Comments (17)
The url is till blocked - any updates?
from triton.
I also faced this HTTP 409 error when building PyTorch 2.4 + ROCm. I think this problem of Triton not building points to AOTriton (AMD's fork) using the old upstream commit containing the now broken URL.
from triton.
For anyone who might be in a similar situation:
I've run into the same issue recently when trying to build PT for ROCm from source. I've tried a couple of ideas, including upgrading LLVM in AMD's triton fork, etc., all in vain. In the end what worked well for me was to follow the Building with custom LLVM part of README.
- Clone and checkout LLVM to commit
49af6502c6dcb4a7f7520178bd14df396f78240c
. - Build with AMDGPU support
- set the env variables as explained in the readme (I used
export ...
) - build pytorch as usual
No more HTTP 409/missing blobs/version mismatch problems - everything went smoothly.
from triton.
I also had mixed results on building Triton (upstream commit cb3d79a185e40c9d8a579bea07747a8a8d157d52
and AMD's triton commit 9b73a543a5545960bcaf2830900b0560eec443c5
) with this particular LLVM commit (49af6502c6dcb4a7f7520178bd14df396f78240c)
. I seem to cannot get past this error installing the Python package:
CMake Error at CMakeLists.txt:61 (find_package):
Found package configuration file:
/root/triton-llvm-build/lib/cmake/mlir/MLIRConfig.cmake
but it set MLIR_FOUND to FALSE so package "MLIR" is considered to be NOT
FOUND. Reason given by package:
The following imported targets are referenced, but are missing:
LLVMNVPTXCodeGen LLVMNVPTXDesc LLVMNVPTXInfo
-- Configuring incomplete, errors occurred
I built LLVM with CMake using the following flags:
-B /root/triton-llvm-build \
-DCMAKE_BUILD_TYPE=Release \
-DLLVM_DEFAULT_TARGET_TRIPLE="x86_64-pc-linux-gnu" \
-DLLVM_TARGETS_TO_BUILD="AMDGPU;NVPTX;Native" \
-DLLVM_ENABLE_PROJECTS="mlir" \
-DLLVM_ENABLE_ASSERTIONS=ON \
-DLLVM_CCACHE_BUILD=ON \
-G Ninja
The find command shows these targets are present, despite what CMake says.
find /root/triton-llvm-build -name "*LLVMNVPTXCodeGen*"
/root/triton-llvm-build/lib/Target/NVPTX/CMakeFiles/LLVMNVPTXCodeGen.dir
/root/triton-llvm-build/lib/libLLVMNVPTXCodeGen.a
find /root/triton-llvm-build -name "*LLVMNVPTXDesc*"
/root/triton-llvm-build/lib/Target/NVPTX/MCTargetDesc/CMakeFiles/LLVMNVPTXDesc.dir
/root/triton-llvm-build/lib/libLLVMNVPTXDesc.a
find /root/triton-llvm-build -name "*LLVMNVPTXInfo*"
/root/triton-llvm-build/lib/Target/NVPTX/TargetInfo/CMakeFiles/LLVMNVPTXInfo.dir
/root/triton-llvm-build/lib/libLLVMNVPTXInfo.a
Environment variables used:
LLVM_LIBRARY_DIR=/root/triton-llvm-build/lib
LLVM_INCLUDE_DIRS=/root/triton-llvm-build/include
LLVM_SYSPATH=/root/triton-llvm-build
from triton.
I'm seeing this too. The link to the tarball is now restricted.
from triton.
In case anyone is looking into this: I've tried accessing https://tritonlang.blob.core.windows.net via an internal redhat.com network, a private network and via an online anonymizer. It looks like the site is locked down for some reason and access to the tarball is restricted as indicated in the initial report by @lisaong
from triton.
Is there a way to provide a 1-time download for the tarball? Perhaps through a direct commit to the triton repo itself?
from triton.
We are also seeing this when trying to build triton in pytorch: https://github.com/pytorch/pytorch/actions/runs/10421322991/job/28863284428
from triton.
That URL was changed back in June:
06e6799
Perhaps torch is on an older commit?
from triton.
I'm also seeing the similar issue. Any updates?
from triton.
This is still an issue, AOTriton still can not be build
from triton.
For anyone attempting to build aotriton from ROCm. aotriton references a DIFFERENT repository than triton-lang, and I have filed an issue here: ROCm#631
from triton.
That URL was changed back in June: 06e6799
Perhaps torch is on an older commit?
Thank you! Using the new URL solved this problem.
from triton.
@jayfurmanek torch is indeed on an older commit. Is there any plan to provide the older llvm-49af6502-ubuntu-x64.tar.gz tarball at the new oatriton download link? If not, is there a way I and others could get a copy?
from triton.
Thanks. I had actually stumbled across this today and was trying it out :) It seems like it works -- I'm just worried that the AMD llvm may have additional flags we're not compiling with. Hopefully someone from the AMD ROCm team can confirm the above instructions are valid.
from triton.
Related Issues (20)
- Fused Attention FP8 correctness on L20(Ada GPU) HOT 1
- What impact does changing the kernel shape's divisibility value have?
- Does Triton have a roadmap plan? HOT 1
- FP8 Casting Tensor Size Alignment Error on AMD MI210 HOT 2
- Output dtype of 09-persistent-matmul.py
- Worse performance on `H100` than `RTX3090`
- [Bug] Assertion `idx < size()' failed. HOT 7
- Seems like these slides are missing, if possible, would you mind update the links or re-upload the files?
- Should @core.extern be part of the libdevice interface? HOT 1
- Cannot find 2.0.0.dev20221202 version HOT 2
- make error
- FlexAttention Segmentation Fault HOT 2
- What is the use of divisibility in coalesce pass
- Confusion about memory of pointers HOT 1
- [Tutorial] Fused-attention `bwd` tutorial incorrect for `causal=True`
- The question of implementing the addmm_ operator based on triton HOT 1
- Nothing found at https://oaitriton.blob.core.windows.net/public/llvm-builds/llvm-ce80c80d-centos-x64.tar.gz
- Numeric errors with tf32 matmul on A100 GPU HOT 2
- DeepSpeed came to Windows natively - pip install deepspeed - where are you Triton?
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from triton.