mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2025-12-24 13:28:13 +08:00
707d1a1fc9
* [RL] Support Rollout Routing Replay
* add routing indices cache
* fix config bug and moe forward bug
* R3 Support GLM
* support eb4.5
* fix merge bug
* Apply suggestion from @Copilot
* Apply suggestion from @Copilot
* Apply suggestion from @Copilot
* Apply suggestion from @Copilot
* add routing replay ci
* support glm topk
* support orther top_k
* fix ci bug
* pre-commit
* only support chatcmpl
* Revert "Revert "[RL] Support Rollout Routing Replay (#5321)" (#5402)"
This reverts commit c45e064f3d.
* Fix XPU and NPU bug
---------
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: Yuanle Liu <yuanlehome@163.com>
671 lines
17 KiB
Python
671 lines
17 KiB
Python
"""
|
|
# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License"
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
"""
|
|
|
|
import json
|
|
import os
|
|
import shutil
|
|
import unittest
|
|
|
|
import numpy as np
|
|
import paddle
|
|
import paddle.device.cuda.graphs as graphs
|
|
from paddle.distributed import fleet
|
|
|
|
from fastdeploy.config import (
|
|
CacheConfig,
|
|
FDConfig,
|
|
GraphOptimizationConfig,
|
|
LoadConfig,
|
|
ModelConfig,
|
|
ParallelConfig,
|
|
RoutingReplayConfig,
|
|
)
|
|
from fastdeploy.model_executor.layers.moe.moe import FusedMoE
|
|
from fastdeploy.model_executor.layers.quantization.block_wise_fp8 import (
|
|
BlockWiseFP8Config,
|
|
)
|
|
from fastdeploy.scheduler import SchedulerConfig
|
|
from fastdeploy.worker.worker_process import init_distributed_environment
|
|
|
|
paddle.set_default_dtype("bfloat16")
|
|
|
|
gate_correction_bias_real_data = paddle.to_tensor(
|
|
[
|
|
32.8339,
|
|
32.8231,
|
|
32.8151,
|
|
32.8131,
|
|
32.8317,
|
|
32.8343,
|
|
32.8356,
|
|
32.8270,
|
|
32.8344,
|
|
32.8342,
|
|
32.8126,
|
|
32.8299,
|
|
32.8282,
|
|
32.8254,
|
|
32.8320,
|
|
32.8280,
|
|
32.8303,
|
|
32.8351,
|
|
32.8364,
|
|
32.8347,
|
|
32.8179,
|
|
32.8349,
|
|
32.8322,
|
|
32.8323,
|
|
32.8360,
|
|
32.8351,
|
|
32.8059,
|
|
32.8352,
|
|
32.8303,
|
|
32.8334,
|
|
32.8283,
|
|
32.8265,
|
|
32.8344,
|
|
32.8307,
|
|
32.8271,
|
|
32.8343,
|
|
32.8326,
|
|
32.8327,
|
|
32.8349,
|
|
32.8356,
|
|
32.8303,
|
|
32.8327,
|
|
32.8310,
|
|
32.8363,
|
|
32.8274,
|
|
32.8335,
|
|
32.8350,
|
|
32.8255,
|
|
32.8298,
|
|
32.8141,
|
|
32.8218,
|
|
32.8362,
|
|
32.8126,
|
|
32.7902,
|
|
32.8314,
|
|
32.8356,
|
|
32.8177,
|
|
32.8333,
|
|
32.8352,
|
|
32.8354,
|
|
32.8334,
|
|
32.8325,
|
|
32.7971,
|
|
32.8319,
|
|
32.8222,
|
|
32.8284,
|
|
32.8288,
|
|
32.8355,
|
|
32.8351,
|
|
32.8356,
|
|
32.8338,
|
|
32.8346,
|
|
32.7737,
|
|
32.8317,
|
|
32.8357,
|
|
32.8345,
|
|
32.8347,
|
|
32.8360,
|
|
32.8289,
|
|
32.8268,
|
|
32.8164,
|
|
32.8324,
|
|
32.8363,
|
|
32.8308,
|
|
32.8352,
|
|
32.8302,
|
|
32.8345,
|
|
32.8298,
|
|
32.8057,
|
|
32.8229,
|
|
32.8355,
|
|
32.8325,
