[Float8Quant] Add rowwise scaling option to float8 dyanmic quant

stack-info: PR: #819, branch: drisspg/stack/11

Author: drisspg

ruff.toml

@@ -8,4 +8,5 @@ include = [
     "torchao/dtypes/nf4tensor.py",
     "test/dtypes/test_nf4.py",
     "torchao/float8/float8_tensor.py",
+    "test/dtypes/test_affine_quantized_float.py"
 ]

test/dtypes/test_affine_quantized_float.py

@@ -1,34 +1,28 @@
 from torchao.utils import (
     TORCH_VERSION_AT_LEAST_2_5,
-    unwrap_tensor_subclass,
 )
 import pytest
 
 if not TORCH_VERSION_AT_LEAST_2_5:
     pytest.skip("Unsupported PyTorch version", allow_module_level=True)
 
-from numpy import full
-from torch.testing._internal.common_utils import (
-    run_tests,
-)
 from torch._inductor.test_case import TestCase as InductorTestCase
 from torch.testing._internal import common_utils
-from torch._dynamo.testing import CompileCounterWithBackend
 
 from torchao.quantization import (
     quantize_,
     float8_weight_only,
     float8_dynamic_activation_float8_weight,
 )
+from torchao.quantization.observer import PerTensor, RowWise
 from torchao.float8.float8_utils import compute_error
 import torch
 import unittest
 import pytest
-import tempfile
 import copy
 import random
-
-from unittest.mock import patch
+from functools import partial
+from typing import Tuple
 
 
 random.seed(0)
@@ -56,6 +50,7 @@ class TestAffineQuantizedFloat8Compile(InductorTestCase):
     @common_utils.parametrize("dtype", [torch.bfloat16, torch.float32])
     @common_utils.parametrize("mode", ["dynamic", "weight-only"])
     @common_utils.parametrize("compile", [True, False])
+    @common_utils.parametrize("granularity", [PerTensor, RowWise])
     # Inputs are (M,..), K, N
     @common_utils.parametrize(
         "sizes",
@@ -68,13 +63,20 @@ class TestAffineQuantizedFloat8Compile(InductorTestCase):
         ],
     )
     def test_fp8_linear_variants(
-        self, dtype: torch.dtype, mode: str, compile: bool, sizes: tuple
+        self, dtype: torch.dtype, mode: str, compile: bool, sizes: Tuple, granularity
     ):
+        if granularity == RowWise and mode == "dynamic":
+            pytest.skip(
+                "RowWise quantization only works for bfloat16 precision input weight and activation for now"
+            )
+
         M, N, K = sizes
         input_tensor = torch.randn(*M, K, dtype=dtype, device="cuda")
 
         mode_map = {
-            "dynamic": float8_dynamic_activation_float8_weight,
+            "dynamic": partial(
+                float8_dynamic_activation_float8_weight, granularity=granularity
+            ),
             "weight-only": float8_weight_only,
         }
 

torchao/dtypes/affine_quantized_tensor.py

@@ -1161,21 +1161,32 @@ def _linear_f16_act_fpx_weight_impl(input_tensor, weight_tensor, bias):
 
     return out.view(*act.shape[:-1], out_dim).to(act.dtype)
 
-def _linear_fp_act_fp8_tensor_wise_weight_check(
+def _linear_fp_act_fp8_weight_check(
     input_tensor: Union[torch.Tensor, AffineQuantizedTensor],
     weight_tensor: Union[torch.Tensor, AffineQuantizedTensor],
     bias: Optional[torch.Tensor],
 ) -> bool:
-    def check_aqt_tensorwise(aqt: Union[torch.Tensor, AffineQuantizedTensor]) -> bool:
+    def check_aqt(aqt: Union[torch.Tensor, AffineQuantizedTensor]) -> bool:
+        row_block_size = (1,) * (aqt.dim() - 1) + (aqt.shape[-1],)
         return (
             isinstance(aqt, AffineQuantizedTensor) and
             isinstance(aqt.layout_type, Float8LayoutType)
             and aqt.layout_tensor.dtype in [torch.float8_e4m3fn, torch.float8_e5m2]
-            and aqt.shape == aqt.block_size
+            and (aqt.shape == aqt.block_size or aqt.block_size == row_block_size)
         )
-    return check_aqt_tensorwise(input_tensor) and check_aqt_tensorwise(weight_tensor)
+    return check_aqt(input_tensor) and check_aqt(weight_tensor)
 
