mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2026-04-23 00:17:25 +08:00
[OP] latent moe deepgemm support (#7537)
* commit * commit * commit * commit * commit * commit * commit * commit
This commit is contained in:
@@ -179,6 +179,8 @@ class MoEMethodBase(QuantMethodBase):
|
|||||||
x: paddle.Tensor,
|
x: paddle.Tensor,
|
||||||
gate: nn.Layer,
|
gate: nn.Layer,
|
||||||
topk_ids_hookfunc: Callable = None,
|
topk_ids_hookfunc: Callable = None,
|
||||||
|
fc1_latent_proj: nn.Layer = None,
|
||||||
|
fc2_latent_proj: nn.Layer = None,
|
||||||
) -> paddle.Tensor:
|
) -> paddle.Tensor:
|
||||||
"""
|
"""
|
||||||
Apply the EP prefill method.
|
Apply the EP prefill method.
|
||||||
@@ -192,6 +194,8 @@ class MoEMethodBase(QuantMethodBase):
|
|||||||
x: paddle.Tensor,
|
x: paddle.Tensor,
|
||||||
gate: nn.Layer,
|
gate: nn.Layer,
|
||||||
topk_ids_hookfunc: Callable = None,
|
topk_ids_hookfunc: Callable = None,
|
||||||
|
fc1_latent_proj: nn.Layer = None,
|
||||||
|
fc2_latent_proj: nn.Layer = None,
|
||||||
) -> paddle.Tensor:
|
) -> paddle.Tensor:
|
||||||
"""
|
"""
|
||||||
Apply the EP decoder method.
|
Apply the EP decoder method.
|
||||||
@@ -232,13 +236,19 @@ class MoEMethodBase(QuantMethodBase):
|
|||||||
if layer.fd_config.scheduler_config.splitwise_role == "mixed" and is_moe_start_layer:
|
if layer.fd_config.scheduler_config.splitwise_role == "mixed" and is_moe_start_layer:
|
||||||
self.ep_prefill_runner.clean_low_latency_buffer()
|
self.ep_prefill_runner.clean_low_latency_buffer()
|
||||||
return self.apply_ep_prefill(
|
return self.apply_ep_prefill(
|
||||||
layer, x, gate, topk_ids_hookfunc=topk_ids_hookfunc, shared_experts=shared_experts
|
layer,
|
||||||
|
x,
|
||||||
|
gate,
|
||||||
|
topk_ids_hookfunc,
|
||||||
|
shared_experts,
|
||||||
|
fc1_latent_proj,
|
||||||
|
fc2_latent_proj,
|
||||||
)
|
)
|
||||||
else:
|
else:
|
||||||
if layer.fd_config.scheduler_config.splitwise_role == "mixed" and is_moe_start_layer:
|
if layer.fd_config.scheduler_config.splitwise_role == "mixed" and is_moe_start_layer:
|
||||||
self.ep_decoder_runner.clean_low_latency_buffer()
|
self.ep_decoder_runner.clean_low_latency_buffer()
|
||||||
return self.apply_ep_decode(
|
return self.apply_ep_decode(
|
||||||
layer, x, gate, topk_ids_hookfunc=topk_ids_hookfunc, shared_experts=shared_experts
|
layer, x, gate, topk_ids_hookfunc, shared_experts, fc1_latent_proj, fc2_latent_proj
|
||||||
)
|
)
|
||||||
else:
|
else:
|
||||||
return self.apply_tp(layer, x, gate, topk_ids_hookfunc, fc1_latent_proj, fc2_latent_proj)
|
return self.apply_tp(layer, x, gate, topk_ids_hookfunc, fc1_latent_proj, fc2_latent_proj)
|
||||||
|
|||||||
@@ -128,6 +128,8 @@ class CutlassMoEMethod(UnquantizedFusedMoEMethod):
|
|||||||
gate: nn.Layer,
|
gate: nn.Layer,
|
||||||
topk_ids_hookfunc: Callable = None,
|
topk_ids_hookfunc: Callable = None,
|
||||||
shared_experts: nn.Layer = None,
|
shared_experts: nn.Layer = None,
|
||||||
|
fc1_latent_proj: nn.Layer = None,
|
||||||
|
fc2_latent_proj: nn.Layer = None,
|
||||||
) -> paddle.Tensor:
|
) -> paddle.Tensor:
|
||||||
"""
|
"""
|
||||||
Apply the EP prefill method.