|
|
32.8350,
|
|
32.8357,
|
|
32.8315,
|
|
32.8327,
|
|
32.8263,
|
|
32.8342,
|
|
32.8165,
|
|
32.8349,
|
|
32.8310,
|
|
32.8101,
|
|
32.8101,
|
|
32.8081,
|
|
32.8341,
|
|
32.8313,
|
|
32.8331,
|
|
32.8299,
|
|
32.8320,
|
|
32.7941,
|
|
32.8277,
|
|
32.8287,
|
|
32.8326,
|
|
32.8331,
|
|
32.8360,
|
|
32.8295,
|
|
32.8255,
|
|
32.8330,
|
|
32.8279,
|
|
32.8210,
|
|
32.7921,
|
|
32.8348,
|
|
32.8271,
|
|
32.8297,
|
|
32.8211,
|
|
32.8353,
|
|
32.8339,
|
|
32.8335,
|
|
32.8275,
|
|
32.8245,
|
|
32.8287,
|
|
32.8352,
|
|
32.8318,
|
|
32.8354,
|
|
32.8110,
|
|
32.8347,
|
|
32.8340,
|
|
32.8322,
|
|
32.8341,
|
|
32.8316,
|
|
32.8328,
|
|
32.8341,
|
|
32.8354,
|
|
32.8264,
|
|
32.8362,
|
|
32.8352,
|
|
32.8293,
|
|
32.8292,
|
|
32.8328,
|
|
32.8316,
|
|
32.8329,
|
|
32.8308,
|
|
32.8307,
|
|
32.8170,
|
|
32.8345,
|
|
32.8356,
|
|
32.8176,
|
|
32.8326,
|
|
32.8288,
|
|
32.8355,
|
|
32.8346,
|
|
32.8337,
|
|
32.8049,
|
|
32.8315,
|
|
32.8337,
|
|
32.8352,
|
|
32.7991,
|
|
32.8304,
|
|
32.8348,
|
|
32.8316,
|
|
32.8358,
|
|
32.8279,
|
|
32.8348,
|
|
32.8326,
|
|
32.8215,
|
|
32.8281,
|
|
32.8344,
|
|
32.8309,
|
|
32.8355,
|
|
32.8337,
|
|
32.8276,
|
|
32.8250,
|
|
32.8340,
|
|
32.8322,
|
|
32.8317,
|
|
32.8274,
|
|
32.8363,
|
|
32.8277,
|
|
32.8345,
|
|
32.8342,
|
|
32.8343,
|
|
32.8355,
|
|
32.8326,
|
|
32.8299,
|
|
32.8322,
|
|
32.8351,
|
|
32.8356,
|
|
32.7925,
|
|
32.8362,
|
|
32.8170,
|
|
32.8323,
|
|
32.8335,
|
|
32.8339,
|
|
32.8193,
|
|
32.8340,
|
|
32.8362,
|
|
32.8323,
|
|
32.8328,
|
|
32.8328,
|
|
32.8296,
|
|
32.8297,
|
|
32.8344,
|
|
32.8254,
|
|
32.8341,
|
|
32.8345,
|
|
32.7967,
|
|
32.8228,
|
|
32.8363,
|
|
32.8356,
|
|
32.8317,
|
|
32.8362,
|
|
32.8302,
|
|
32.8356,
|
|
32.8239,
|
|
32.8304,
|
|
32.8323,
|
|
32.8335,
|
|
32.8196,
|
|
32.8354,
|
|
32.6991,
|
|
32.8350,
|
|
32.8337,
|
|
32.8314,
|
|
32.8274,
|
|
32.8232,
|
|
32.8305,
|
|
32.8349,
|
|
32.8246,
|
|
32.8343,
|
|
32.8339,
|
|
32.7849,
|
|
32.8359,
|
|
32.8353,
|
|
32.8352,
|
|
32.8348,
|
|
32.8095,
|
|
32.8301,
|
|
32.8350,
|
|
32.8340,
|
|
32.8353,
|
|
32.8343,
|
|
32.8344,
|
|
32.8312,
|
|
32.8350,
|
|
32.8327,
|
|
32.8231,
|
|
32.8325,
|
|
32.8352,
|
|
32.8352,
|
|
32.8293,
|
|
32.8357,
|
|
32.8337,
|
|
32.8335,
|
|
32.8348,
|
|
32.8321,
|
|
32.8153,
|
|
32.8352,
|
|
32.8265,
|
|
32.8326,
|
|
32.8361,
|
|
32.8357,
|
|
32.8312,
|
|
32.8347,
|
|
32.8152,
|
|
32.8340,
|
|
32.8272,
|
|
32.8352,
|
|
32.8331,
|
|
32.8324,
|
|
32.7952,
|
|
32.8170,
|
|
32.8356,
|
|
32.8360,
|
|
32.8298,
|
|
32.8356,
|
|
32.8331,
|
|
32.8317,
|
|
32.8349,
|
|
32.8269,
|
|
32.8323,
|
|
32.8354,
|
|
32.8350,
|
|
32.8226,
|
|
32.8002,
|
|
32.8205,
|
|
32.8329,
|
|
32.8319,
|
|
32.8297,
|
|
32.8282,
|
|
32.8356,
|
|
32.8303,
|
|
32.8349,
|
|
32.8337,
|
|
32.8247,
|
|
32.8279,
|
|
32.8309,
|
|
32.8225,
|
|
32.8337,
|
|
32.8356,
|
|
32.8105,
|
|
32.8353,
|
|
32.8361,
|
|
32.8297,
|
|
32.8313,
