[Refactor] Remove dead code quantization 2 (#52821)

Signed-off-by: yewentao256 <zhyanwentao@126.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
Wentao Ye
2026-08-25 20:28:44 -07:00
committed by GitHub
co-authored by mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent 7156c63bef
commit a5e0045433
14 changed files with 0 additions and 526 deletions
@@ -38,9 +38,6 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
kFp8StaticChannelSym,
kFp8StaticTensorSym,
)
from vllm.model_executor.layers.quantization.utils.w8a8_utils import (
cutlass_block_fp8_supported,
)
__all__ = ["CompressedTensorsW8A8Fp8"]
@@ -67,7 +64,6 @@ class CompressedTensorsW8A8Fp8(CompressedTensorsScheme):
self.weight_block_size = self.weight_quant.block_structure
if self.weight_block_size is not None:
self.cutlass_block_fp8_supported = cutlass_block_fp8_supported()
self.use_aiter_and_is_supported = rocm_aiter_ops.is_linear_fp8_enabled()
assert not self.is_static_input_scheme
self.act_q_group_shape = GroupShape(1, self.weight_block_size[0])
@@ -67,7 +67,6 @@ from vllm.model_executor.layers.quantization.utils.quant_utils import (
kFp8StaticTensorSym,
)
from vllm.model_executor.layers.quantization.utils.w8a8_utils import (
cutlass_block_fp8_supported,
cutlass_fp8_supported,
normalize_e4m3fn_to_e4m3fnuz,
)
@@ -252,7 +251,6 @@ class Fp8LinearMethod(LinearMethodBase):
def __init__(self, quant_config: Fp8Config):
self.quant_config = quant_config
self.is_scale_e8m0 = getattr(quant_config, "is_scale_e8m0", False)
self.cutlass_block_fp8_supported = cutlass_block_fp8_supported()
self.out_dtype = torch.get_default_dtype()
self.input_dtype = get_current_vllm_config().model_config.dtype
@@ -9,7 +9,6 @@ if TYPE_CHECKING:
from vllm.model_executor.layers.fused_moe import FusedMoEMethodBase
from vllm.model_executor.layers.linear import LinearMethodBase
from vllm.model_executor.layers.quantization import QuantizationMethods
from ..config_parser import INCLayerConfig
from ..inc import INCConfig
@@ -22,7 +21,6 @@ class INCScheme(ABC):
- can_handle(): when does this scheme apply?
- get_linear_method(): required — how to quantize Linear layers
- get_moe_method(): optional — how to quantize MoE layers
- get_kvcache_method(): optional — how to quantize KV cache
Schemes that don't support MoE/KVCache inherit the default raise.
"""
@@ -56,20 +54,6 @@ class INCScheme(ABC):
f"Layer config: {layer_config}"
)
def get_kvcache_method(
self,
config: "INCConfig",
layer: "torch.nn.Module",
prefix: str,
layer_config: "INCLayerConfig",
) -> "QuantizationMethods":
"""Optional. Override if this scheme supports KV cache quantization.
Default raises NotImplementedError."""
raise NotImplementedError(
f"{type(self).__name__} does not support KV cache quantization. "
f"Layer config: {layer_config}"
)
class INCLinearScheme(ABC):
@classmethod
@@ -1064,8 +1064,6 @@ class QuarkOCP_MX_MoEMethod(QuarkMoEMethod):
else:
self.input_dtype = None
self.fp4_dtype = getattr(torch, "float4_e2m1fn_x2", None)
self.ocp_mx_scheme = OCP_MX_Scheme.from_quant_dtype(
self.input_dtype, self.weight_dtype
)
@@ -60,8 +60,6 @@ class QuarkOCP_MX(QuarkScheme):
input_quant_spec: dict[str, Any] | None,
dynamic_mxfp4_quant: bool = False,
):
self.weight_quant_spec = weight_quant_spec
self.input_quant_spec = input_quant_spec
self.dynamic_mxfp4_quant = dynamic_mxfp4_quant
self.weight_dtype = weight_quant_spec["dtype"].replace("fp", "mxfp")
self.input_dtype: str | None = None
@@ -38,7 +38,6 @@ def cutlass_group_gemm_supported() -> bool:
return ops.cutlass_group_gemm_supported(capability)
CUTLASS_FP8_SUPPORTED = cutlass_fp8_supported()
CUTLASS_BLOCK_FP8_SUPPORTED = cutlass_block_fp8_supported()