Getting Very Close to getting AMD working; stuck on libc req's #12229
Replies: 4 comments 1 reply
-
Hi @src-r-r! Considering that your problem doesn't seem to be directly related to spaCy, I recommend you give the CuPy issue tracker or forum a shot. You'll likely get better advice there 🙂 |
Beta Was this translation helpful? Give feedback.
-
Hi @src-r-r , it's really promising that you've gotten this far! It would be great to have it working on AMD. Please do keep us updated if you're able to make progress :) |
Beta Was this translation helpful? Give feedback.
-
Hello, I have a vested interest in taking this over and hopefully finish the process. Currently we're at version 6.0 of ROCm and Cupy currently has experimental support for 5.7. So we will pin to 5.7 for ROCm in tests for now until Cupy supports 6.0 So here's my testing environment; here's my Python, ROCm and pip packages (in a virtualenv)
Finally here's my Env Variables that are relevant to the deployment of spacy
When I run the available GPU test on my AMD system;
installed cupy using steps detailed at:
Testing script for cuda/rocm
Installing spacy
testing with basics spacy train
testing without the HSA variable to prove the GPU is being targetted;
So it would seem we have Rocm5.7, Cupy w/Rocm support, and the issue collects as
However looking at rocm6 hipRand: I think its possible that everything is correct, but we simply can't actually use an equivalent ROCm function. But I have reached the limit of my understanding of this so far. Another alternative issue is maybe Thinc is not supporting ROCm and so it can't load the GPU into buffer. I did test with different sets of packages, such as removing HIP libraries and alternate ROCm libs, results in combinations of GPU not found and "HIP" GPU not found. This is the farthest I've gotten, where its actually trying to init the GPU for processing. Happy to test other scripts or debug as needed. I am dedicated to this issue as its the only thing holding me back from using GPU for much of my spacy needs. Thank you |
Beta Was this translation helpful? Give feedback.
-
Can we remove the answered tracker? I feel the subject isn't answered, as Spacy does not run on AMD yet. Otherwise I can open a new topic |
Beta Was this translation helpful? Give feedback.
-
This might be a little outside the scope, but here is what I discovered.
For reference my OS is Ubuntu 22.04 and my card is AMD Radeon RX580.
If you check in the Cupy issues, it looks like there's an open ticket to have ROCM v 5.4 merged into the branch.
So if you're using ROCM I'd highly recommend uninstalling the core ROCM driver and instead install the ROCM-install script (removal instructions are also in the docs).
The last supported version of ROCM appears to be 5.0. Not 5.3 or 5.4, but 5.0.
Trust me, I tried to run cupy with ROCM 5.3 and ran into symbol errors, leading me to believe that it's not forward-compatible.
So then I used
rocm-install
to install v. 5.0.2. But here I ran into issues.ROCM v. 5.0.2 requires stdlibc6-dev and stdc++-7-dev, which is really old (by C standards), and conflicts with the current gcc version 11.
That's alright. I can always install gcc from source.
But even then I'm running into issues.
dpkg
is a wash as well.Or maybe I'm going down a rabbit hole with the whole GCC thing.
Anyone have any ideas? About ready to build cupy for 5.4 myself!
Beta Was this translation helpful? Give feedback.
All reactions