|
|
32.8313,
|
|
32.8363,
|
|
32.8357,
|
|
32.8357,
|
|
32.8363,
|
|
32.7806,
|
|
32.8306,
|
|
32.8347,
|
|
32.8248,
|
|
32.8334,
|
|
32.8356,
|
|
32.8324,
|
|
32.8327,
|
|
32.8284,
|
|
32.8351,
|
|
32.8349,
|
|
32.8351,
|
|
32.8171,
|
|
32.8317,
|
|
32.8363,
|
|
32.8346,
|
|
32.8335,
|
|
32.8307,
|
|
32.7907,
|
|
32.8229,
|
|
32.8346,
|
|
32.8298,
|
|
32.8336,
|
|
32.8313,
|
|
32.8349,
|
|
32.8219,
|
|
32.8354,
|
|
32.8337,
|
|
32.8294,
|
|
32.8306,
|
|
32.8322,
|
|
32.8290,
|
|
32.8333,
|
|
32.8327,
|
|
32.8279,
|
|
32.8283,
|
|
32.8338,
|
|
32.8310,
|
|
32.8351,
|
|
32.8171,
|
|
32.8310,
|
|
32.8323,
|
|
32.8324,
|
|
32.8215,
|
|
32.8314,
|
|
32.8333,
|
|
32.8353,
|
|
32.8184,
|
|
32.8344,
|
|
32.8280,
|
|
32.8352,
|
|
32.8361,
|
|
32.8308,
|
|
32.8271,
|
|
32.8335,
|
|
32.8236,
|
|
32.8350,
|
|
32.8325,
|
|
32.8330,
|
|
32.8228,
|
|
32.8352,
|
|
32.8258,
|
|
32.8343,
|
|
32.8338,
|
|
32.8292,
|
|
],
|
|
dtype="float32",
|
|
)
|
|
|
|
|
|
class MockForwardMeta:
|
|
def __init__(self):
|
|
# chunked MoE related.
|
|
self.moe_num_chunk = 1
|
|
self.max_moe_num_chunk = 1
|
|
|
|
|
|
class FuseMoEWrapper(paddle.nn.Layer):
|
|
def __init__(
|
|
self,
|
|
model_config: ModelConfig,
|
|
tp_size: int = 1,
|
|
tp_rank: int = 0,
|
|
ep_size: int = 1,
|
|
ep_rank: int = 0,
|
|
prefix: str = "layer0",
|
|
nnodes: int = 1,
|
|
):
|
|
super().__init__()
|
|
self.model_config = model_config
|
|
|
|
self.tp_size = tp_size
|
|
self.ep_size = ep_size
|
|
self.ep_rank = ep_rank
|
|
|
|
self.prefix = prefix
|
|
self.fd_config = FDConfig(
|
|
model_config=self.model_config,
|
|
parallel_config=ParallelConfig(
|
|
{
|
|
"tensor_parallel_size": self.tp_size,
|
|
"expert_parallel_size": self.ep_size,
|
|
"expert_parallel_rank": self.ep_rank,
|
|
"data_parallel_size": self.ep_size,
|
|
}
|
|
),
|
|
quant_config=BlockWiseFP8Config(weight_block_size=[128, 128]),
|
|
# quant_config=WINT8Config({}),
|
|
# quant_config=WINT4Config({}),
|
|
scheduler_config=SchedulerConfig({}),
|
|
cache_config=CacheConfig({}),
|
|
graph_opt_config=GraphOptimizationConfig({}),
|
|
load_config=LoadConfig({}),
|
|
ips=",".join(["0"] * nnodes),
|
|
routing_replay_config=RoutingReplayConfig({}),
|
|
)
|
|
self.fd_config.parallel_config.tp_group = None
|
|
self.fd_config.parallel_config.tensor_parallel_rank = tp_rank
|
|
self.fd_config.parallel_config.expert_parallel_size = self.ep_size
|
|
if self.ep_size > 1:
|
|
self.fd_config.parallel_config.ep_group = fleet.get_hybrid_communicate_group().get_model_parallel_group()
|
|
self.fd_config.scheduler_config.splitwise_role = "mixed"
|
|
self.fd_config.model_config.moe_phase.phase = "decode"
|
|
|
|
weight_key_map = {
|
|
"gate_weight_key": f"{self.prefix}.gate.weight",
|
|
"gate_correction_bias_key": f"{self.prefix}.moe_statics.e_score_correction_bias",
|
|
"up_gate_proj_expert_weight_key": f"{self.prefix}.experts.{{}}.up_gate_proj.weight",
