Copyright © 2023-2024 Advanced Micro Devices, Inc. All rights reserved.
- About zentorch
- Installation
- Usage
- Logging and Debugging
- Performance tuning and Benchmarking
- Additional Utilities
The latest ZenDNN Plugin for PyTorch* (zentorch) 5.0 is here!
This powerful upgrade continues to redefine deep learning performance on AMD EPYC™ CPUs, combining relentless optimization, innovative features, and industry-leading support for modern workloads.
zentorch 5.0 takes deep learning to new heights with significant enhancements for bfloat16 performance, expanded support for cutting-edge models like Llama 3.1 and 3.2, Microsoft Phi, and more as well as support for INT4 quantized datatype. This includes the advanced Activation-Aware Weight Quantization (AWQ) algorithm, driving remarkable accuracy in low-precision computations.
Combined with PyTorch's torch.compile, zentorch transforms deep learning pipelines into finely-tuned, AMD-specific engines, delivering unparalleled efficiency and speed for large-scale inference workloads.
The zentorch 5.0 plugs seamlessly with PyTorch version 2.4.0, offering a high-performance experience for deep learning on AMD EPYC™ platforms.
We welcome feedback, suggestions, and bug reports. Should you have any of the these, please kindly file an issue on the ZenDNN Plugin for PyTorch Github page here
AMD copyrighted code in ZenDNN is subject to the Apache-2.0, MIT, or BSD-3-Clause licenses; consult the source code file headers for the applicable license. Third party copyrighted code in ZenDNN is subject to the licenses set forth in the source code file headers of such code.
zentorch consists of three parts. They are
- ZenDNN Integration Code
- zentorch backend
- Build System
ZenDNN is integrated into zentorch using CPP code which interfaces ATen API to ZenDNN's API. This code exports torch compatible API similar to ATen IR of PyTorch. The exported API is made available for usage from python code using TORCH_LIBRARY, TORCH_LIBRARY_FRAGMENT and TORCH_LIBRARY_IMPL. Integration code is linked and compiled into zentorch using CppExtension provided by PyTorch.
The following ops are integrated as of now:
- Embedding bag op
- Embedding op
- Matmul ops
- Custom Fusion ops
- Rope op
- MHA op
We have registered a custom backend to torch.compile called zentorch. This backend integrates ZenDNN optimizations after AOTAutograd through a function called optimize. This function operates on the FX based graph at the ATEN IR to produce an optimized FX based graph as the output.
The static libraries for ZenDNN, AOCL BLIS and the cpp Extension modules that bind the ZenDNN operators with Python are built using setup.py script.
CMake downloads the ZenDNN , AOCL BLIS and FBGEMM during configure stage. It generates a config.h with GIT hashes of ZenDNN , AOCL BLIS and FBGEMM. It builds ZenDNN , AOCL BLIS and FBGEMM as static libraries.
The CPP code, being an extension module, is built through CppExtension. It takes static libraries of the ZenDNN , AOCL BLIS and FBGEMM libraries. setup.py also adds in various attributes to the zentorch for debugging and providing additional information.
zentorch uses following libraries for its functionality.
zentorch can be installed using binary wheel file or can be built from source itself.
- Create conda or python environment and activate it.
- Uninstall any existing zentorch installations.
pip uninstall zentorch- Install Pytorch v2.4.0
conda install pytorch==2.4.0 cpuonly -c pytorch- Use one of two methods to install zentorch:
Using pip utility
pip install zentorch==5.0.0or
Using the release package.
Download the package from AMD developer portal from here.
Run the following commands to unzip the package and install the binary.
unzip ZENTORCH_v5.0.0_Python_v3.8.zip
cd ZENTORCH_v5.0.0_Python_v3.8/
pip install zentorch-5.0.0-cp38-cp38-manylinux_2_28_x86_64.whlNotes:
- In above steps, we have taken an example for release package with Python version 3.8.
- Dependent packages 'numpy' and 'torch' will be installed by 'zentorch' if not already present.
