update doc for 2.0.0 release (#1486)

update tuning guide add known issues add release notes update cppsdk installation guide
intel · Mar 22, 2023 · 99a70a6 · 99a70a6
1 parent c52d951
commit 99a70a6
Show file tree

Hide file tree

Showing 5 changed files with 137 additions and 65 deletions.
diff --git a/docs/tutorials/getting_started.md b/docs/tutorials/getting_started.md
@@ -51,7 +51,7 @@ with torch.no_grad(), torch.cpu.amp.autocast():
   model(data)
 ##########################################
 
-############ T##orchDynamo ###############
+############## TorchDynamo ###############
 model = ipex.optimize(model)
 
 model = torch.compile(model, backend="ipex")

diff --git a/docs/tutorials/installation.md b/docs/tutorials/installation.md
@@ -152,7 +152,7 @@ docker pull intel/intel-optimized-pytorch:latest
 
 |Version|Pre-cxx11 ABI|cxx11 ABI|
 |--|--|--|
-| 2.0.0 | [libintel-ext-pt-2.0.0+cpu.run](http://intel-optimized-pytorch.s3.cn-north-1.amazonaws.com.cn/libipex/cpu/libintel-ext-pt-2.0.0%2Bcpu.run) | [libintel-ext-pt-cxx11-abi-2.0.0+cpu.run](http://intel-optimized-pytorch.s3.cn-north-1.amazonaws.com.cn/libipex/cpu/libintel-ext-pt-cxx11-abi-2.0.0%2Bcpu.run) |
+| 2.0.0 | [libintel-ext-pt-2.0.0+cpu.run](https://intel-extension-for-pytorch.s3.amazonaws.com/libipex/cpu/libintel-ext-pt-2.0.0%2Bcpu.run) | [libintel-ext-pt-cxx11-abi-2.0.0+cpu.run](https://intel-extension-for-pytorch.s3.amazonaws.com/libipex/cpu/libintel-ext-pt-cxx11-abi-2.0.0%2Bcpu.run) |
 | 1.13.100 | [libintel-ext-pt-1.13.100+cpu.run](http://intel-optimized-pytorch.s3.cn-north-1.amazonaws.com.cn/libipex/cpu/libintel-ext-pt-1.13.100%2Bcpu.run) | [libintel-ext-pt-cxx11-abi-1.13.100+cpu.run](http://intel-optimized-pytorch.s3.cn-north-1.amazonaws.com.cn/libipex/cpu/libintel-ext-pt-cxx11-abi-1.13.100%2Bcpu.run) |
 | 1.13.0 | [libintel-ext-pt-1.13.0+cpu.run](http://intel-optimized-pytorch.s3.cn-north-1.amazonaws.com.cn/libipex/cpu/libintel-ext-pt-1.13.0%2Bcpu.run) | [libintel-ext-pt-cxx11-abi-1.13.0+cpu.run](http://intel-optimized-pytorch.s3.cn-north-1.amazonaws.com.cn/libipex/cpu/libintel-ext-pt-cxx11-abi-1.13.0%2Bcpu.run) |
 | 1.12.300 | [libintel-ext-pt-1.12.300+cpu.run](http://intel-optimized-pytorch.s3.cn-north-1.amazonaws.com.cn/libipex/cpu/libintel-ext-pt-1.12.300%2Bcpu.run) | [libintel-ext-pt-cxx11-abi-1.12.300+cpu.run](http://intel-optimized-pytorch.s3.cn-north-1.amazonaws.com.cn/libipex/cpu/libintel-ext-pt-cxx11-abi-1.12.300%2Bcpu.run) |

diff --git a/docs/tutorials/performance_tuning/known_issues.md b/docs/tutorials/performance_tuning/known_issues.md
@@ -1,13 +1,76 @@
 Known Issues
 ============
 
+## Usage
+
 - There might be Python packages having PyTorch as their hard dependency. If you installed `+cpu` version of PyTorch, installation of these packages might replace the `+cpu` version with the default version released on Pypi.org. If anything goes wrong, please reinstall the `+cpu` version back.
 
 - If you found the workload runs with Intel® Extension for PyTorch\* occupies a remarkably large amount of memory, you can try to reduce the occupied memory size by setting the `--weights_prepack` parameter of the `ipex.optimize()` function to `False`.
 
-- Supporting of EmbeddingBag with INT8 when bag size > 1 is working in progress.
+- If inference is done with a custom function, `conv+bn` folding feature of the `ipex.optimize()` function doesn't work.
+
+  ```
+  import torch
+  import intel_pytorch_extension as ipex
+
+  class Module(torch.nn.Module):
+      def __init__(self):
+          super(Module, self).__init__()
+          self.conv = torch.nn.Conv2d(1, 10, 5, 1)
+          self.bn = torch.nn.BatchNorm2d(10)
+          self.relu = torch.nn.ReLU()
+
+      def forward(self, x):
+          x = self.conv(x)
+          x = self.bn(x)
+          x = self.relu(x)
+          return x
+
+      def inference(self, x):
+          return self.forward(x)
+
+  if __name__ == '__main__':
+      m = Module()
+      m.eval()
+      m = ipex.optimize(m, dtype=torch.float32, level="O0")
+      d = torch.rand(1, 1, 112, 112)
+      with torch.no_grad():
+        m.inference(d)
+  ```
+
+  This is a PyTorch FX limitation. You can avoid this error by calling `m = ipex.optimize(m, level="O0")`, which doesn't apply ipex optimization, or disable `conv+bn` folding by calling `m = ipex.optimize(m, level="O1", conv_bn_folding=False)`.
+
+## TorchDynamo
+
+- The support of torch.compile() with ipex as the backend is still an experimental feature. If the workload fails to run or demonstrates poor performance, you can use the `torch.jit` APIs and graph optimization APIs of ipex. Currently, the below HuggingFace models fail to run using torch.compile() with ipex backend due to memory issues:
+  - masked-language-modeling+xlm-roberta-base
+  - casual-language-modeling+gpt2
+  - casual-language-modeling+xlm-roberta-base
+  - summarization+t5-base
+  - text-classification+allenai-longformer-base-409
+
+## Dynamic Shape
+
+- When working with an NLP model inference with dynamic input data length appling with TorchScript (either `torch.jit.trace` or `torch.jit.script`), performance with Intel® Extension for PyTorch\* is possible to be less than that without Intel® Extension for PyTorch\*. In this case, adding the workarounds below would help solve this issue.
+  - Python interface
+    ```python
+	torch._C._jit_set_texpr_fuser_enabled(False)
+	```
+  - C++ interface
+    ```c++
+    #include <torch/csrc/jit/passes/tensorexpr_fuser.h>
+	torch::jit::setTensorExprFuserEnabled(false);
+	```
+
+## INT8
 
-- Compiling with gcc 11 might result in `illegal instruction` error.
+- Low performance with INT8 support for dynamic shapes
+
+  The support for dynamic shapes in Intel® Extension for PyTorch\* INT8 integration is still work in progress. When the input shapes are dynamic, for example inputs of variable image sizes in an object detection task or of variable sequence lengths in NLP tasks, the Intel® Extension for PyTorch\* INT8 path may slow down the model inference. In this case, use stock PyTorch INT8 functionality.
+
+  **Note**: Using Runtime Extension feature if batch size cannot be divided by number of streams, because mini batch size on each stream are not equivalent, scripts run into this issues.
+
+- Supporting of EmbeddingBag with INT8 when bag size > 1 is working in progress.
 
 - `RuntimeError: Overflow when unpacking long` when a tensor's min max value exceeds int range while performing int8 calibration. Please customize QConfig to use min-max calibration method.
 
@@ -31,55 +94,24 @@ Known Issues
   run_benchmark(freezed_model, input)
   ```
 
+## BFloat16
+
 - BF16 AMP(auto-mixed-precision) runs abnormally with the extension on the AVX2-only machine if the topology contains `Conv`, `Matmul`, `Linear`, and `BatchNormalization`
 
+## Runtime Extension
+
 - Runtime extension of MultiStreamModule doesn't support DLRM inference, since the input of DLRM (EmbeddingBag specifically) can't be simplely batch split.
 
 - Runtime extension of MultiStreamModule has poor performance of RNNT Inference comparing with native throughput mode. Only part of the RNNT models (joint_net specifically) can be jit traced into graph. However, in one batch inference, `joint_net` is invoked multi times. It increases the overhead of MultiStreamModule as input batch split, thread synchronization and output concat.
 
+## Correctness
+
 - Incorrect Conv and Linear result if the number of OMP threads is changed at runtime
 
   The oneDNN memory layout depends on the number of OMP threads, which requires the caller to detect the changes for the # of OMP threads while this release has not implemented it yet.
 
-- Low performance with INT8 support for dynamic shapes
-
-  The support for dynamic shapes in Intel® Extension for PyTorch\* INT8 integration is still work in progress. When the input shapes are dynamic, for example inputs of variable image sizes in an object detection task or of variable sequence lengths in NLP tasks, the Intel® Extension for PyTorch\* INT8 path may slow down the model inference. In this case, use stock PyTorch INT8 functionality.
-
-  **Note**: Using Runtime Extension feature if batch size cannot be divided by number of streams, because mini batch size on each stream are not equivalent, scripts run into this issues.
+## Float32 Training
 
 - Low throughput with DLRM FP32 Train
 
   A 'Sparse Add' [PR](https://github.com/pytorch/pytorch/pull/23057) is pending on review. The issue will be fixed when the PR is merged.
-
-- If inference is done with a custom function, `conv+bn` folding feature of the `ipex.optimize()` function doesn't work.
-
-  ```
-  import torch
-  import intel_pytorch_extension as ipex
-
-  class Module(torch.nn.Module):
-      def __init__(self):
-          super(Module, self).__init__()
-          self.conv = torch.nn.Conv2d(1, 10, 5, 1)
-          self.bn = torch.nn.BatchNorm2d(10)
-          self.relu = torch.nn.ReLU()
-
-      def forward(self, x):
-          x = self.conv(x)
-          x = self.bn(x)
-          x = self.relu(x)
-          return x
-
-      def inference(self, x):
-          return self.forward(x)
-
-  if __name__ == '__main__':
-      m = Module()
-      m.eval()
-      m = ipex.optimize(m, dtype=torch.float32, level="O0")
-      d = torch.rand(1, 1, 112, 112)
-      with torch.no_grad():
-        m.inference(d)
-  ```
-
-  This is a PyTorch FX limitation. You can avoid this error by calling `m = ipex.optimize(m, level="O0")`, which doesn't apply ipex optimization, or disable `conv+bn` folding by calling `m = ipex.optimize(m, level="O1", conv_bn_folding=False)`.