chore: review

Author: Diogo-V

test/sparsity/test_marlin.py

@@ -4,6 +4,7 @@
 
 from torch import nn
 from torch.testing._internal.common_utils import TestCase, run_tests
+from torchao.utils import TORCH_VERSION_AT_LEAST_2_5, unwrap_tensor_subclass
 from torchao.dtypes import MarlinSparseLayoutType
 from torchao.sparsity.sparse_api import apply_fake_sparsity
 from torchao.quantization.quant_api import int4_weight_only, quantize_
@@ -12,20 +13,22 @@
     unpack_from_marlin_24,
     inject_24
 )
-from torchao.quantization.utils import (
-    get_group_qparams_symmetric,
-    groupwise_affine_quantize_tensor_from_qparams,
+from torchao.quantization.quant_primitives import (
+    choose_qparams_affine,
+    quantize_affine,
+    ZeroPointDomain,
+    MappingType,
 )
 
 
 class SparseMarlin24(TestCase):
 
-    @pytest.mark.skipif(not torch.cuda.is_available(), reason="Need CUDA available")
-    def test_quant_sparse_marlin_layout_eager(self):
+    def setUp(self):
+        super().setUp()
         torch.manual_seed(0)
 
-        input = torch.randn((32, 16, 4096), dtype=torch.float16, device="cuda")
-        model = (
+        self.input = torch.randn((32, 16, 4096), dtype=torch.float16, device="cuda")
+        self.model = (
             nn.Sequential(
                 nn.Linear(4096, 21504),
                 nn.Linear(21504, 4096),
@@ -37,48 +40,38 @@ def test_quant_sparse_marlin_layout_eager(self):
             .cuda()
         )
 
-        apply_fake_sparsity(model)
-        model_copy = copy.deepcopy(model)
+    @pytest.mark.skipif(not torch.cuda.is_available(), reason="Need CUDA available")
+    def test_quant_sparse_marlin_layout_eager(self):
+        apply_fake_sparsity(self.model)
+        model_copy = copy.deepcopy(self.model)
 
         # Quantized
         quantize_(model_copy.bfloat16(), int4_weight_only())
-        dense_result = model_copy(input.bfloat16()).half()
+        dense_result = model_copy(self.input.bfloat16()).half()
 
         # Sparse + quantized
-        quantize_(model, int4_weight_only(layout_type=MarlinSparseLayoutType()))
-        sparse_result = model(input)
+        quantize_(self.model, int4_weight_only(layout_type=MarlinSparseLayoutType()))
+        sparse_result = self.model(self.input)
 
         assert torch.allclose(dense_result, sparse_result, atol=3e-1), "Results are not close"
 
     @pytest.mark.skipif(not torch.cuda.is_available(), reason="Need CUDA available")
     def test_quant_sparse_marlin_layout_compile(self):
-        torch.manual_seed(0)
-
-        input = torch.randn((32, 16, 4096), dtype=torch.float16, device="cuda")
-        model = (
-            nn.Sequential(
-                nn.Linear(4096, 21504),
-                nn.Linear(21504, 4096),
-                nn.ReLU(),
-                nn.Linear(4096, 21504),
-                nn.Linear(21504, 4096),
-            )
-            .half()
-            .cuda()
-        )
-
-        apply_fake_sparsity(model)
-        model_copy = copy.deepcopy(model)
+        apply_fake_sparsity(self.model)
+        model_copy = copy.deepcopy(self.model)
 
         # Quantized
         quantize_(model_copy.bfloat16(), int4_weight_only())
         model_copy.foward = torch.compile(model_copy.forward, fullgraph=True)
-        dense_result = model_copy(input.bfloat16()).half()
+        dense_result = model_copy(self.input.bfloat16()).half()
 
         # Sparse + quantized
-        quantize_(model, int4_weight_only(layout_type=MarlinSparseLayoutType()))
-        model.forward = torch.compile(model.forward, fullgraph=True)
-        sparse_result = model(input)
+        quantize_(self.model, int4_weight_only(layout_type=MarlinSparseLayoutType()))
+        if not TORCH_VERSION_AT_LEAST_2_5:
+            unwrap_tensor_subclass(self.model)
+
+        self.model.forward = torch.compile(self.model.forward, fullgraph=True)
+        sparse_result = self.model(self.input)
 
         assert torch.allclose(dense_result, sparse_result, atol=3e-1), "Results are not close"
 