- If you get the error: ImportError: /lib64/libstdc++.so.6: version `GLIBCXX_.a.b.cc' not found (required by <path_to_conda>/envs/<env_name>/lib/python<py_version>/site-packages/zentorch-5.0.0-pyx.y-linux-x86_64.egg/zentorch/_C.cpython-xy-x86_64-linux-gnu.so), export LD_PRELOAD as:
- export LD_PRELOAD=<path_to_conda>/envs/<env_name>/lib/libstdc++.so.6:$LD_PRELOAD
Run the following commands:
git clone https://github.com/amd/ZenDNN-pytorch-plugin.git
cd ZenDNN_PyTorch_Plugin/Note: Repository defaults to master branch, to build the version 5.0 checkout the branch r5.0.
git checkout r5.0Build setup downloads the ZenDNN, AOCL BLIS and FBGEMM repos into third_party folder.
conda create -n pt-zentorch python=3.8 -y
conda activate pt-zentorch# Pip command
pip install torch==2.4.0 --index-url https://download.pytorch.org/whl/cpuor
# Conda command
conda install pytorch==2.4.0 cpuonly -c pytorch -yNote: The CPU version of torch/pytorch only supports CPU version of torchvision.
Note: cmake & ninja are required for cpp extension builds, will be installed through build script.
pip install -r requirements.txtpython setup.py bdist_wheelNote: The wheel file will be generated in dist folder in ZenDNN_PyTorch_Plugin directory
cd dist
pip install zentorch-5.0.0-cp38-cp38-linux_x86_64.whlpython test/test_zentorch.pypython test/test_zentorch.py --seed seed_valuepython setup.py clean --all# Using torch.compile with 'zentorch' as backend
import torch
import zentorch
compiled_model = torch.compile(model, backend='zentorch')
with torch.no_grad():
output = compiled_model(input)Note: If same model is optimized with
torch.compilefor multiple backends within single script, it is recommended to usetorch._dynamo.reset()before calling thetorch.compileon that model. This is applicable if torch version is less than 2.3.
Note: zentorch is able to do the zentorch op replacements in both non-inference and inference modes. But some of the zentorch optimizations are only supported for the inference mode, so it is recommended to use
torch.no_grad()if you are running the model for inference only.
For CNN models, set dynamic=False when calling for torch.compile as below:
model = torch.compile(model, backend='zentorch', dynamic=False)
with torch.no_grad():
output = model(input)For HuggingFace NLP models, optimize them as below:
model = torch.compile(model, backend='zentorch')
with torch.no_grad():
output = model(input)For HuggingFace Generative LLM models, usage of zentorch.llm.optimize is recommended. All the optimizations included in this API are specifically targeted for Generative Large Language Models from HuggingFace. If a model is not a valid Generative Large Language Model from HuggingFace, the following warning will be displayed and zentorch.llm.optimize will act as a dummy function with no optimizations being applied to the model that is passed: “Cannot detect the model transformers family by model.config.architectures. Please pass a valid HuggingFace LLM model to the zentorch.llm.optimize API.” This check confirms the presence of the "config" and "architectures" attributes of the model to get the model id. Considering the check, two scenarios the zentorch.llm.optimize can still act as a dummy function:
- HuggingFace has a plethora of models, of which Generative LLMs are a subset of. So, even if the model has the attributes of "config" and "architectures", the model id might not be yet present in the supported models list from zentorch. In this case zentorch.llm.optimize will act as a dummy function.
- A model can be a valid generative LLM from HuggingFace or not, might miss the "config" and "architectures" attributes. In this case also, the zentorch.llm.optimize API will act as a dummy function.
If the model passed is valid, all the supported optimizations will be applied, and performant execution is ensured. To check the supported models, run the following command:
python -c 'import zentorch; print("\n".join([f"{i+1:3}. {item}" for i, item in enumerate(zentorch.llm.SUPPORTED_MODELS)]))'If a model id other than the listed above are passed, zentorch.llm.optimize will not apply the above specific optimizations to the model and a warning will be displayed as follows: “Complete set of optimizations are currently unavailable for this model.” Control will pass to the zentorch custom backend to torch.compile for applying optimizations.
For leveraging the best performance of zentorch_llm_optimize, user has to install IPEX corresponding to the PyTorch version that is installed in the environment. The PyTorch version for performant execution of supported LLMs should be greater than or equal to 2.3.0. Recommended version for optimal performance is using PyTorch 2.4.
model = zentorch.llm.optimize(model, dtype)
model = torch.compile(model, backend='zentorch')
with torch.no_grad():
output = model(input)model = zentorch.llm.optimize(model, dtype)
model.forward = torch.compile(model.forward, backend='zentorch')
with torch.no_grad():
output = model.forward(input)- Optimize the `model.forward` with torch.compile instead of `model.generate`.
