From 56ddd5de175295268ae7880479d03b86ee249e07 Mon Sep 17 00:00:00 2001 From: Siyuan Liu Date: Thu, 6 Jun 2024 21:56:55 -0700 Subject: [PATCH] Add int8 per channel weight-only quantized matmul (#7201) Co-authored-by: Siyuan Liu --- docs/quantized_ops.md | 114 ++++++++++++++++++ test/quantized_ops/test_quantized_matmul.py | 105 ++++++++++++++++ test/run_tests.sh | 1 + .../experimental/xla_quantized_matmul.py | 89 ++++++++++++++ 4 files changed, 309 insertions(+) create mode 100644 docs/quantized_ops.md create mode 100644 test/quantized_ops/test_quantized_matmul.py create mode 100644 torch_xla/experimental/xla_quantized_matmul.py diff --git a/docs/quantized_ops.md b/docs/quantized_ops.md new file mode 100644 index 00000000000..f2345e2bc15 --- /dev/null +++ b/docs/quantized_ops.md @@ -0,0 +1,114 @@ +Quantized Operations for XLA device (Experimental feature) +-------------------------- + +This document outlines how to utilize quantized operations to enable quantization on XLA devices. + +XLA Quantized ops offer a high-level abstraction for quantized operations (e.g., blockwise int4 quantized matrix multiplication). These ops are analogous to quantized CUDA kernels ([example](https://github.com/vllm-project/vllm/blob/main/csrc/quantization/gptq/q_gemm.cu)) in the CUDA ecosystem, providing similar functionality and performance benefits within the XLA framework. + +**NOTE:** Currently this is classified as experimental feature. It's API specifics +will change in the next (2.5) release. + + +## How to use: + +XLA quantized operations can be used as `torch op`, or a `torch.nn.Module` that wraps the `torch.op`. These 2 options give model developers the flexibility to choose the best way to integrate XLA quantized ops into their solution. + +Both `torch op` and `nn.Module` are compatible with `torch.compile( backend='openxla')`. + +### Call XLA quantized op in model code + +Users can call XLA quantized ops in the same way as calling other regular PyTorch ops. This provides maximum flexibility in integrating XLA quantized ops into their applications. The quantized ops work in both eager mode and Dynamo, with regular PyTorch CPU tensor and XLA tensor. + +**Note** Please check the docstring of the quantized ops for the layout of the quantized weights. + +```Python +import torch +import torch_xla.core.xla_model as xm +import torch_xla.experimental.xla_quantized_matmul + +N_INPUT_FEATURES=10 +N_OUTPUT_FEATURES=20 +x = torch.randn((3, N_INPUT_FEATURES), dtype=torch.bfloat16) +w_int = torch.randint(-128, 127, (N_OUTPUT_FEATURES, N_INPUT_FEATURES), dtype=torch.int8) +scaler = torch.randn((N_OUTPUT_FEATURES,), dtype=torch.bfloat16) + +# Call with torch CPU tensor (For debugging purpose) +matmul_output = torch.ops.xla.quantized_matmul(x, w_int, scaler) + +device = xm.xla_device() +x_xla = x.to(device) +w_int_xla = w_int.to(device) +scaler_xla = scaler.to(device) + +# Call with XLA Tensor to run on XLA device +matmul_output_xla = torch.ops.xla.quantized_matmul(x_xla, w_int_xla, scaler_xla) + +# Use with torch.compile(backend='openxla') +def f(x, w, s): + return torch.ops.xla.quantized_matmul(x, w, s) + +f_dynamo = torch.compile(f, backend="openxla") +dynamo_out_xla = f_dynamo(x_xla, w_int_xla, scaler_xla) +``` + +It's common to wrap the quantized op into a custom `nn.Module` in model developers model code: + +```Python +class MyQLinearForXLABackend(torch.nn.Module): + def __init__(self): + self.weight = ... + self.scaler = ... + + def load_weight(self, w, scaler): + # Load quantized Linear weights + # Customized way to preprocess the weights + ... + self.weight = processed_w + self.scaler = processed_scaler + + + def forward(self, x): + # Do some random stuff with x + ... + matmul_output = torch.ops.xla.quantized_matmul(x, self.weight, self.scaler) + # Do some random stuff with matmul_output + ... +``` + +### Module Swap + +Alternatively, users can also use the `nn.Module` that wraps the XLA quantized ops and do module swap in the model code: + +```Python +orig_model = MyModel() +# Quantize the model and get quantized weights +q_weights = quantize(orig_model) +# Process