@@ -87,34 +80,26 @@ def test_pack_unpack_equivalence(self):
         num_bits = 4
         group_size = 128
         shape = (11008, 4096)
-        max_q_val = 2**num_bits - 1
-        half_q_val = (max_q_val + 1) // 2
+        block_size = (1, group_size)
+        target_dtype = torch.int32
+        quant_min = 0
+        quant_max = 15
+        eps = 1e-6
+        zero_point_dtype = torch.bfloat16
+        mapping_type = MappingType.SYMMETRIC
+        preserve_zero = True
+        zero_point_domain = ZeroPointDomain.INT
+        scale_dtype = None
 
         w = torch.rand(shape, dtype=torch.float16, device="cuda")
-        size_k, size_n = w.shape
 
         # Inject 2:4 sparsity mask
         w_24, _ = inject_24(w, *w.shape)
 
         # Quantize weights 
-        w_24 = w_24.reshape((-1, group_size, size_n))
-        w_24 = w_24.permute(1, 0, 2)
-        w_24 = w_24.reshape((group_size, -1))
-
-        # Compute scale for each group
-        scales = torch.max(torch.abs(w_24), 0, keepdim=True)[0]
-        scales *= 2 / max_q_val  # 2 => symmetric
-
-        # Quantize
-        w_q_24 = torch.round(w_24 / scales).int()
-        w_q_24 += half_q_val
-        w_q_24 = torch.clamp(w_q_24, 0, max_q_val)
-
-        # Shape back to original shape
-        w_q_24 = w_q_24.reshape((group_size, -1, size_n))
-        w_q_24 = w_q_24.permute(1, 0, 2)
-        w_q_24 = w_q_24.reshape((size_k, size_n)).contiguous()
-        scales = scales.reshape((-1, size_n)).contiguous()
+        scales, zeros = choose_qparams_affine(w_24, mapping_type, block_size, target_dtype, quant_min, quant_max, eps, scale_dtype, zero_point_dtype, preserve_zero, zero_point_domain)
+        w_q_24 = quantize_affine(w_24, block_size, scales, zeros, target_dtype, quant_min, quant_max, zero_point_domain)
+        scales = scales.reshape(-1, w_q_24.shape[1])
 
         # Test pack/unpack equivalence
         q_w_comp, packed_scales, meta = pack_to_marlin_24(

torchao/sparsity/README.md

@@ -58,6 +58,23 @@ For more information about accelerting BERT with semi-sturcutred sparsity, pleas
 | F1 (%) | 86.93 | 86.49 | -0.44 |
 | Time (bs=16) | 19.35 | 15.74 | 1.23x |
 
+# Implemented APIs
+
+## Quantization + Sparsity
+
+### Sparse Marlin 2:4
+
+Sparse-Marlin 2:4 is an optimized GPU kernel that extends the Mixed Auto-Regressive Linear (Marlin) dense kernel to support 4-bit quantized weights and 2:4 sparsity, improving performance in matrix multiplication and accumulation. Full documentation can be found [here](https://github.com/IST-DASLab/Sparse-Marlin).
+
+```py
+from torchao.quantization.quant_api import quantize_, int4_weight_only
+from torchao.dtypes import MarlinSparseLayoutType
+
+# Your FP16 model
+model = model.cuda().half()
+
+quantize_(model, int4_weight_only(layout_type=MarlinSparseLayoutType()))
+```
 
 # Design
 

torchao/sparsity/marlin/utils.py

@@ -96,7 +96,11 @@ def get_perms_24(num_bits: int) -> Tuple[torch.Tensor, List[int], List[int]]:
     """Precompute permutations for Marlin24 weight and scale shuffling
     
     Marlin works on [16*2,64] tiles. The goal of the permutations is to reorder the weight data so that it is compatible
-    with the tensor-core format.
+    with the tensor-core format that is described here:
+    https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#matrix-fragments-for-mma-m16n8k16-with-floating-point-type
+    
+    As a result of this reordering, the vector loads inside the kernel will get the data as it is needed for tensor-core
+    (without the need to use ldmatrix instructions)
     
     Args:
         num_bits (int): Number of bits to pack.