[Kernel] Add Exllama as a backend for compressed-tensors (#9395)

vllm/envs.py

@@ -66,6 +66,7 @@ if TYPE_CHECKING:
     VLLM_SKIP_P2P_CHECK: bool = False
     VLLM_ALLOW_DEPRECATED_BLOCK_MANAGER_V1: bool = False
     VLLM_TORCH_COMPILE_LEVEL: int = 0
+    VLLM_DISABLED_KERNELS: List[str] = []
 
 
 def get_default_cache_root():
@@ -430,6 +431,14 @@ environment_variables: Dict[str, Callable[[], Any]] = {
     "VLLM_ALLOW_DEPRECATED_BLOCK_MANAGER_V1":
     lambda: os.environ.get("VLLM_ALLOW_DEPRECATED_BLOCK_MANAGER_V1", "0"
                            ) == "1",
+
+    # List of quantization kernels that should be disabled, used for testing
+    # and performance comparisons. Currently only affects MPLinearKernel
+    # selection
+    # (kernels: MacheteLinearKernel, MarlinLinearKernel, ExllamaLinearKernel)
+    "VLLM_DISABLED_KERNELS":
+    lambda: [] if "VLLM_DISABLED_KERNELS" not in os.environ else os.environ[
+        "VLLM_DISABLED_KERNELS"].split(","),
 }
 
 # end-env-vars-definition

vllm/model_executor/layers/quantization/kernels/MPLinearKernel.py

@@ -42,6 +42,10 @@ class MPLinearKernel(ABC):
         self.config = c
         self.w_q_name = w_q_param_name
         self.w_s_name = w_s_param_name
+        if c.zero_points:
+            assert w_zp_param_name is not None
+        if c.has_g_idx:
+            assert w_gidx_param_name is not None
         self.w_zp_name = w_zp_param_name
         self.w_gidx_name = w_gidx_param_name
 

vllm/model_executor/layers/quantization/kernels/__init__.py

-import os
 from typing import List, Optional, Type
 
+import vllm.envs as envs
+from vllm.model_executor.layers.quantization.kernels.exllama import (
+    ExllamaLinearKernel)
 from vllm.model_executor.layers.quantization.kernels.machete import (
     MacheteLinearKernel)
 from vllm.model_executor.layers.quantization.kernels.marlin import (
@@ -13,6 +15,7 @@ from vllm.platforms import current_platform
 _POSSIBLE_KERNELS: List[Type[MPLinearKernel]] = [
     MacheteLinearKernel,
     MarlinLinearKernel,
+    ExllamaLinearKernel,
 ]
 
 
@@ -45,8 +48,7 @@ def choose_mp_linear_kernel(
 
     failure_reasons = []
     for kernel in _POSSIBLE_KERNELS:
-        if kernel.__name__ in os.environ.get("VLLM_DISABLED_KERNELS", "")\
-            .split(","):
+        if kernel.__name__ in envs.VLLM_DISABLED_KERNELS:
             failure_reasons.append(
                 f' {kernel.__name__} disabled by environment variable')
             continue