the quantized weight to the format that XLA quantized op expects. +q_weights_for_xla = process_for_xla(q_weights) + +# Do module swap +q_linear = XlaQuantizedLinear(self.linear.in_features, + self.linear.out_features) +q_linear.load_quantized_weight(q_weights_for_xla) +orig_model.linear = q_linear +``` + +## Supported Quantized Operations: + +### Matrix Multiply + +| Weight Quantization Type | Activation Quantization Type | Dtype | Supported | +|---|---|---|---| +| per-channel | N/A | W8A16 | Yes | +| per-channel | N/A | W4A16 | No | +| per-channel | per-token | W8A8 | No | +| per-channel | per-token | W4A8 | No | +| blockwise | N/A | W8A16 | No | +| blockwise | N/A | W4A16 | No | +| blockwise | per-token | W8A8 | No | +| blockwise | per-token | W4A8 | No | + +**Note** `W[X]A[Y]` refers to Weight in `X`-bit, Activation in `Y`-bit. If `X/Y` is 4 or 8, it refers to `int4/8`. 16 for `bfloat16` format. + +### Embedding +To be added diff --git a/test/quantized_ops/test_quantized_matmul.py b/test/quantized_ops/test_quantized_matmul.py new file mode 100644 index 00000000000..e0bcb8b1f32 --- /dev/null +++ b/test/quantized_ops/test_quantized_matmul.py @@ -0,0 +1,105 @@ +import re +import unittest + +import torch +import torch_xla +import torch_xla.core.xla_model as xm +import torch_xla.experimental.xla_quantized_matmul +from torch_xla.experimental.xla_quantized_matmul import XlaQuantizedLinear +from torch.ao.quantization.utils import determine_qparams + +torch.manual_seed(123456) + +device = xm.xla_device() + + +class M(torch.nn.Module): + + def __init__(self, input_dim, output_dim): + super(M, self).__init__() + # Define a linear layer + self.linear = torch.nn.Linear(input_dim, output_dim, bias=False) + + def weight_quantization_rtn(self, + linear, + quant_method=torch.per_channel_symmetric): + ''' + Quantize linear weight using Round-To-Nearest(RTN) algorithm. + ''' + assert isinstance(self.linear, torch.nn.Linear) + w_fp = linear.weight.data + min_val, max_val = torch.aminmax(w_fp, dim=1) # min_val, max_val [out_dim] + n_bits = 8 + int_min = -2**(n_bits - 1) + int_max = 2**(n_bits - 1) - 1 + scaler, zero_point = determine_qparams( + min_val, + max_val, + int_min, + int_max, + dtype=torch.int8, + eps=torch.Tensor([1e-5]), + has_customized_qrange=False, + qscheme=quant_method) + w_int = torch.ops.quantized_decomposed.quantize_per_channel( + w_fp, scaler, zero_point, 0, int_min, int_max, torch.int8) + return w_int, scaler.to(w_fp.dtype), zero_point + + def replace_with_xla_quantized_matmul(self): + assert isinstance(self.linear, torch.nn.Linear) + w_int, scaler, _ = self.weight_quantization_rtn(self.linear) + q_linear = XlaQuantizedLinear(self.linear.in_features, + self.linear.out_features) + q_linear.load_quantized_weight(w_int, scaler) + self.linear = q_linear + + def forward(self, x): + # Forward pass through the linear layer + return self.linear(x) + + +class QuantizedTest(unittest.TestCase): + + def test_q_linear_module_per_channel(self): + + with torch.no_grad(): + m = M(5, 8) + x = torch.randn(3, 5) + out_fp = m(x) + m.replace_with_xla_quantized_matmul() + out_quant = m(x) + + m = m.to(device) + x = x.to(device) + out_quant_xla = m(x) + self.assertTrue(torch.allclose(out_fp, out_quant, atol=0.01)) + self.assertTrue(torch.allclose(out_quant_xla.cpu(), out_quant)) + + def test_q_linear_module_dynamo(self): + + with torch.no_grad(): + m = M(5, 8) + x = torch.randn(3, 5) + out_fp = m(x) + m.replace_with_xla_quantized_matmul() + out_quant = m(x) + m = m.to(device) + m_dynamo = torch.compile(m, backend="openxla") + out_quant_dynamo = m_dynamo(x.to(device)) + self.assertTrue(torch.allclose(out_fp, out_quant, atol=0.01)) + self.assertTrue(torch.allclose(out_quant_dynamo.cpu(), out_quant)) + + def test_q_linear_hlo(self): + with torch.no_grad(): + x = torch.randn((3, 5), dtype=torch.bfloat16).to(device) + w_int = torch.randint(-128, 127, (8, 5), dtype=torch.int8).to(device) + scaler = torch.randn((8,), dtype=torch.bfloat16).to(device) + + output = torch.ops.xla.quantized_matmul(x, w_int, scaler) + hlo = torch_xla._XLAC._get_xla_tensors_hlo([output]) + print(hlo) + self.assertTrue(re.search(r'bf16.