|
Apply the EP prefill method.
|
||||||
@@ -275,6 +277,8 @@ class CutlassMoEMethod(UnquantizedFusedMoEMethod):
|
|||||||
gate: nn.Layer,
|
gate: nn.Layer,
|
||||||
topk_ids_hookfunc: Callable = None,
|
topk_ids_hookfunc: Callable = None,
|
||||||
shared_experts: nn.Layer = None,
|
shared_experts: nn.Layer = None,
|
||||||
|
fc1_latent_proj: nn.Layer = None,
|
||||||
|
fc2_latent_proj: nn.Layer = None,
|
||||||
) -> paddle.Tensor:
|
) -> paddle.Tensor:
|
||||||
"""
|
"""
|
||||||
Apply the EP decoder method.
|
Apply the EP decoder method.
|
||||||
|
|||||||
@@ -332,6 +332,8 @@ class DeepGemmFusedMoeMethod(MoEMethodBase):
|
|||||||
gate: nn.Layer,
|
gate: nn.Layer,
|
||||||
topk_ids_hookfunc: Callable = None,
|
topk_ids_hookfunc: Callable = None,
|
||||||
shared_experts: nn.Layer = None,
|
shared_experts: nn.Layer = None,
|
||||||
|
fc1_latent_proj: nn.Layer = None,
|
||||||
|
fc2_latent_proj: nn.Layer = None,
|
||||||
) -> paddle.Tensor:
|
) -> paddle.Tensor:
|
||||||
"""
|
"""
|
||||||
Apply the EP prefill method.
|
Apply the EP prefill method.
|
||||||
@@ -339,7 +341,7 @@ class DeepGemmFusedMoeMethod(MoEMethodBase):
|
|||||||
gate_out = gate(x)
|
gate_out = gate(x)
|
||||||
gate_out = gate_out.cast("float32")
|
gate_out = gate_out.cast("float32")
|
||||||
|
|
||||||
hidden_size = x.shape[1]
|
hidden_size = layer.hidden_size
|
||||||
|
|
||||||
# 1. Select topk experts and weights
|
# 1. Select topk experts and weights
|
||||||
topk_idx, topk_weights = self.ep_prefill_runner.moe_select(layer, gate_out)
|
topk_idx, topk_weights = self.ep_prefill_runner.moe_select(layer, gate_out)
|
||||||
@@ -347,6 +349,9 @@ class DeepGemmFusedMoeMethod(MoEMethodBase):
|
|||||||
if topk_ids_hookfunc is not None:
|
if topk_ids_hookfunc is not None:
|
||||||
topk_ids_hookfunc(topk_ids=topk_idx)
|
topk_ids_hookfunc(topk_ids=topk_idx)
|
||||||
|
|
||||||
|
if fc1_latent_proj:
|
||||||
|
x = fc1_latent_proj(x)
|
||||||
|
|
||||||
# 2. Dynamic compute blockwise quantization scales
|
# 2. Dynamic compute blockwise quantization scales
|
||||||
if not fastdeploy.envs.FD_USE_PHI_FP8_QUANT:
|
if not fastdeploy.envs.FD_USE_PHI_FP8_QUANT:
|
||||||
x_fp8, x_scale_tensor = fastdeploy.model_executor.ops.gpu.per_token_quant(
|
x_fp8, x_scale_tensor = fastdeploy.model_executor.ops.gpu.per_token_quant(
|
||||||
@@ -643,6 +648,9 @@ class DeepGemmFusedMoeMethod(MoEMethodBase):
|
|||||||
if shared_experts is not None:
|
if shared_experts is not None:
|
||||||
tmp_ffn_out += s_x
|
tmp_ffn_out += s_x
|
||||||
|
|
||||||
|
if fc2_latent_proj:
|
||||||
|
tmp_ffn_out = fc2_latent_proj(tmp_ffn_out)
|
