[Feature] Support Ernie FP8 on sm100 (#5593)

* Deepgemm暂时可用版本

* dense部分 e8m0 ok

* EB模型E8M0跑通的版本

* code check

* support 21b-tp2, dev_paddle

* 单机4.5T ep OK的版本

* 修复删除的代码,单机4.5T ep(非cudagraph)

* eb tp

* Support SM100 block-wise FP8 inference

* refine codes, support deepgemm on sm100

* add thirdparty PFCC/DeepGEMM

* fix ep decode

* 使用deepep ue8m0, 解决精度问题

* 修复FP8 TP精度

* Deepgemm升级适配Hopper逻辑

* add ue8m0 kernel

* add ue8m0 kernel

* fix custom_ops/gpu_ops/cpp_extensions.cc

* eb 输出正常

* eb5 text is right

* 目测精度一致

* 自测精度对齐

* 替换masked_per_token_quant, ep精度OK

* 性能提升约30%

* 暂时跑通ep但是有问题

* 自测一致

* rm test fun

* fix ep event

* 图优化算子更新Deepgemm

* fix build

* 暂时绕过deepgemm CI编译问题

* 根据SM区分deepgemm版本

* remove useless code

---------

Co-authored-by: ckl117 <ckl117@163.com>
Co-authored-by: K11OntheBoat <“ruianmaidanglao@163.com”>
Co-authored-by: fxyfxy777 <fxyfxy777@163.com>

fastdeploy/model_executor/layers/utils.py

@@ -254,7 +254,10 @@ def per_block_cast_to_fp8(x: Tensor, block_size: list = [128, 128]) -> Tuple[Ten
     Only used in deep_gemm block wise quant weight.
     copy from FastDeploy/custom_ops/gpu_ops/fp8_deep_gemm/tests/test_core.py.
     """
-    from fastdeploy.model_executor.ops.gpu.deep_gemm import ceil_div
+    try:
+        from deep_gemm import ceil_div
+    except ModuleNotFoundError:
+        from fastdeploy.model_executor.ops.gpu.deep_gemm import ceil_div
 
     assert x.dim() == 2
     m, n = x.shape
@@ -551,6 +554,12 @@ def vocab_range_from_global_vocab_size(global_vocab_size: int, rank: int, world_
     return vocab_range_from_per_partition_vocab_size(per_partition_vocab_size, rank, offset=offset)
 
 
+def get_sm_version():
+    prop = paddle.device.cuda.get_device_properties()
+    cc = prop.major * 10 + prop.minor
+    return cc
+
+
 def modules_to_convert(prefix: str, fd_config: FDConfig):
     import fnmatch