Add HQQ support

torchao/dtypes/affine_quantized_tensor.py

@@ -2,13 +2,15 @@
 from typing import Dict, Callable, Any, Tuple, Optional
 from collections import defaultdict
 import functools
+import math
 from torchao.quantization.quant_primitives import (
     choose_qparams_affine,
     quantize_affine,
     dequantize_affine,
     ZeroPointDomain,
     MappingType,
     int_scaled_matmul,
+    quantize_affine_hqq,
 )
 from torchao.quantization.utils import (
     pack_tinygemm_scales_and_zeros,
@@ -25,6 +27,7 @@
     PlainLayoutType,
     is_device,
 )
+
 from dataclasses import dataclass
 from torchao.utils import TORCH_VERSION_AFTER_2_5
 
@@ -75,7 +78,6 @@ def _get_to_kwargs(self, *args, **kwargs):
 ##############################
 # Tensor Subclass Definition #
 ##############################
-
 class AffineQuantizedTensor(torch.Tensor):
     """
     Affine quantized tensor subclass. Affine quantization means we quantize the floating point tensor with an affine transformation:
@@ -190,14 +192,23 @@ def from_float(
         preserve_zero: bool = True,
         zero_point_domain: ZeroPointDomain = ZeroPointDomain.INT,
         layout_type: LayoutType = PlainLayoutType(),
+        use_hqq: bool = False,
     ):
         original_shape = input_float.shape
-        input_float = layout_type.pre_process(input_float)
 
-        scale, zero_point = choose_qparams_affine(input_float, mapping_type, block_size, target_dtype, quant_min, quant_max, eps, scale_dtype, zero_point_dtype, preserve_zero, zero_point_domain)
-        int_data = quantize_affine(input_float, block_size, scale, zero_point, target_dtype, quant_min, quant_max, zero_point_domain)
-        int_data = layout_type.post_process(int_data)
+        if(use_hqq):
+            assert zero_point_domain == ZeroPointDomain.FLOAT and mapping_type == MappingType.ASYMMETRIC and quant_min==0, "Invalid input parameters for HQQ quantization."
+            nbits = int(math.log2(quant_max + 1))
+            axis  = 1 if (block_size[0]==1) else 0
+            group_size = max(block_size)
+            int_data, scale, zero_point, _ = quantize_affine_hqq(input_float, nbits=nbits, group_size=group_size, axis=axis, compute_dtype=input_float.dtype, device=input_float.device, verbose=False, raw_output=False)
 
+        else:
+            input_float = layout_type.pre_process(input_float)
+            scale, zero_point = choose_qparams_affine(input_float, mapping_type, block_size, target_dtype, quant_min, quant_max, eps, scale_dtype, zero_point_dtype, preserve_zero, zero_point_domain)
+            int_data = quantize_affine(input_float, block_size, scale, zero_point, target_dtype, quant_min, quant_max, zero_point_domain)
+            int_data = layout_type.post_process(int_data)
+
         layout_tensor_ctr = get_layout_tensor_constructor(type(layout_type))
         layout_tensor = layout_tensor_ctr(int_data, scale, zero_point, layout_type)
         return cls(