vllm/model_executor/layers/quantization/kernels/exllama.py

+from typing import Optional, Tuple
+
+import torch
+
+from vllm import _custom_ops as ops
+from vllm.model_executor.layers.quantization.utils.quant_utils import (
+    pack_quantized_values_into_int32)
+from vllm.model_executor.parameter import (BasevLLMParameter,
+                                           permute_param_layout_)
+from vllm.scalar_type import scalar_types
+
+from .MPLinearKernel import MPLinearKernel, MPLinearLayerConfig
+
+
+class ExllamaLinearKernel(MPLinearKernel):
+    SUPPORTED_QUANT_TYPES = [scalar_types.uint4b8, scalar_types.uint8b128]
+    # In theory supports `scalar_types.uint2b2, scalar_types.uint3b4` too but
+    # currently untested so not added to the list
+
+    @classmethod
+    def get_min_capability(cls) -> int:
+        return 60
+
+    @classmethod
+    def can_implement(cls,
+                      c: MPLinearLayerConfig) -> Tuple[bool, Optional[str]]:
+        if c.has_g_idx and\
+            c.partition_weight_shape[0] != c.full_weight_shape[0]:
+            return False, "Act reordering currently not supported by Exllama, "\
+                          "when the input features are partitioned across "\
+                          "devices"
+
+        if c.partition_weight_shape[1] % (32 // c.weight_type.size_bits) != 0:
+            return False, "Output features must be a multiple of the pack " \
+                            "factor (32 / num_bits) so that we can correctly " \
+                            "pack the zero points"
+
+        if c.act_type != torch.float16:
+            return False, "Exllama only supports float16 activations"
+
+        if c.weight_type not in cls.SUPPORTED_QUANT_TYPES:
+            return False, f"Quant type ({c.weight_type}) not supported by "\
+                           "Exllama, supported types are: "\
+                           f"{cls.SUPPORTED_QUANT_TYPES}"
+
+        if c.full_weight_shape[0] % c.group_size != 0:
+            return False, f"Group size ({c.group_size}) does not evenly divide"\
+                           " the number of input features "\
+                           f"({c.full_weight_shape[0]})"
+
+        return True, None
+
+    def process_weights_after_loading(self, layer: torch.nn.Module):
+        c = self.config
+
+        # For Exllama, we need to set a zero-point tensor if there is not one
+        if not c.zero_points:
+            self.w_zp_name = "qzeros"
+            device = getattr(layer, self.w_q_name).device
+            groups = c.partition_weight_shape[0] // c.group_size
+            out_features = c.partition_weight_shape[1]
+
+            if c.weight_type.has_bias():
+                # if the type has a bias we have to create a zeros tensor that
+                # contains the bias values repeated for each group (-1 due to
+                # a bug in the original GPTQ checkpoint format leading to
+                # exllama kernel adding 1 to the zero points during inference)
+                # Documentation of the bug can be found here:
+                #  https://garden.danieldk.eu/GPTQ-Checkpoint-Format
+                zeros = torch.full((groups, out_features),
+                                   c.weight_type.bias - 1,
+                                   dtype=torch.int32,
+                                   device=device)
+            else:
+                raise NotImplementedError(
+                    "A 0 zero-point is not supported by Exllama due to "
+                    "a bug in the original GPTQ checkpoint format leading to "
+                    "exllama kernel adding 1 to the zero points during "
+                    "inference")
+            zeros = pack_quantized_values_into_int32(zeros,
+                                                     c.weight_type,
+                                                     packed_dim=1)
+            setattr(layer, self.w_zp_name,
+                    torch.nn.Parameter(zeros, requires_grad=False))
+
+        if c.has_g_idx:
+
+            def transform_w_g_idx(x):
+                # Exllama wants the permutation array instead of the group
+                # indices
+                return torch.argsort(x).to(torch.int)
+
+            self._transform_param(layer, self.w_gidx_name, transform_w_g_idx)
+        else:
+            self.w_gidx_name = "g_idx"
+            empty_g_idx = torch.nn.Parameter(torch.empty((0, ),
+                                                         dtype=torch.int,
+                                                         device=device),
+                                             requires_grad=False)
+            setattr(layer, self.w_gidx_name, empty_g_idx)
+
+        def transform_w_q(x):
+            assert isinstance(x, BasevLLMParameter)
+            assert self.w_gidx_name is not None
+            g_idx = getattr(layer, self.w_gidx_name)
+
+            permute_param_layout_(x, input_dim=0, output_dim=1, packed_dim=0)
+            x_cont = x.data.contiguous()
+            ops.gptq_shuffle(x_cont, g_idx, c.weight_type.size_bits)
+            return x_cont
+
+        def transform_w_s(x):
+            assert isinstance(x, BasevLLMParameter)
+            permute_param_layout_(x, input_dim=0, output_dim=1)
+            x.data = x.data.contiguous()
+            return x.to(dtype=c.act_type)
+
+        # Repack weights and scales for Machete
+        self._transform_param(layer, self.w_q_name, transform_w_q)
+        self._transform_param(layer, self.w_s_name, transform_w_s)
+
+    def apply_weights(self,
+                      layer: torch.nn.Module,
+                      x: torch.Tensor,
+                      bias: Optional[torch.Tensor] = None) -> torch.Tensor:
+        c = self.config
+
+        x_2d = x.reshape(-1, x.shape[-1])
+        out_shape = x.shape[:-1] + (c.partition_weight_shape[1], )
+
+        w_q, w_s, w_zp, w_g_idx = self._get_weight_params(layer)
+
+        assert w_zp is not None, "Zero points are required by Exllama"
+        assert w_g_idx is not None, "Group index is required by Exllama"
+        output = ops.gptq_gemm(x_2d, w_q, w_zp, w_s, w_g_idx, True,
+                               c.weight_type.size_bits)
+
+        if bias is not None:
+            output.add_(bias)
+        return output.reshape(out_shape)