*dot.*bf16.*s8', hlo) is not None) + + +if __name__ == '__main__': + unittest.main() diff --git a/test/run_tests.sh b/test/run_tests.sh index 409c36c9c27..531a9b0c8a9 100755 --- a/test/run_tests.sh +++ b/test/run_tests.sh @@ -221,6 +221,7 @@ function run_xla_op_tests3 { run_xla_hlo_debug "$CDIR/stablehlo/test_stablehlo_inference.py" run_test "$CDIR/stablehlo/test_stablehlo_compile.py" run_test "$CDIR/stablehlo/test_unbounded_dynamism.py" + run_test "$CDIR/quantized_ops/test_quantized_matmul.py" run_test "$CDIR/spmd/test_xla_sharding.py" run_test "$CDIR/spmd/test_xla_sharding_hlo.py" run_test "$CDIR/spmd/test_xla_virtual_device.py" diff --git a/torch_xla/experimental/xla_quantized_matmul.py b/torch_xla/experimental/xla_quantized_matmul.py new file mode 100644 index 00000000000..ef6734de7c5 --- /dev/null +++ b/torch_xla/experimental/xla_quantized_matmul.py @@ -0,0 +1,89 @@ +import torch +import torch.nn.functional as F +import torch_xla +from torch.library import impl +from torch_xla.core.xla_model import XLA_LIB + +XLA_LIB.define( + "quantized_matmul(Tensor x, Tensor w, Tensor scale, int? blocksize=-1, bool? quantize_activation=False) -> Tensor" +) + + +def _check_per_channel_quant_weight_dtype_shapes(input_dim, output_dim, w, + w_scaler): + assert w.dtype == torch.int8, f"Weight dtype is expected to be torch.int8, got {w.dtype}." + assert w.dim( + ) == 2, f"Weight tensor is expected to be 2D, got {w.dim()}D Tensor." + assert output_dim == w.shape[0] and input_dim == w.shape[ + 1], f"Weight shape is expected to be [output_dim, input_dim], output_dim: {output_dim}, input_dim: {input_dim}, but got {w.shape}." + assert w_scaler.dim() == 1 and w_scaler.shape[0] == w.shape[ + 0], f"weight scaler shape is expect to be [out_channel,], got {w_scaler.shape}, weight shape {w.shape}." + + +@impl(XLA_LIB, "quantized_matmul", "XLA") +def quantized_matmul_xla(x: torch.Tensor, + w: torch.Tensor, + scaler: torch.Tensor, + blocksize: int = -1): + """Quantized Matrix Multiply op on XLA devices. + + Args: + x: torch.Tensor - Activation of Matmul [..., in_channel]. + w: torch.Tensor - Weight Tensor. + per-channel quant: torch.int8 x [out_channel, in_channel]. + scaler: torch.Tensor - Weight scaler. + per-channel quant: [out_channel,]. + blocksize: blocksize for blockwise quantization, -1 for per-channel quantization. + """ + assert blocksize == -1, "blockwise quantization is not supported yet." + # Per-channel quant. + _check_per_channel_quant_weight_dtype_shapes(x.shape[-1], scaler.shape[0], w, + scaler) + return F.linear(x, w) * scaler + + +@impl(XLA_LIB, "quantized_matmul", "CompositeExplicitAutograd") +def quantized_matmul(x: torch.Tensor, + w: torch.Tensor, + scaler: torch.Tensor, + blocksize: int = -1): + assert blocksize == -1, "blockwise quantization is not supported yet." + # Per-channel quant. + _check_per_channel_quant_weight_dtype_shapes(x.shape[-1], scaler.shape[0], w, + scaler) + w = w.to(x.dtype) + return torch.mul(F.linear(x, w), scaler) + + +class XlaQuantizedLinear(torch.nn.Module): + + def __init__(self, input_dim, output_dim, blocksize=-1): + super().__init__() + assert blocksize == -1, "Only per-channel quantization is supported." + self.input_dim = input_dim + self.output_dim = output_dim + self.blocksize = blocksize + self.register_buffer('weight', + torch.zeros(output_dim, input_dim).to(torch.int8)) + self.register_buffer('weight_scaler', torch.zeros(output_dim)) + + def load_quantized_weight(self, weight, weight_scaler): + ''' + Weight shape: [output_channel, input_channel] + Weight scaler shape: [output_channel] + ''' + if self.blocksize == -1: + # Per-channel quant. + _check_per_channel_quant_weight_dtype_shapes(self.input_dim, + self.output_dim, weight, + weight_scaler) + self.weight = weight + self.weight_scaler = weight_scaler + else: + assert False, "Only per-channel quantization is supported." + + def forward(self, x): + if self.blocksize == -1: + return torch.ops.xla.quantized_matmul(x, self.weight, self.weight_scaler) + else: + assert False, "Only per-channel quantization is supported."