[Core]Refactor gptq_marlin ops (#4466)

This commit is contained in:
Kunshang Ji 2024-04-30 12:14:47 +00:00 committed by GitHub
parent fa32207842
commit 26f2fb5113
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 17 additions and 1 deletions

View File

@ -167,6 +167,22 @@ def aqlm_dequant(codes: torch.Tensor, codebooks: torch.Tensor,
return vllm_ops.aqlm_dequant(codes, codebooks, codebook_partition_sizes)
# gptq_marlin
def gptq_marlin_repack(b_q_weight: torch.Tensor, perm: torch.Tensor,
size_k: int, size_n: int) -> torch.Tensor:
return vllm_ops.gptq_marlin_repack(b_q_weight, perm, size_k, size_n)
def gptq_marlin_gemm(a: torch.Tensor, b_q_weight: torch.Tensor,
b_scales: torch.Tensor, g_idx: torch.Tensor,
perm: torch.Tensor, workspace: torch.Tensor, size_m: int,
size_n: int, size_k: int,
is_k_full: bool) -> torch.Tensor:
return vllm_ops.gptq_marlin_gemm(a, b_q_weight, b_scales, g_idx, perm,
workspace, size_m, size_n, size_k,
is_k_full)
# fp8
def scaled_fp8_quant(
input: torch.Tensor,

View File

@ -6,7 +6,7 @@ import numpy
import torch
from torch.nn.parameter import Parameter
from vllm._C import ops
from vllm import _custom_ops as ops
from vllm.model_executor.layers.linear import (LinearBase, LinearMethodBase,
set_weight_attrs)
from vllm.model_executor.layers.quantization.base_config import (