Running LLMs on PyTorch with ROCm#44
Conversation
|
@bbrosenba , @ChineduN for visibility |
A test in strix halo would be good. |
jammm
left a comment
There was a problem hiding this comment.
Looks good overall. Check my comments re. environment setup for README.md.
Please test on strix halo for both Windows and Linux.
Done, Strix Halo Windows and Linux Tested. |
|
@danielholanda The team reviewed and clearly highlighted the installation clarity needed in pytorch installs for ROCm. |
This string is correct here^ |
@DavidPerryamd, to avoid duplication, we have this for https://github.com/amd/halo_playbooks/blob/e4aa27c81e43a0f411288d28b65a827ab594531b/playbooks/dependencies/pytorch.md pytorch installation
|
…md/halo_playbooks into iswarya/rocm-inference-pytorch
There was a problem hiding this comment.
Hey - I'm Adam from the solutions team, and I'll be helping with UX Review.
Good job overall - just a couple more minor things that I commented.
The main question I have is: do users have access to the full files in assets/? If not, we need to provide them the full code on the website. This can be provided either by: 1 long code snippet (with copy button), or, preferably, several code snippets that guide the user through each important learning goal.
- ex. run_llm.py is pretty simple and straightforward. Maybe we can shorten the code (lines-wise) and put it all in one big code block.
- summarizer.py is more complex - we could separate it into a couple sections with comments, and explain key parts, like the system prompt,
@adamlam2-amd To give valuable learning and one click try to the user, would be great to retain the code snippet as short as it already is Recording.2026-01-23.154156.mp4 |
|
@jammm We are likely merging very soon. Please let us know if you have any additional recommendations or if this looks good. |
|
@adamlam2-amd @iswaryaalex Added the code preview feature directly to this PR |
jammm
left a comment
There was a problem hiding this comment.
LGTM after the feedback is addressed. The part about screenshots would be nice to have but not critical for initial release so is optional.
adamlam2-amd
left a comment
There was a problem hiding this comment.
Thanks for all the hard work @iswaryaalex . Thanks @danielholanda for the code preview addition.
|
All feedback comments are addressed. Merging this playbook |
…md/halo_playbooks into iswarya/rocm-inference-pytorch


This tutorial is to quickly bringup user using a super easy example
I have run these examples on my STRX Halo environment.
Need to check if the setup instructions are suitable for Halo Box