vllm/model_executor/layers/quantization/kernels/machete.py

@@ -8,7 +8,7 @@ from vllm.model_executor.layers.quantization.utils.machete_utils import (
     MACHETE_SUPPORTED_GROUP_SIZES, check_machete_supports_shape,
     query_machete_supported_quant_types)
 from vllm.model_executor.layers.quantization.utils.quant_utils import (
-    pack_weights_into_int32, unpack_weights_into_int32)
+    pack_quantized_values_into_int32, unpack_quantized_values_into_int32)
 from vllm.model_executor.parameter import (BasevLLMParameter,
                                            permute_param_layout_)
 
@@ -71,13 +71,13 @@ class MacheteLinearKernel(MPLinearKernel):
             assert isinstance(x, BasevLLMParameter)
             permute_param_layout_(x, input_dim=0, output_dim=1, packed_dim=0)
             if c.has_g_idx:
-                x_unpacked = unpack_weights_into_int32(x.data,
-                                                       c.weight_type,
-                                                       packed_dim=0)
+                x_unpacked = unpack_quantized_values_into_int32(x.data,
+                                                                c.weight_type,
+                                                                packed_dim=0)
                 x_perm = x_unpacked[perm, :]
-                x.data = pack_weights_into_int32(x_perm,
-                                                 c.weight_type,
-                                                 packed_dim=0)
+                x.data = pack_quantized_values_into_int32(x_perm,
+                                                          c.weight_type,
+                                                          packed_dim=0)
             x.data = ops.machete_prepack_B(x.data.t().contiguous().t(),
                                            self.config.weight_type)
             return x

vllm/model_executor/layers/quantization/utils/quant_utils.py

@@ -20,9 +20,9 @@ FUSED_LAYER_NAME_MAPPING = {
 }
 
 
-def pack_weights_into_int32(w_q: torch.Tensor,
-                            wtype: ScalarType,
-                            packed_dim: int = 0):
+def pack_quantized_values_into_int32(w_q: torch.Tensor,
+                                     wtype: ScalarType,
+                                     packed_dim: int = 0):
     # move dim to pack to the end
     perm = (*[i for i in range(len(w_q.shape)) if i != packed_dim], packed_dim)
     inv_perm = tuple(perm.index(i) for i in range(len(perm)))
@@ -42,9 +42,9 @@ def pack_weights_into_int32(w_q: torch.Tensor,
     return res.permute(inv_perm)
 
 
-def unpack_weights_into_int32(w_q: torch.Tensor,
-                              wtype: ScalarType,
-                              packed_dim: int = 0):
+def unpack_quantized_values_into_int32(w_q: torch.Tensor,
+                                       wtype: ScalarType,
+                                       packed_dim: int = 0):
     # move dim to pack to the end
     perm = (*[i for i in range(len(w_q.shape)) if i != packed_dim], packed_dim)
     inv_perm = tuple(perm.index(i) for i in range(len(perm)))

vllm/scalar_type.py

@@ -27,6 +27,8 @@ class scalar_types:
     float6_e3m2f = ScalarType.float_(3, 2, True, NanRepr.NONE.value)
 
     # "gptq" types
+    uint2b2 = ScalarType.uint(2, 2)
+    uint3b4 = ScalarType.uint(3, 4)
     uint4b8 = ScalarType.uint(4, 8)
     uint8b128 = ScalarType.uint(8, 128)