Adopt inductor fusion and define quantization fusion pass (#4168)
### 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:
https://github.com/vllm-project/vllm/commit/86e178f7c4d8c3b0eaf3c8e3f810a83f63b90e24
---------
Signed-off-by: Icey <1790571317@qq.com>
Signed-off-by: wxsIcey <1790571317@qq.com>
2025-12-04 10:29:48 +08:00
|
|
|
#
|
|
|
|
|
# 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.
|
|
|
|
|
#
|
|
|
|
|
|
|
|
|
|
from torch import fx as fx
|
2026-02-27 16:05:21 +08:00
|
|
|
from vllm.compilation.passes.inductor_pass import get_pass_context
|
|
|
|
|
from vllm.compilation.passes.vllm_inductor_pass import VllmInductorPass
|
Adopt inductor fusion and define quantization fusion pass (#4168)
### 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:
https://github.com/vllm-project/vllm/commit/86e178f7c4d8c3b0eaf3c8e3f810a83f63b90e24
---------
Signed-off-by: Icey <1790571317@qq.com>
Signed-off-by: wxsIcey <1790571317@qq.com>
2025-12-04 10:29:48 +08:00
|
|
|
from vllm.config import VllmConfig
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
class GraphFusionPassManager:
|
|
|
|
|
"""
|
|
|
|
|
A pass manager for graph fusion passes.
|
|
|
|
|
It handles the configuration and execution of passes.
|
[Lint]Style: Convert `vllm-ascend/compilation` to ruff format (#5912)
### What this PR does / why we need it?
Convert `vllm-ascend/compilation` to ruff format.
### Does this PR introduce _any_ user-facing change?
During this migration, we encountered some **errors** in our CI and
testing environments, such as:
```
vllm_ascend/utils.py:653: in <module>
def register_ascend_customop(vllm_config: VllmConfig | None = None):
^^^^^^^^^^^^^^^^^
E TypeError: unsupported operand type(s) for |: 'NoneType' and 'NoneType'
```
**1. Root Cause Analysis:**
The project uses a common pattern to break circular dependencies:
```python
if TYPE_CHECKING:
from vllm.config import VllmConfig
else:
VllmConfig = None # Placeholder assigned at runtime
```
When Python parses the function definition `def
register_ascend_customop(vllm_config: VllmConfig | None)`, it attempts
to evaluate the expression `VllmConfig | None`.
Since `VllmConfig` is assigned `None` at runtime, the expression
effectively becomes `None | None`. In Python, `None` is an instance of
`NoneType`. While the `|` operator is implemented for Type objects
(classes), it is not supported for `NoneType` instances, leading to the
`TypeError` shown above.
**2. Solution:**
To maintain the modern `|` syntax required by our new linting standards
while preserving our dependency management strategy, I have introduced:
```python
from __future__ import annotations
```
at the top of the affected files. This enables **Postponed Evaluation of
Annotations (PEP 563)**.
**3. Impact and Benefits:**
- By enabling `annotations`, Python no longer executes the `VllmConfig |
None` operation during module load. Instead, it stores the annotation as
a string literal, completely avoiding the `None | None` calculation.
- We can keep the `VllmConfig = None` placeholders. This ensures that
other modules can still import these symbols without triggering an
`ImportError`, maintaining a stable dependency graph.
- IDEs and static type checkers (MyPy/Pyright) continue to resolve the
types correctly. This allows us to use modern syntax without sacrificing
type safety or runtime stability.
- The only side effect is that `__annotations__` will now return strings
instead of type objects. Since this module does not use runtime type
enforcement or reflection, this change has zero negative impact on
existing functionality.
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/11b6af5280d6d6dfb8953af16e67b25f819b3be9
---------
Signed-off-by: MrZ20 <2609716663@qq.com>
2026-01-16 20:57:46 +08:00
|
|
|
The counterpart in vllm is PostGradPassManager. Since torch_npu
|
Adopt inductor fusion and define quantization fusion pass (#4168)
### 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:
https://github.com/vllm-project/vllm/commit/86e178f7c4d8c3b0eaf3c8e3f810a83f63b90e24
---------
Signed-off-by: Icey <1790571317@qq.com>
Signed-off-by: wxsIcey <1790571317@qq.com>
2025-12-04 10:29:48 +08:00
|
|
|
does not support triton for now, we define our own pass manager.