|
|
"down_proj_expert_weight_key": f"{self.prefix}.experts.{{}}.down_proj.weight",
|
|
}
|
|
|
|
self.fused_moe = FusedMoE(
|
|
fd_config=self.fd_config,
|
|
moe_intermediate_size=self.fd_config.model_config.moe_intermediate_size,
|
|
num_experts=self.fd_config.model_config.moe_num_experts,
|
|
top_k=self.fd_config.model_config.moe_k,
|
|
# avoiding invoke clean_low_latency_buffer in mixed ep.
|
|
layer_idx=666,
|
|
weight_key_map=weight_key_map,
|
|
topk_method="noaux_tc",
|
|
topk_group=4,
|
|
n_group=8,
|
|
gate_correction_bias=paddle.zeros([self.fd_config.model_config.moe_num_experts], paddle.float32),
|
|
# gate_correction_bias = gate_correction_bias_real_data
|
|
)
|
|
moe_layer = self.fused_moe
|
|
|
|
up_gate_proj_weight_shape = [
|
|
moe_layer.num_local_experts,
|
|
moe_layer.hidden_size,
|
|
moe_layer.moe_intermediate_size * 2,
|
|
]
|
|
down_proj_weight_shape = [
|
|
moe_layer.num_local_experts,
|
|
moe_layer.moe_intermediate_size,
|
|
moe_layer.hidden_size,
|
|
]
|
|
|
|
up_gate_proj_weight = paddle.randn(up_gate_proj_weight_shape, paddle.bfloat16)
|
|
down_proj_weight = paddle.randn(down_proj_weight_shape, paddle.bfloat16)
|
|
|
|
local_expert_ids = list(
|
|
range(moe_layer.expert_id_offset, moe_layer.expert_id_offset + moe_layer.num_local_experts)
|
|
)
|
|
state_dict = {}
|
|
up_gate_proj_expert_weight_key = moe_layer.weight_key_map.get("up_gate_proj_expert_weight_key")
|
|
down_proj_expert_weight_key = moe_layer.weight_key_map.get("down_proj_expert_weight_key")
|
|
for expert_idx in local_expert_ids:
|
|
down_proj_expert_weight_key_name = down_proj_expert_weight_key.format(expert_idx)
|
|
up_gate_proj_expert_weight_key_name = up_gate_proj_expert_weight_key.format(expert_idx)
|
|
state_dict[up_gate_proj_expert_weight_key_name] = up_gate_proj_weight[
|
|
expert_idx - moe_layer.expert_id_offset
|
|
]
|
|
state_dict[down_proj_expert_weight_key_name] = down_proj_weight[expert_idx - moe_layer.expert_id_offset]
|
|
|
|
moe_layer.load_state_dict(state_dict)
|
|
|
|
|
|
class TestFusedMoE(unittest.TestCase):
|
|
def setUp(self) -> None:
|
|
self.architectures = ["Ernie4_5_MoeForCausalLM"]
|
|
self.hidden_size = 4096
|
|
self.moe_intermediate_size = 2048
|
|
self.moe_num_experts = 160
|
|
self.moe_k = 8
|
|
self.num_layers = 2
|
|
self.num_attention_heads = -1
|
|
self.model_config = self.build_model_config()
|
|
|
|
def build_model_config(self) -> ModelConfig:
|
|
model_name_or_path = self.build_config_json()
|
|
return ModelConfig(
|
|
{
|
|
"model": model_name_or_path,
|
|
"max_model_len": 2048,
|
|
}
|
|
)
|
|
|
|
def build_config_json(self) -> str:
|
|
config_dict = {
|
|
"architectures": self.architectures,
|
|
"hidden_size": self.hidden_size,
|
|
"moe_intermediate_size": self.moe_intermediate_size,
|
|
"moe_num_experts": self.moe_num_experts,
|
|
"moe_k": self.moe_k,
|
|
"num_attention_heads": self.num_attention_heads,
|
|
"dtype": "bfloat16",
|
|
}
|
|
|
|
tmp_dir = f"./tmpwedfewfef{paddle.distributed.get_rank()}"
|
|
os.makedirs(tmp_dir, exist_ok=True)
|
|
