apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

[Others] support import deepgemm/deepep from fleet ops (#6351)

Browse Source 
* update paddleformers to v1.0

* only change import fleetpath
This commit is contained in:
JYChen
2026-02-09 11:53:13 +08:00
committed by GitHub
parent 74762b0fb2
commit 9bcd863902
4 changed files with 46 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/model_executor/layers/quantization/block_wise_fp8.py
+4 -8
View File
@@ -42,14 +42,10 @@ from ..utils import get_sm_version, get_tensor, per_block_cast_to_fp8
from .quant_base import QuantConfigBase, QuantMethodBase
if current_platform.is_cuda():
if get_sm_version() == 100:
# SM100 should use PFCC DeepGemm
paddle.compat.enable_torch_proxy(scope={"deep_gemm"})
from deep_gemm import fp8_gemm_nt
else:
from fastdeploy.model_executor.ops.gpu.deep_gemm import (
gemm_fp8_fp8_bf16_nt as fp8_gemm_nt,
)
try:
fp8_gemm_nt = fastdeploy.model_executor.layers.quantization.fp8_utils.deep_gemm.fp8_gemm_nt
except:
fp8_gemm_nt = fastdeploy.model_executor.layers.quantization.fp8_utils.deep_gemm.gemm_fp8_fp8_bf16_nt
else:
fp8_gemm_nt = None