|
|
|
|
|
"""
|
|
|
|
|
|
|
|
|
|
def __init__(self):
|
|
|
|
|
self.passes: list[VllmInductorPass] = []
|
|
|
|
|
|
2026-01-07 18:42:55 +08:00
|
|
|
def __call__(self, graph: fx.Graph) -> fx.Graph:
|
|
|
|
|
compile_range = get_pass_context().compile_range
|
|
|
|
|
|
Adopt inductor fusion and define quantization fusion pass (#4168)
### 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:
https://github.com/vllm-project/vllm/commit/86e178f7c4d8c3b0eaf3c8e3f810a83f63b90e24
---------
Signed-off-by: Icey <1790571317@qq.com>
Signed-off-by: wxsIcey <1790571317@qq.com>
2025-12-04 10:29:48 +08:00
|
|
|
for pass_ in self.passes:
|
2026-01-07 18:42:55 +08:00
|
|
|
if pass_.is_applicable_for_range(compile_range):
|
Adopt inductor fusion and define quantization fusion pass (#4168)
### 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:
https://github.com/vllm-project/vllm/commit/86e178f7c4d8c3b0eaf3c8e3f810a83f63b90e24
---------
Signed-off-by: Icey <1790571317@qq.com>
Signed-off-by: wxsIcey <1790571317@qq.com>
2025-12-04 10:29:48 +08:00
|
|
|
pass_(graph)
|
2026-01-07 18:42:55 +08:00
|
|
|
graph.recompile()
|
Adopt inductor fusion and define quantization fusion pass (#4168)
### 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:
https://github.com/vllm-project/vllm/commit/86e178f7c4d8c3b0eaf3c8e3f810a83f63b90e24
---------
Signed-off-by: Icey <1790571317@qq.com>
Signed-off-by: wxsIcey <1790571317@qq.com>
2025-12-04 10:29:48 +08:00
|
|
|
return graph
|
|
|
|
|
|
|
|
|
|
def add(self, pass_: VllmInductorPass):
|
|
|
|
|
assert isinstance(pass_, VllmInductorPass)
|
|
|
|
|
self.passes.append(pass_)
|
|
|
|
|
|
|
|
|
|
def configure(self, config: VllmConfig):
|
|
|
|
|
# By default, we enable the graph fusion and quantization fusion pass.
|
[Lint]Style: Convert `vllm-ascend/compilation` to ruff format (#5912)
### What this PR does / why we need it?
Convert `vllm-ascend/compilation` to ruff format.
### Does this PR introduce _any_ user-facing change?
During this migration, we encountered some **errors** in our CI and
testing environments, such as:
```
vllm_ascend/utils.py:653: in <module>
def register_ascend_customop(vllm_config: VllmConfig | None = None):
^^^^^^^^^^^^^^^^^
E TypeError: unsupported operand type(s) for |: 'NoneType' and 'NoneType'
```
**1. Root Cause Analysis:**
The project uses a common pattern to break circular dependencies:
```python
if TYPE_CHECKING:
from vllm.config import VllmConfig
else:
VllmConfig = None # Placeholder assigned at runtime
```
When Python parses the function definition `def
register_ascend_customop(vllm_config: VllmConfig | None)`, it attempts
to evaluate the expression `VllmConfig | None`.
Since `VllmConfig` is assigned `None` at runtime, the expression
effectively becomes `None | None`. In Python, `None` is an instance of
`NoneType`. While the `|` operator is implemented for Type objects
(classes), it is not supported for `NoneType` instances, leading to the
`TypeError` shown above.
**2. Solution:**
To maintain the modern `|` syntax required by our new linting standards
while preserving our dependency management strategy, I have introduced:
```python
from __future__ import annotations
```
at the top of the affected files. This enables **Postponed Evaluation of
Annotations (PEP 563)**.
**3. Impact and Benefits:**
- By enabling `annotations`, Python no longer executes the `VllmConfig |
None` operation during module load. Instead, it stores the annotation as
a string literal, completely avoiding the `None | None` calculation.
- We can keep the `VllmConfig = None` placeholders. This ensures that
other modules can still import these symbols without triggering an
`ImportError`, maintaining a stable dependency graph.
- IDEs and static type checkers (MyPy/Pyright) continue to resolve the
types correctly. This allows us to use modern syntax without sacrificing
type safety or runtime stability.
