FIX: Update CUDA and improve status page with diagnostics

[mmcky]

fixes #10 #5

[netlify]

✅ Deploy Preview for timely-seahorse-68815c ready!

Name	Link
🔨 Latest commit	b790186
🔍 Latest deploy log	https://app.netlify.com/sites/timely-seahorse-68815c/deploys/65c048bbbd1b61000847f480
😎 Deploy Preview	https://deploy-preview-11--timely-seahorse-68815c.netlify.app
📱 Preview on mobile	Toggle QR Code... Use your smartphone camera to open QR code link.

---

To edit notification comments on pull requests, go to your Netlify site configuration.

[github-actions]

🚀 Deployed on https://65c0652199ee79e1b17e9933--timely-seahorse-68815c.netlify.app

[mmcky]

• rerun with jax[_local] installed instead of jax[_pip] as per 9f96343

[mmcky]

• re-run with latest docker (switched numpyro and jax) -- it looks like numpyro is causing linking issues in jax.

[mmcky]

@HumphreyYang this is now installing torch and pyro-ppl in the Docker container then installing jax to ensure appropriate CUDA linking for JAX. It is running with gpu enabled as reported in https://65bc2ecf929de00cbf2d4378--timely-seahorse-68815c.netlify.app/status

I am surprised though to see the runtime now 5000s so maybe torch is not working as expected.

• build a CUDA=12.1 docker image with torch, pyro-ppl and jax to see if we can improve the runtime with proper driver linking. We need to install jax in the 12.1 context without 12.3 driver downloads as a whl

Update: the above configuration still has an issue with pytorch and jax in the same conda environment. I have just come across the following:
jax-ml/jax#18032
so it is a known issue that is currently not resolved. I think what we have now using CUDA=12.3 might be the best we can achieve without loading two separate conda environments (at the lecture lecture).

[mmcky]

• test CUDA=11.8 as both work in that context

down from 7 to 1 incompatible dependency between the two

torch 2.2.0+cu118 requires nvidia-cudnn-cu11==8.7.0.84; platform_system == "Linux" and platform_machine == "x86_64", but you have nvidia-cudnn-cu11 8.9.6.50 which is incompatible.

• need to review where docker is running out of memory for this test. Doesn't appear linked to the hdd allocation

[mmcky]

• wait for [bayes_nonconj] Migrate from pytorch to just use JAX #13 as it will remove two dependencies causing the issues of software conflicts. Then we can use the latest CUDA for JAX

[mmcky]

@kp992 I will merge this PR to remove the CUDA failure message that is currently live on stats.quantecon.org and then I will open an issue to update the docker image to use mmcky/quantecon-lecture-python:cuda-12.3.1-anaconda-2023-09-py311 once we have removed pytorch and pyro successfully.

[kp992]

Sounds good to me. Thanks @mmcky