torchao/prototype/hqq/core.py

@@ -0,0 +1,184 @@
+import torch
+import math
+from torch import Tensor, float16, float32
+from typing import Union
+
+
+# Shrinking operator (proximal operator for the lp norm)
+def shrink_lp_op(x: Tensor, beta: float, lp_norm: float) -> Tensor:
+    if lp_norm == 1:
+        return torch.sign(x) * torch.nn.functional.relu(torch.abs(x) - 1.0 / beta)
+    else:
+        return torch.sign(x) * torch.nn.functional.relu(
+            torch.abs(x) - (1.0 / beta) * torch.pow(torch.abs(x), lp_norm - 1)
+        )
+
+
+# Proximal solver || W - dequantize(quantize(W))||_p^p
+@torch.inference_mode()
+def optimize_weights_proximal_legacy(
+    tensor: Tensor,
+    scale: Tensor,
+    zero: Tensor,
+    min_max: list,
+    axis: int = 0,
+    dtype: Union[torch.dtype, None] = None,
+    device: Union[str, None] = None,
+    verbose: bool = False,
+    opt_params: dict = {
+        "lp_norm": 0.7,
+        "beta": 1e1,
+        "kappa": 1.01,
+        "iters": 20,
+        "early_stop": True,
+    },
+) -> tuple:
+    lp_norm, beta, kappa, iters, early_stop = (
+        opt_params["lp_norm"],
+        opt_params["beta"],
+        opt_params["kappa"],
+        opt_params["iters"],
+        opt_params["early_stop"],
+    )
+
+    device = tensor.device if (device is None) else torch.device(device)
+
+    if dtype is None:
+        dtype = float16 if (device.type == "cuda") else float32
+
+    W_f = tensor.to(dtype=dtype, device=device)
+    scale = scale.to(dtype=dtype, device=device)
+    zero = zero.to(dtype=dtype, device=device)
+
+    best_error = 1e4
+    for i in range(iters):
+        W_q = torch.round(W_f * scale + zero).clamp(min_max[0], min_max[1])
+        W_r = (W_q - zero) / scale
+        W_e = shrink_lp_op(W_f - W_r, beta, lp_norm)
+        zero = torch.mean(W_q - (W_f - W_e) * scale, axis=axis, keepdim=True)
+        beta *= kappa
+
+        current_error = float(torch.abs(W_f - W_r).mean())
+        if verbose:
+            print("Iter " + str(i + 1), " | Error: " + str(current_error))
+        if early_stop:
+            if current_error < best_error:
+                best_error = current_error
+            else:
+                break
+
+    scale = scale.to(tensor.device)
+    zero = zero.to(tensor.device)
+    del W_f, W_q, W_r, W_e
+    torch.cuda.empty_cache()
+
+    W_q = torch.round(tensor * scale + zero).clamp(min_max[0], min_max[1])
+    return W_q, scale, zero
+
+
+# Default: fast with early stopping
+optimize_weights_proximal = optimize_weights_proximal_legacy
+
+
+# Mainly used to check if the group-size is divisible by numel()
+def is_divisible(val1: int, val2: int) -> bool:
+    return int(val2 * math.ceil(val1 / val2)) == val1
+
+
+# Converts hqq format W_dequant = (W_q - zero)*scale into affinequantized format: (W_q - mid_point)*scale_ao + zero_ao
+def convert_to_affinequantized_format(W_q, scale, zero, nbits, shape):
+    quant_min = 0
+    quant_max = 2**nbits - 1
+    mid_point = (quant_max + quant_min + 1) / 2
+    zero_ao = ((mid_point - zero.float()) * scale.float()).to(zero.dtype)
+    scale_ao = scale
+    W_q_ao = W_q.view(shape)
+    return W_q_ao, scale_ao, zero_ao
+
+
+# Main HQQ Quantizer - simplified, no bitpacking.
+class HQQQuantizer:
+    optimize_weights = optimize_weights_proximal
+
+    @classmethod
+    def quantize(
+        cls,
+        tensor: Tensor,
+        nbits: float = 4,
+        group_size: int = 64,
+        optimize: bool = True,
+        axis: int = 1,
+        compute_dtype: torch.dtype = float16,
+        device: str = "cuda",
+        verbose: bool = False,  # to check the optimizer error
+        raw_output: bool = False,  # If True, it will return the quant params in hqq lib format
+    ) -> tuple:
+        assert axis in [0, 1], "axis should be either 0 or 1"
+        if group_size is not None:
+            assert is_divisible(tensor.numel(), group_size), (
+                "group_size should be divisble by the total tensor dimensions. shape: "
+                + str(tensor.shape)
+                + ", group_size: "
+                + str(group_size)
+            )
+
+        W = tensor.to(device=device, dtype=torch.float32)
+        shape = W.shape
+
+        # Reshape for grouping
+        if group_size is not None:
+            W = (
+                W.reshape([-1, group_size])
+                if (axis == 1)
+                else W.reshape([group_size, -1])
+            )
+
+        # Get min/max values
+        _min = W.min(axis=axis, keepdim=True)[0]
+        _max = W.max(axis=axis, keepdim=True)[0]
+
+        max_v = round(2**nbits - 1)
+        min_v = 0
+        min_max = [min_v, max_v]
+
+        # Clamp to avoid fp16 issues
+        scale = (max_v / (_max - _min)).clamp(max=2e4)
+        zero = -_min * scale
+
+        # Round zero as in: https://github.com/casper-hansen/AutoAWQ/blob/main/awq/quantize/quantizer.py#L42C9-L42C14
+        if nbits in [4]:
+            zero = torch.round(zero)
+
+        # Fine-tune weights
+        if optimize:
+            W_q, scale, zero = HQQQuantizer.optimize_weights(
+                tensor=W,
+                scale=scale,
+                zero=zero,
+                min_max=min_max,
+                axis=axis,
+                device=device,
+                verbose=verbose,
+            )
+        else:
+            W_q = torch.round(W * scale + zero).clamp(min_max[0], min_max[1])
+
+        # Store meta-data (we invert the scale for dequantization)
+        scale = 1.0 / scale
+
+        # Convert to affienquantized format
+        if raw_output is False:
+            W_q, scale, zero = convert_to_affinequantized_format(
+                W_q, scale, zero, nbits, shape
+            )
+
+        # Make sure all the weights are in the right compute_dtype/device
+        W_q = W_q.to(dtype=torch.uint8, device=device)
+        scale = scale.to(dtype=compute_dtype, device=device)
+        zero = zero.to(dtype=compute_dtype, device=device)
+
+        # cleanup
+        del W, _min, _max
+        torch.cuda.empty_cache()
+
+        return W_q, scale, zero, shape