- But still generate the output through a call to `model.generate`.
model = zentorch.llm.optimize(model, dtype)
model.forward = torch.compile(model.forward, backend='zentorch')
with torch.no_grad():
output = model.generate(input)Huggingface models are quantized using AMD's Quark tool. After downloading the zip file, install Quark and follow the below steps:
- Enter the examples/torch/language_modeling directory
- Run the following command
OMP_NUM_THREADS=<physical-cores-num> numactl --physcpubind=<physical-cores-list> python quantize_quark.py --model_dir <hugging_face_model_id> --device cpu --data_type bfloat16 --model_export quark_safetensors --quant_algo awq --quant_scheme w_int4_per_group_sym --group_size -1 --num_calib_data 128 --dataset pileval_for_awq_benchmark --seq_len 128 --output_dir <output_dir> --pack_method orderAs currently HF does not support AWQ format for CPU, an additional codeblock needs to be added to your inference script for loading the WOQ models.
config = AutoConfig.from_pretrained(model_id, trust_remote_code=True, torch_dtype=torch.bfloat16)
model = AutoModelForCausalLM.from_config(config, trust_remote_code=True, torch_dtype=torch.bfloat16)
model = zentorch.load_woq_model(model, safetensor_path)Here, safetensor_path refers to the "<output_dir>" path of the quantized model. After the loading steps, the model can be executed in a similar fashion as the cases# 1-3 listed in section 3.4.
Logging for ZenDNN is disabled by default but can be enabled by using the environment variable ZENDNN_LOG_OPTS before running any tests. Its behavior can be specified by setting ZENDNN_LOG_OPTS to a comma-delimited list of ACTOR:DBGLVL pairs. An example to turn on info logging is given below.
export ZENDNN_LOG_OPTS=ALL:2To enable the profiling logs zendnn_primitive_create and zendnn_primitive_execute, you can use:
export ZENDNN_PRIMITIVE_LOG_ENABLE=1For further details on ZenDNN logging mechanism, refer to ZenDNN user-guide from this page.
For zentorch, CPP specific logging can be enabled by setting the environment variable TORCH_CPP_LOG_LEVEL. This has four levels: INFO, WARNING, ERROR and FATAL in decreasing order of verbosity. Similarly, python logging can be enabled by setting the environment variable ZENTORCH_PY_LOG_LEVEL, this has five levels: DEBUG, INFO, WARNING, ERROR and CRITICAL, again in decreasing order of verbosity. An example to enable INFO level logs for cpp and DEBUG level for python (most verbose) is given below:
export TORCH_CPP_LOG_LEVEL=INFO
export ZENTORCH_PY_LOG_LEVEL=DEBUGThe default level of logs is WARNING for both cpp and python sources but can be overridden as discussed above.
NOTE: The log levels are the same as those provided by the python logging module.
INFO: Since all OPs implemented in zentorch are registered with torch using the TORCH_LIBRARY(), TORCH_LIBRARY_FRAGMENT() and TORCH_LIBRARY_IMPL() macros in bindings, the PyTorch profiler can be used without any modifications to measure the op level performance.
PyTorch offers a debugging toolbox that comprises a built-in stats and trace function. This functionality facilitates the display of the time spent by each compilation phase, output code, output graph visualization, and IR dump. TORCH_COMPILE_DEBUG invokes this debugging tool that allows for better problem-solving while troubleshooting the internal issues of TorchDynamo and TorchInductor. This functionality works for the models optimized using zentorch, so it can be leveraged to debug these models as well. To enable this functionality, users can either set the environment variable TORCH_COMPILE_DEBUG=1 or specify the environment variable with the runnable file (e.g., test.py) as input.
# test.py contains model optimized by torch.compile with 'zentorch' as backend
TORCH_COMPILE_DEBUG=1 python test.pyFor more information about TORCH_COMPILE_DEBUG refer to the official PyTorch documentaion available.
zentorch v5.0.0 is supported with ZenDNN v5.0. Please see the Tuning Guidelines section of ZenDNN User Guide for performance tuning. ZenDNN User Guide can be downloaded from here
To check the version of zentorch use the following command:
python -c 'import zentorch; print(zentorch.__version__)'To check the build config of zentorch use the following command:
python -c 'import zentorch; print(*zentorch.__config__.split("\n"), sep="\n")'