||||||
|
|
||||||
return tmp_ffn_out
|
return tmp_ffn_out
|
||||||
|
|
||||||
def apply_ep_decode(
|
def apply_ep_decode(
|
||||||
@@ -652,6 +660,8 @@ class DeepGemmFusedMoeMethod(MoEMethodBase):
|
|||||||
gate: nn.Layer,
|
gate: nn.Layer,
|
||||||
topk_ids_hookfunc: Callable = None,
|
topk_ids_hookfunc: Callable = None,
|
||||||
shared_experts: nn.Layer = None,
|
shared_experts: nn.Layer = None,
|
||||||
|
fc1_latent_proj: nn.Layer = None,
|
||||||
|
fc2_latent_proj: nn.Layer = None,
|
||||||
) -> paddle.Tensor:
|
) -> paddle.Tensor:
|
||||||
"""
|
"""
|
||||||
Apply the EP decoder method.
|
Apply the EP decoder method.
|
||||||
@@ -665,6 +675,9 @@ class DeepGemmFusedMoeMethod(MoEMethodBase):
|
|||||||
topk_ids_hookfunc(topk_ids=topk_idx)
|
topk_ids_hookfunc(topk_ids=topk_idx)
|
||||||
|
|
||||||
# 2. EP Dispatch
|
# 2. EP Dispatch
|
||||||
|
if fc1_latent_proj:
|
||||||
|
x = fc1_latent_proj(x)
|
||||||
|
|
||||||
permute_input, token_nums_per_expert, handle = self.ep_decoder_runner.dispatch(
|
permute_input, token_nums_per_expert, handle = self.ep_decoder_runner.dispatch(
|
||||||
x, topk_idx, topk_weights, use_fp8=True, use_ue8m0=self.quant_config.deepgemm_scale_ue8m0
|
x, topk_idx, topk_weights, use_fp8=True, use_ue8m0=self.quant_config.deepgemm_scale_ue8m0
|
||||||
)
|
)
|
||||||
@@ -728,6 +741,10 @@ class DeepGemmFusedMoeMethod(MoEMethodBase):
|
|||||||
|
|
||||||
if shared_experts is not None:
|
if shared_experts is not None:
|
||||||
out += s_x
|
out += s_x
|
||||||
|
|
||||||
|
if fc2_latent_proj:
|
||||||
|
out = fc2_latent_proj(out)
|
||||||
|
|
||||||
return out
|
return out
|
||||||
|
|
||||||
def apply_tp(
|
def apply_tp(
|
||||||
|
|||||||
@@ -0,0 +1,108 @@
|
|||||||
|
"""
|
||||||
|
# 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 unittest
|
||||||
|
|
||||||
|
import numpy as np
|
||||||
|
import paddle
|
||||||
|
|
||||||
|
from fastdeploy.model_executor.layers.quantization.fp8_utils import deep_gemm
|
||||||
|
|
||||||
|
paddle.set_default_dtype("bfloat16")
|
||||||
|
|
||||||
|
|
||||||
|
class TestDeepGemmPrefill(unittest.TestCase):
|
||||||
|
def setUp(self):
|
||||||
|
pass
|
||||||
|
|
||||||
|
def one_invoke(self, num_experts, M, N, K):
|
||||||
|
token_num_in_eatch_batch = (paddle.zeros([num_experts], dtype="int32") + M).numpy().tolist()
|
||||||
|
total_m = sum(token_num_in_eatch_batch)
|
||||||
|
block_size = 128
|
||||||
|
|
||||||
|
raw_x = paddle.randn([total_m, K], dtype="bfloat16").cast(paddle.float8_e4m3fn)
|
||||||
|
raw_x_scale = paddle.randn([total_m, K // block_size], dtype="float32")
|
||||||
|
|
||||||
|