with open(f"./{tmp_dir}/config.json", "w") as f:
|
|
json.dump(config_dict, f)
|
|
self.model_name_or_path = os.path.join(os.getcwd(), tmp_dir)
|
|
return self.model_name_or_path
|
|
|
|
def test_fused_moe(self):
|
|
init_distributed_environment()
|
|
|
|
gating = paddle.nn.Linear(self.model_config.hidden_size, self.model_config.moe_num_experts)
|
|
gating.to(dtype=paddle.float32) # it's dtype is bfloat16 default, but the forward input is float32
|
|
gating.weight.set_value(paddle.rand(gating.weight.shape, dtype=paddle.float32))
|
|
|
|
os.environ["FD_USE_DEEP_GEMM"] = "0"
|
|
ep_size = paddle.distributed.get_world_size()
|
|
ep_rank = paddle.distributed.get_rank()
|
|
|
|
tp_rank = 0
|
|
tp_size = 1
|
|
|
|
nnodes = (ep_size + 7) // 8
|
|
|
|
# 这行代码必须保留，否则影响均匀性！
|
|
paddle.seed(ep_rank + 100)
|
|
|
|
num_layers = self.num_layers
|
|
real_weight_layers = num_layers // 2
|
|
fused_moe = [None] * real_weight_layers
|
|
for i in range(real_weight_layers):
|
|
fused_moe[i] = FuseMoEWrapper(self.model_config, tp_size, tp_rank, ep_size, ep_rank, nnodes=nnodes)
|
|
|
|
moe_cuda_graphs = [None] * 100
|
|
cache_hidden_states = [None] * 100
|
|
is_decoder = fused_moe[0].fd_config.model_config.moe_phase.phase == "decode"
|
|
test_token_nums = [4096 * i for i in [1, 2, 4, 8]]
|
|
if is_decoder:
|
|
test_token_nums = [10, 20, 40, 60, 80, 100, 128, 160, 192, 256]
|
|
for idx, num_tokens in enumerate(test_token_nums):
|
|
|
|
cache_hidden_states[idx] = paddle.rand((num_tokens, self.model_config.hidden_size), dtype=paddle.bfloat16)
|
|
|
|
def fake_model_run():
|
|
for j in range(num_layers):
|
|
out = fused_moe[j % real_weight_layers].fused_moe(
|
|
cache_hidden_states[idx], gating, forward_meta=MockForwardMeta()
|
|
)
|
|
|
|
return out
|
|
|
|
if is_decoder:
|
|
moe_cuda_graphs[idx] = graphs.CUDAGraph()
|
|
moe_cuda_graphs[idx].capture_begin()
|
|
|
|
fake_model_run()
|
|
|
|
if is_decoder:
|
|
moe_cuda_graphs[idx].capture_end()
|
|
|
|
num_tests = 20
|
|
start_events = [paddle.device.cuda.Event(enable_timing=True) for _ in range(num_tests)]
|
|
end_events = [paddle.device.cuda.Event(enable_timing=True) for _ in range(num_tests)]
|
|
for i in range(num_tests):
|
|
start_events[i].record()
|
|
|
|
if is_decoder:
|
|
moe_cuda_graphs[idx].replay()
|
|
else:
|
|
fake_model_run()
|
|
|
|
end_events[i].record()
|
|
paddle.device.cuda.synchronize()
|
|
|
|
times = np.array([round(s.elapsed_time(e), 1) for s, e in zip(start_events, end_events)])[1:]
|
|
print("num_token:", num_tokens)
|
|
print(times[-5:])
|
|
rdma_GB = 3.0 * num_tokens * self.moe_k * self.hidden_size / (1e9)
|
|
times_s = (times[-1] / num_layers) / (1e3)
|
|
print(times[-1], round(rdma_GB / times_s, 1))
|
|
|
|
tmp_layer = fused_moe[0].fused_moe
|
|
memory_GB = (
|
|
tmp_layer.num_local_experts
|
|
* tmp_layer.hidden_size
|
|
* tmp_layer.moe_intermediate_size
|
|
* 3
|
|
/ (1e9)
|
|
* num_layers
|
|
)
|
|
print(round(memory_GB / times[-1], 1), "TB/s")
|
|
|
|
shutil.rmtree(self.model_name_or_path)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|