 
+def preprocess_input_tensor(inpt_data: torch.Tensor, input_scale: torch.Tensor, input_shape: Tuple[int]):
+    """ Ensures input tensor is correctly formated for _scaled_mm """
+    if input_scale.size(0) != 1:
+        input_scale = input_scale.unsqueeze(-1)
+    
+    if input_scale.dim() > 2:
+        input_scale = input_scale.reshape(-1, input_scale.shape[-1])
+    
+    return inpt_data, input_scale
+
 def _linear_fp_act_fp8_weight_impl(
     input_tensor: AffineQuantizedTensor,
     weight_tensor: AffineQuantizedTensor,
@@ -1186,28 +1197,34 @@ def _linear_fp_act_fp8_weight_impl(
         preprocess_data,
         Float8MMConfig,
         addmm_float8_unwrapped_inference,
+        _is_rowwise_scaled
     )
 
     scaled_mm_config = weight_tensor.layout_type.mm_config
-    scaled_mm_config = scaled_mm_config if scaled_mm_config is not None else Float8MMConfig()
+    out_shape = get_out_shape(input_tensor.shape, weight_tensor.shape)
 
+    # Weight tensor preprocessing
     w_layout = weight_tensor.layout_tensor
-    w_data = weight_tensor.layout_tensor.float8_data
-    w_data = w_data.T if w_layout.transposed else w_data
+    assert not w_layout.transposed, "Weight tensor must be contiguous"
+    w_data = w_layout.float8_data
     w_scale = w_layout.scale
-    w_scale = w_scale if w_layout.transposed else w_scale
-
-    out_shape = get_out_shape(input_tensor.shape, weight_tensor.shape)
 
+    # Input tensor preprocessing
     inpt_data = input_tensor.layout_tensor.float8_data
-    # Handle case where input tensor is more than 2D
-    inpt_data = inpt_data.reshape(-1, input_tensor.shape[-1])
     input_scale = input_tensor.layout_tensor.scale
-    if input_scale.dim() > 2:
-        input_scale = input_scale.reshape(-1, input_scale.shape[-1])
+    # Handle case where input tensor is more than 2D
+    inpt_data = inpt_data.reshape(-1, inpt_data.shape[-1])
+
+    # Handle rowwise case
+    if _is_rowwise_scaled(weight_tensor):
+        assert _is_rowwise_scaled(input_tensor), "Input tensor must be rowwise block size"
+        w_scale = w_scale.unsqueeze(-1).T
+        inpt_data, input_scale = preprocess_input_tensor(inpt_data, input_scale, input_tensor.shape)
 
+    # Preprocess data
     inpt_data, w_data = preprocess_data(inpt_data, w_data.T, scaled_mm_config)
 
+    # Perform the computation
     return addmm_float8_unwrapped_inference(
         inpt_data,
         input_scale,
@@ -1223,7 +1240,7 @@ def _register_aqt_quantized_linear_dispatches():
     for dispatch_condition, impl in [
         (_linear_int8_act_int8_weight_check, _linear_int8_act_int8_weight_impl),
         (_linear_int8_act_int8_weight_semi_structured_sparse_check, _linear_int8_act_int8_weight_semi_structured_sparse_impl),
-        (_linear_fp_act_fp8_tensor_wise_weight_check, _linear_fp_act_fp8_weight_impl),
+        (_linear_fp_act_fp8_weight_check, _linear_fp_act_fp8_weight_impl),
         (_linear_bf16_act_uint4_weight_check, _linear_bf16_act_uint4_weight_impl),
         (_linear_fp_act_int8_weight_check, _linear_fp_act_int8_weight_impl),
         (_linear_f16_act_fpx_weight_check, _linear_f16_act_fpx_weight_impl),