raw_w = paddle.randn([num_experts, N, K], dtype="bfloat16").cast(paddle.float8_e4m3fn)
|
||||||
|
raw_w_scale = paddle.randn([num_experts, N // block_size, K // block_size], dtype="float32")
|
||||||
|
|
||||||
|
m_indices = np.zeros([total_m], dtype="int32")
|
||||||
|
|
||||||
|
baseline_out = paddle.empty([total_m, N], dtype="bfloat16")
|
||||||
|
for i in range(num_experts):
|
||||||
|
start = sum(token_num_in_eatch_batch[:i])
|
||||||
|
end = start + token_num_in_eatch_batch[i]
|
||||||
|
|
||||||
|
this_expert_token = raw_x[start:end].contiguous().cast("float32")
|
||||||
|
this_expert_token_scale = (
|
||||||
|
raw_x_scale[start:end]
|
||||||
|
.contiguous()
|
||||||
|
.reshape([0, 0, 1])
|
||||||
|
.tile([1, 1, block_size])
|
||||||
|
.reshape([0, -1])
|
||||||
|
.cast("float32")
|
||||||
|
)
|
||||||
|
tmp0 = this_expert_token * this_expert_token_scale
|
||||||
|
|
||||||
|
this_expert_weight = raw_w[i].contiguous().cast("float32")
|
||||||
|
this_expert_weight_scale = (
|
||||||
|
raw_w_scale[i]
|
||||||
|
.contiguous()
|
||||||
|
.reshape([0, 1, -1, 1])
|
||||||
|
.tile([1, block_size, 1, block_size])
|
||||||
|
.reshape([N, K])
|
||||||
|
.cast("float32")
|
||||||
|
)
|
||||||
|
tmp1 = this_expert_weight * this_expert_weight_scale
|
||||||
|
|
||||||
|
out = paddle.matmul(tmp0, tmp1, False, True)
|
||||||
|
baseline_out[start:end] = out
|
||||||
|
|
||||||
|
m_indices[start:end] = i
|
||||||
|
|
||||||
|
deepgemm_output = paddle.zeros_like(baseline_out)
|
||||||
|
|
||||||
|
m_indices = paddle.to_tensor(m_indices, dtype="int32")
|
||||||
|
|
||||||
|
for i in range(10):
|
||||||
|
a = paddle.zeros([1024, 1024, 1024]) + 1
|
||||||
|
del a
|
||||||
|
deep_gemm.m_grouped_gemm_fp8_fp8_bf16_nt_contiguous(
|
||||||
|
(raw_x, raw_x_scale.transpose([1, 0]).contiguous().transpose([1, 0])),
|
||||||
|
(raw_w, raw_w_scale),
|
||||||
|
deepgemm_output,
|
||||||
|
m_indices,
|
||||||
|
)
|
||||||
|
|
||||||
|
print(baseline_out - deepgemm_output)
|
||||||
|
|
||||||
|
def test_main(self):
|
||||||
|
# import paddle.profiler as profiler
|
||||||
|
# p = profiler.Profiler(
|
||||||
|
# targets=[profiler.ProfilerTarget.CPU, profiler.ProfilerTarget.GPU],
|
||||||
|
# on_trace_ready=profiler.export_chrome_tracing("./profile_log"),
|
||||||
|
# )
|
||||||
|
# p.start()
|
||||||
|
# p.step()
|
||||||
|
|
||||||
|
self.one_invoke(48, 128 * 20, 2048, 4096)
|
||||||
|
self.one_invoke(96, 128 * 20, 2048, 2048)
|
||||||
|
|
||||||
|
# p.stop()
|
||||||
|
|
||||||
|
|
||||||
|
if __name__ == "__main__":
|
||||||
|
unittest.main()
|
||||||
Reference in New Issue
Block a user