178ca1607e
### What this PR does / why we need it?
The main goal of this PR to alleviate the high maintenance burden from
model duplication when we are going to do the model optimization. Some
of our optimized models diverges a little from the vllm's modeling, but
needs to rewrite several part of original one, brings negligible
maintenance bruden to the vllm-ascend.In order to solve that, we propose
to leverage `torch.compile` and `inductor pattern matcher`,
automatically fuse the pattern we want to merge. For more details can
refer to the RFC https://github.com/vllm-project/vllm-ascend/issues/4239
This pr integrates `AddRMSNorm` and the `Quant` operator, which can
improve the inference speed of models using `w8a8 `quantization.
### Does this PR introduce _any_ user-facing change?
Yes, add new additional_config
### How was this patch tested?
```python
def main():
prompts = [
"The president of the United States is Mr.",
]
# Create a sampling params object.
sampling_params = SamplingParams(max_tokens=100, temperature=0.6, top_k=40, top_p=0.95)
# Create an LLM.
llm = LLM(
model="/root/.cache/modelscope/hub/models/vllm-ascend/Qwen3-8B-W8A8",
# enforce_eager=True,
tensor_parallel_size=1,
trust_remote_code=True,
gpu_memory_utilization=0.7,
quantization="ascend",
)
# Generate texts from the prompts.
outputs = llm.generate(prompts, sampling_params)
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```
```text
Prompt: 'The president of the United States is Mr.', Generated text: ' Trump. The president of the United States is Mr. Biden. Which of the following statements is correct? \n\nA. Mr. Trump is Mr. Biden. \nB. Mr. Trump is not Mr. Biden. \nC. The president of the United States is not Mr. Trump. \nD. The president of the United States is not Mr. Biden.\n\nThe question presents a contradiction: it states that "The president of the United States is Mr. Trump" and "The president of'
```
- vLLM version: 86e178f7c4d8c3b0eaf3c8e3f810a83f63b90e24
- vLLM main:
86e178f7c4
---------
Signed-off-by: Icey <1790571317@qq.com>
Signed-off-by: wxsIcey <1790571317@qq.com>
114 lines
4.2 KiB
Python
114 lines
4.2 KiB
Python
#
|
|
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
|
|
# This file is a part of the vllm-ascend project.
|
|
#
|
|
#
|
|
# 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 logging
|
|
|
|
import torch
|
|
import torch._inductor.pattern_matcher as pm
|
|
from torch._inductor.pattern_matcher import PatternMatcherPass
|
|
from vllm.compilation.vllm_inductor_pass import VllmInductorPass
|
|
from vllm.config import VllmConfig
|
|
|
|
|
|
class AddRMSNormQuantPattern:
|
|
|
|
def __init__(self, vllm_config: VllmConfig, eps: float = 1e-6):
|
|
self.vllm_config = vllm_config
|
|
self.eps = eps
|
|
|
|
def get_inputs(self):
|
|
"""
|
|
Generate example inputs for the AddRMSNormQuant fusion pattern.
|
|
"""
|
|
rms_norm_input = torch.randn(2, 4, device="npu")
|
|
residual = torch.randn(2, 4, device="npu")
|
|
rms_norm_weight = torch.randn(4, device="npu")
|
|
scale = torch.tensor([1.0], device="npu")
|
|
offset = torch.tensor([0.0], device="npu")
|
|
return [rms_norm_input, residual, rms_norm_weight, scale, offset]
|
|
|
|
def register(self, pm_pass: PatternMatcherPass):
|
|
|
|
def pattern(rms_norm_input: torch.Tensor, residual: torch.Tensor,
|
|
rms_norm_weight: torch.Tensor, scale: torch.Tensor,
|
|
offset: torch.Tensor):
|
|
"""
|
|
Pattern for AddRMSNormQuant fusion.
|
|
"""
|
|
output = torch.ops.npu.npu_add_rms_norm(rms_norm_input, residual,
|
|
rms_norm_weight, self.eps)
|
|
out0 = output[0]
|
|
out1 = output[2]
|
|
quantized_output = torch.ops.npu.npu_quantize(
|
|
out0, scale, offset, torch.qint8, -1, False)
|
|
return quantized_output, out1
|
|
|
|
def replacement(rms_norm_input: torch.Tensor, residual: torch.Tensor,
|
|
rms_norm_weight: torch.Tensor, scale: torch.Tensor,
|
|
offset: torch.Tensor):
|
|
"""
|
|
Replacement for the AddRMSNormQuant fusion.
|
|
"""
|
|
output = torch.ops.npu.npu_add_rms_norm_quant(
|
|
rms_norm_input,
|
|
residual,
|
|
rms_norm_weight,
|
|
1. /
|
|
scale, # The inverse of scale is required by npu_add_rms_norm_quant kernel which is opposite to the npu_quantize kernel.
|
|
offset,
|
|
epsilon=self.eps)
|
|
quantized_output = output[0]
|
|
out1 = output[2]
|
|
return quantized_output, out1
|
|
|
|
pm.register_replacement(pattern, replacement, self.get_inputs(),
|
|
pm.fwd_only, pm_pass)
|
|
|
|
|
|
class AddRMSNormQuantFusionPass(VllmInductorPass):
|
|
"""
|
|
A pass for fusing AddRMSNorm and W8A8 quantization operations on Ascend.
|
|
"""
|
|
|
|
def __init__(self, vllm_config: VllmConfig):
|
|
super().__init__(vllm_config)
|
|
self.pattern_match_passes: PatternMatcherPass = PatternMatcherPass(
|
|
pass_name="rmsnorm_quant_fusion_pass")
|
|
|
|
dtype = vllm_config.model_config.dtype
|
|
if dtype not in (torch.bfloat16, torch.float16):
|
|
logging.info("Quant fusion not enabled: unsupported dtype %s",
|
|
dtype)
|
|
return
|
|
|
|
common_epsilons = [1e-5, 1e-6]
|
|
for eps in common_epsilons:
|
|
AddRMSNormQuantPattern(vllm_config,
|
|
eps=eps).register(self.pattern_match_passes)
|
|
|
|
def __call__(self, graph: torch.fx.Graph):
|
|
self.begin()
|
|
self.matched_count = self.pattern_match_passes.apply(graph)
|
|
logging.debug("Replaced %s patterns", self.matched_count)
|
|
self.end_and_log()
|
|
|
|
def is_applicable(self, runtime_shape: int | None = None) -> bool:
|
|
"""
|
|
Check if the pass is applicable for the current configuration.
|
|
"""
|
|
return True
|