torchao/float8/inference.py

@@ -97,3 +97,11 @@ def addmm_float8_unwrapped_inference(
         use_fast_accum=use_fast_accum,
     )
     return output
+
+
+def _is_rowwise_scaled(x) -> bool:
+    """Checks if an AQT tensor is rowwise scaled
+    Args:
+        x: AffineQuantizedTensor tensor
+    """
+    return x.block_size == (1,) * (x.dim() - 1) + (x.shape[-1],)

torchao/quantization/observer.py

@@ -52,6 +52,17 @@ class PerAxis(GranularityType):
     """
     axis: int
 
+@dataclass(frozen=True)
+class RowWise(GranularityType):
+    """
+    Represents row-wise granularity in quantization.
+
+    This is a special case of per-axis quantization and is unique to Float8 matmuls
+    where the input is quantized with a block_size of (1, input.shape[1]). And the weight
+    is quantized with a block_size of (weight.shape[0], 1).
+    """
+    pass
+
 # borrowed from torch.ao.quantization.observer
 class _PartialWrapper:
     def __init__(self, p):

torchao/quantization/quant_api.py

@@ -19,7 +19,7 @@
 import torchao
 import torch.nn as nn
 import torch.nn.functional as F
-from typing import Any, Callable, Union, Dict, Optional
+from typing import Any, Callable, Union, Dict, Optional, Literal
 import types
 
 from torchao.dtypes.uintx.Uintx import UintxLayoutType
@@ -60,6 +60,8 @@
 from .autoquant import autoquant, AutoQuantizableLinearWeight
 from torchao.float8.inference import Float8MMConfig
 
+from torchao.quantization.observer import GranularityType, PerTensor, PerAxis, RowWise
+
 logger = logging.getLogger(__name__)
 
 __all__ = [
@@ -550,6 +552,7 @@ def apply_float8wo_quant(weight):
 def float8_dynamic_activation_float8_weight(
     activation_dtype: torch.dtype = torch.float8_e4m3fn,
     weight_dtype: torch.dtype = torch.float8_e4m3fn,
+    granularity: Literal[PerTensor, RowWise] = PerTensor,
     mm_config: Optional[Float8MMConfig] = None
 ):
     """
@@ -566,20 +569,35 @@ def float8_dynamic_activation_float8_weight(
     if mm_config is None:
         mm_config = Float8MMConfig(use_fast_accum=True)
 
+    def get_block_size(x: torch.Tensor, granularity: Literal[PerTensor, RowWise]):
+        if granularity == PerTensor:
+            return x.shape
+        elif granularity == RowWise:
+            return (1,) * (x.dim() - 1) + (x.shape[-1],)
+        else:
+            raise ValueError(f"Unsupported granularity: {granularity}")
+
     #TODO we are hardcoding TensorWise scaling, will follow up PR for Tensorwise scaling
     def apply_float8_dynamic_activation_quant(weight: torch.Tensor):
+        if granularity == RowWise:
+            assert weight.dtype == torch.bfloat16, "RowWise quantization only works for bfloat16 precision input weight and activation"
+
+        block_size = get_block_size(weight, granularity)
         quantized_weight = to_affine_quantized_floatx(
             input_float=weight,
-            block_size=weight.shape,
+            block_size=block_size,
             target_dtype=weight_dtype,
             scale_dtype=torch.float32,
             layout_type=Float8LayoutType(mm_config=mm_config),
         )
 
         def input_quant_func(x: torch.Tensor):
+            if granularity == RowWise:
+                assert weight.dtype == torch.bfloat16, "RowWise quantization only works for bfloat16 precision input weight and activation"
+            block_size = get_block_size(x, granularity)
             activation = to_affine_quantized_floatx(
                 input_float=x,
-                block_size=x.shape,
+                block_size=block_size,
                 target_dtype=activation_dtype,
                 scale_dtype=torch.float32,
                 layout_type=Float8LayoutType(mm_config=None),  # Config is stored on weight