried using it on an intel i7 cpu but it is not getting initialized.
The text was updated successfully, but these errors were encountered: