Description:
1. Quantize Linear Layer Weights to 4-bits:
Quantize the weights of the Linear layer to 4 bits, using symmetric quantization.
Pack two 4-bit weights into one uint8 container.
Choose a quantization scheme (channel-wise or group-wise), with the group size being a multiple of 32.
2. Prepare Quantized Weights, Scales, and Optional Bias:
After quantizing, obtain the quantized_weights, scales, and groupsize.
If the original Linear layer has a bias, prepare it as well.
3. Pack the Weights Efficiently:
Use torch.ops.aten._dyn_quant_pack_4bit_weight to optimally pack the weights, scales, and optional bias.
```python
packed_weights = torch.ops.aten._dyn_quant_pack_4bit_weight(weight, scales_and_zeros, bias, groupsize, in_features, out_features)
```
Input parameters should include:
in_features and out_features (the same as the Linear layer’s corresponding parameters).
4. Perform Dynamic Quantized Matrix Multiplication:
Use torch.ops.aten._dyn_quant_matmul_4bit to perform matrix multiplication with quantized weights.
```python
output = torch.ops.aten._dyn_quant_matmul_4bit(input, packed_weights, groupsize, in_features, out_features)
```
Inputs required include:
The input tensor, packed_weights , groupsize, and the in_features and out_features.
API Usage: https://github.com/pytorch/pytorch/issues/143289
Model Perf :
7B Transformer model:
Prefill : 340 t/s
Decode : 40 t/s
2B Transformer model
Prefill : 747 t/s
Decode : 80 t/s
Tests:
python test/test_linalg.py -k test__dyn_quant_pack_4bit_weight
Ran 1 test in 0.016s
OK
python test/test_linalg.py -k test__dyn_quant_matmul_4bit
Ran 8 tests in 0.077s
OK
python test/test_linalg.py -k test_compile_dyn_quant_matmul_4bit
Ran 8 tests in 11.454s
Change-Id: Ia1672bad5e6ec94e64d8bb1971395d60f4b3a452
Fixes #ISSUE_NUMBER
Pull Request resolved: https://github.com/pytorch/pytorch/pull/134124
Approved by: https://github.com/digantdesai, https://github.com/malfet
Description:
1. Quantize Linear Layer Weights to 4-bits:
Quantize the weights of the Linear layer to 4 bits, using symmetric quantization.
Pack two 4-bit weights into one uint8 container.
Choose a quantization scheme (channel-wise or group-wise), with the group size being a multiple of 32.
2. Prepare Quantized Weights, Scales, and Optional Bias:
After quantizing, obtain the quantized_weights, scales, and groupsize.
If the original Linear layer has a bias, prepare it as well.
3. Pack the Weights Efficiently:
Use torch.ops.aten._dyn_quant_pack_4bit_weight to optimally pack the weights, scales, and optional bias.
```python
packed_weights = torch.ops.aten._dyn_quant_pack_4bit_weight(weight, scales_and_zeros, bias, groupsize, in_features, out_features)
```
Input parameters should include:
in_features and out_features (the same as the Linear layer’s corresponding parameters).
4. Perform Dynamic Quantized Matrix Multiplication:
Use torch.ops.aten._dyn_quant_matmul_4bit to perform matrix multiplication with quantized weights.
```python
output = torch.ops.aten._dyn_quant_matmul_4bit(input, packed_weights, groupsize, in_features, out_features)
```
Inputs required include:
The input tensor, packed_weights , groupsize, and the in_features and out_features.
API Usage: https://github.com/pytorch/pytorch/issues/143289
Model Perf :
7B Transformer model:
Prefill : 340 t/s
Decode : 40 t/s
2B Transformer model
Prefill : 747 t/s
Decode : 80 t/s
Tests:
python test/test_linalg.py -k test__dyn_quant_pack_4bit_weight
Ran 1 test in 0.016s
OK
python test/test_linalg.py -k test__dyn_quant_matmul_4bit
Ran 8 tests in 0.077s
OK
python test/test_linalg.py -k test_compile_dyn_quant_matmul_4bit
Ran 8 tests in 11.454s
Change-Id: Ia1672bad5e6ec94e64d8bb1971395d60f4b3a452
Fixes #ISSUE_NUMBER
Pull Request resolved: https://github.com/pytorch/pytorch/pull/134124
Approved by: https://github.com/digantdesai, https://github.com/malfet
Summary:
Early prototype of adding CK backend for aten.bmm. Currently, it is very limited in that:
1. BF16 only
2. A single CK instance
3. NT layout only
4. Alpha=1, Beta=0 only
Reviewed By: xw285cornell, zjing14
Differential Revision: D65954695
Pull Request resolved: https://github.com/pytorch/pytorch/pull/140758
Approved by: https://github.com/bradleyhd
# Motivation
This PR intends to update torch-xpu-ops commit pin. It mainly includes the following two highlighted changes:
1. split the DLL library into 4 smaller libraries to avoid the 2G limitation on Windows;
2. some new operators added, for example, `cdist`, `pdist`, `maxunpool2d`, `maxunpood3d`, `upsample_trilinear3d, `Bessel operators`, etc...
# Additional Context
We have to supply XPU device check logic in `cdist` and `pdist` ops.
This PR depends on https://github.com/pytorch/pytorch/pull/139050 to fix Windows build issue.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/139041
Approved by: https://github.com/EikanWang, https://github.com/ezyang
# Motivation
Fix https://github.com/pytorch/pytorch/issues/138577.
# Solution
1. All UTs in `test/inductor/test_compiled_optimizers.py` are fixed by https://github.com/pytorch/pytorch/pull/134170
2. UT in `test/inductor/test_pattern_matcher.py` is introduced by https://github.com/pytorch/pytorch/pull/138089, we will skip this UT due to the unsupported feature `max_autotune_gemm_backends:Triton`.
3. We have a new impl related to `histc`, so we remove the expected failure from `test/inductor/test_torchinductor_opinfo.py`
4. We support `avg_pool3d` for `fp16` data type, so we remove the expected failure from `test/inductor/test_torchinductor_opinfo.py`
5. CUDA-bias code is introduced by https://github.com/pytorch/pytorch/issues/138472, we just generalize it to `GPU_TYPE`.
# Additional Context
> Why update torch-xpu-ops commit pin here?
We have to update commit pin to avoid the build failure raised by the code change [C10_UNUSED](https://github.com/pytorch/pytorch/pull/138364).
> What does the feature of torch-xpu-ops update?
1. Add some foreach ops, like `unary ops` and `foreach_clamp_max` etc;
2. Add some maxpool ops forward and backward, like `averge_pool3d` and `max_pool3d`
3. Add some other ops, like `log_normal_`, `index_copy`, and `mode` etc;
4. fix build failure related to `C10_UNUSED`;
Pull Request resolved: https://github.com/pytorch/pytorch/pull/138548
Approved by: https://github.com/malfet, https://github.com/EikanWang
Intel GPU aten library(libtorch_xpu) utilizes `torchgen` to generate structure kernels. Currently, the generated structure kernels are decorated by `TORCH_API` to control the visibility, while `TORCH_API` is controlled by the `CAFFE2_BUILD_MAIN_LIB` macro. However, we cannot enable `CAFFE2_BUILD_MAIN_LIB` for the Intel GPU ATen library naively. Because the macro not only serves for the `TORCH_API` semantic. It means that the semantic of `TORCH_API` is symbol `hidden`.
https://github.com/pytorch/pytorch/blob/main/c10/macros/Export.h#L95-L99
Therefore, we need to use ` TORCH_XPU_API` to decorate the produced structure kernels.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/137794
Approved by: https://github.com/atalman
ghstack dependencies: #137873
Summary: Tests Fix Clear On Fork by forking a process after a profile has already been done. Afterwards we check that all the PID/TID are as expected.
Test Plan: Ran buck2 test 'fbcode//mode/dev' fbcode//caffe2/test:profiler -- --exact 'caffe2/test:profiler - test_forked_process (profiler.test_profiler.TestProfiler)'
Differential Revision: D63992036
Pull Request resolved: https://github.com/pytorch/pytorch/pull/137511
Approved by: https://github.com/sanrise, https://github.com/aaronenyeshi
* This fixes a major CMake/Bazel configuration bug where we were leaving CUTLASS performance on the table, especially with FlashAttention. This now enables using MMA instructions on SM90+, which should close the gap between SDPA and the external FA2. Note these operations only affect H100 and newer GPUs. Thankfully, this seems to have been updated recently into being a noop on the CUTLASS side. Still better set the CMake variable properly.
* Also enables additional new shape kernels added in the recent CUTLASS 3.5.1+ update. This was the original motivatin of the PR before I realized the basic MMA kernels were accidentally disabled since we didn't go through the submodule's CMake/Bazels.
* Adds a bit to compile time and code size, but well worth it considering it speeds up our internal flash attention significantly on H100s at the cost of some minor additional compile time.
* These kernels and settings will be needed for Flash Attention 3 whenever we add that too.
Fixes#133695
Pull Request resolved: https://github.com/pytorch/pytorch/pull/133686
Approved by: https://github.com/ezyang
Updates pybind11 submodule. The major patchnote is an experimental new function that is added to all pybind11 objects that will make them more compatible across pybind11 version, settings, and frameworks (such as nanobind) called cpp_conduit. No code changes needed on our end except to update
Pull Request resolved: https://github.com/pytorch/pytorch/pull/136087
Approved by: https://github.com/malfet
