[XPU][MXFP4] add mxfp4 quant op for XPU (#39857)

Signed-off-by: Zhu, Zufang <zufang.zhu@intel.com>
This commit is contained in:
zofia
2026-04-15 12:28:19 +00:00
committed by GitHub
parent 68be0f853e
commit fc701c8058
2 changed files with 50 additions and 0 deletions
+46
View File
@@ -144,6 +144,46 @@ def _xpu_mxfp8_quantize_fake(
return x.to(dtype), x_s.to(torch.float8_e8m0fnu)
def _xpu_mxfp4_quantize_impl(
x: torch.Tensor,
) -> tuple[torch.Tensor, torch.Tensor]:
MXFP4_BLOCK_SIZE = 32
eps = 1e-10
assert x.ndim == 2, "input must be 2-D"
assert x.shape[-1] % MXFP4_BLOCK_SIZE == 0, (
f"last dimension {x.shape[-1]} must be divisible by group_size "
f"{MXFP4_BLOCK_SIZE}"
)
assert x.is_contiguous(), "input groups must be contiguous"
M, N = x.shape
# Packed FP4 output: two nibbles per byte
x_q = torch.empty(M, N // 2, device=x.device, dtype=torch.uint8)
x_s = torch.empty(M, N // MXFP4_BLOCK_SIZE, device=x.device, dtype=torch.float32)
torch.ops._C.per_token_group_quant_mxfp4(x, x_q, x_s, MXFP4_BLOCK_SIZE, eps)
x_q = x_q.view(torch.float4_e2m1fn_x2)
x_s = x_s.to(dtype=torch.float8_e8m0fnu, memory_format=torch.preserve_format)
return x_q, x_s
def _xpu_mxfp4_quantize_fake(
x: torch.Tensor,
) -> tuple[torch.Tensor, torch.Tensor]:
MXFP4_BLOCK_SIZE = 32
M, N = x.shape
# Packed FP4 output: two nibbles per byte
x_q = torch.empty(M, N // 2, device=x.device, dtype=torch.uint8)
x_s = torch.empty(M, N // MXFP4_BLOCK_SIZE, device=x.device, dtype=torch.float32)
x_q = x_q.view(torch.float4_e2m1fn_x2)
x_s = x_s.to(dtype=torch.float8_e8m0fnu, memory_format=torch.preserve_format)
return x_q, x_s
# Global flag to ensure ops are registered only once
_OPS_REGISTERED = False
@@ -555,6 +595,12 @@ class xpu_ops:
fake_impl=_xpu_mxfp8_quantize_fake,
)
direct_register_custom_op(
op_name="xpu_mxfp4_quantize",
op_func=_xpu_mxfp4_quantize_impl,
fake_impl=_xpu_mxfp4_quantize_fake,
)
_OPS_REGISTERED = True
@@ -162,3 +162,7 @@ try:
quant_dequant_mxfp4 = torch.ops.vllm.quant_dequant_mxfp4
except AttributeError as error:
raise error
def xpu_mxfp4_quantize(x: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
return torch.ops.vllm.xpu_mxfp4_quantize(x)