torchao/prototype/hqq/example.py

@@ -0,0 +1,62 @@
+import torch
+from torchao.prototype.hqq.core import HQQQuantizer
+from torchao.dtypes.affine_quantized_tensor import (
+    AffineQuantizedTensor,
+    ZeroPointDomain,
+    PlainAQTLayout,
+    PlainLayoutType,
+    TensorCoreTiledAQTLayout,
+    TensorCoreTiledLayoutType,
+    MappingType,
+)
+
+#Parameters
+device, compute_dtype = "cuda:0", torch.bfloat16
+group_size, axis = 64, 1
+
+linear_layer = torch.nn.Linear(4096, 11800, bias=False, device=device)
+x = torch.randn((1, linear_layer.in_features), dtype=torch.float, device=device)/20.
+y_ref = linear_layer(x)
+W = linear_layer.weight.data.clone().to(device=device, dtype=compute_dtype)
+del linear_layer.weight 
+################################################################################################
+
+for nbits in list(range(2, 9))[::-1]:
+    print('------------------------------------------------------------------------------')
+    q_tensor_default = AffineQuantizedTensor.from_float(
+            input_float=W,
+            mapping_type=MappingType.ASYMMETRIC,
+            block_size=[1, group_size],
+            target_dtype=torch.uint8,
+            quant_min=0,
+            quant_max=2**nbits - 1,
+            preserve_zero=False,#Important
+            zero_point_domain= ZeroPointDomain.FLOAT,
+            layout_type=PlainLayoutType(),
+            )
+
+    linear_layer.weight = q_tensor_default
+    print("nbits", nbits, "| Default dequantization error", (W - q_tensor_default.dequantize()).abs().mean().item())
+    print("nbits", nbits, '| Default Dot product error', (y_ref - linear_layer(x.to(compute_dtype))).abs().mean().item())
+    # 4-bit Default dequantization error 0.001953125
+    # 4-bit Default Dot product error 0.0057801781222224236
+
+
+    q_tensor_hqq = AffineQuantizedTensor.from_float(
+            input_float=W,
+            mapping_type=MappingType.ASYMMETRIC,
+            block_size=[1, group_size],
+            target_dtype=torch.uint8,
+            quant_min=0,
+            quant_max=2**nbits - 1,
+            preserve_zero=False,#Important
+            zero_point_domain= ZeroPointDomain.FLOAT,
+            layout_type=PlainLayoutType(),
+            use_hqq=True,
+            )
+
+    linear_layer.weight = q_tensor_hqq
+    print("nbits", nbits, "| HQQ dequantization error", (W - q_tensor_hqq.dequantize()).abs().mean().item())
+    print("nbits", nbits, '| HQQ Dot product error', (y_ref - linear_layer(x.to(compute_dtype))).abs().mean().item())
+    # 4-bit HQQ dequantization error 0.0004863739013671875
+    # 4-bit HQQ Dot product error 0.0014263123739510775