- The only side effect is that `__annotations__` will now return strings
instead of type objects. Since this module does not use runtime type
enforcement or reflection, this change has zero negative impact on
existing functionality.
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/11b6af5280d6d6dfb8953af16e67b25f819b3be9
---------
Signed-off-by: MrZ20 <2609716663@qq.com>
2026-01-16 20:57:46 +08:00
|
|
|
self.ascend_compilation_config: dict = config.additional_config.get("ascend_compilation_config", {})
|
2025-12-11 17:53:43 +08:00
|
|
|
if self.ascend_compilation_config.get("fuse_norm_quant", True):
|
[Lint]Style: Convert `vllm-ascend/compilation` to ruff format (#5912)
### What this PR does / why we need it?
Convert `vllm-ascend/compilation` to ruff format.
### Does this PR introduce _any_ user-facing change?
During this migration, we encountered some **errors** in our CI and
testing environments, such as:
```
vllm_ascend/utils.py:653: in <module>
def register_ascend_customop(vllm_config: VllmConfig | None = None):
^^^^^^^^^^^^^^^^^
E TypeError: unsupported operand type(s) for |: 'NoneType' and 'NoneType'
```
**1. Root Cause Analysis:**
The project uses a common pattern to break circular dependencies:
```python
if TYPE_CHECKING:
from vllm.config import VllmConfig
else:
VllmConfig = None # Placeholder assigned at runtime
```
When Python parses the function definition `def
register_ascend_customop(vllm_config: VllmConfig | None)`, it attempts
to evaluate the expression `VllmConfig | None`.
Since `VllmConfig` is assigned `None` at runtime, the expression
effectively becomes `None | None`. In Python, `None` is an instance of
`NoneType`. While the `|` operator is implemented for Type objects
(classes), it is not supported for `NoneType` instances, leading to the
`TypeError` shown above.
**2. Solution:**
To maintain the modern `|` syntax required by our new linting standards
while preserving our dependency management strategy, I have introduced:
```python
from __future__ import annotations
```
at the top of the affected files. This enables **Postponed Evaluation of
Annotations (PEP 563)**.
**3. Impact and Benefits:**
- By enabling `annotations`, Python no longer executes the `VllmConfig |
None` operation during module load. Instead, it stores the annotation as
a string literal, completely avoiding the `None | None` calculation.
- We can keep the `VllmConfig = None` placeholders. This ensures that
other modules can still import these symbols without triggering an
`ImportError`, maintaining a stable dependency graph.
- IDEs and static type checkers (MyPy/Pyright) continue to resolve the
types correctly. This allows us to use modern syntax without sacrificing
type safety or runtime stability.
- The only side effect is that `__annotations__` will now return strings
instead of type objects. Since this module does not use runtime type
enforcement or reflection, this change has zero negative impact on
existing functionality.
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/11b6af5280d6d6dfb8953af16e67b25f819b3be9
---------
Signed-off-by: MrZ20 <2609716663@qq.com>
2026-01-16 20:57:46 +08:00
|
|
|
from .passes.norm_quant_fusion_pass import AddRMSNormQuantFusionPass
|
|
|
|
|
|
Adopt inductor fusion and define quantization fusion pass (#4168)
### 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:
https://github.com/vllm-project/vllm/commit/86e178f7c4d8c3b0eaf3c8e3f810a83f63b90e24
---------
Signed-off-by: Icey <1790571317@qq.com>
Signed-off-by: wxsIcey <1790571317@qq.com>
2025-12-04 10:29:48 +08:00
|
|
|
self.passes.append(AddRMSNormQuantFusionPass(config))
|
2025-12-17 08:53:44 +08:00
|
|
|
|
|
|
|
|
if self.ascend_compilation_config.get("fuse_qknorm_rope", True):
|
|
|
|
|
from .passes.qknorm_rope_fusion_pass import QKNormRopeFusionPass
|
[Lint]Style: Convert `vllm-ascend/compilation` to ruff format (#5912)
### What this PR does / why we need it?
Convert `vllm-ascend/compilation` to ruff format.
### Does this PR introduce _any_ user-facing change?
During this migration, we encountered some **errors** in our CI and
testing environments, such as:
```
vllm_ascend/utils.py:653: in <module>
def register_ascend_customop(vllm_config: VllmConfig | None = None):
^^^^^^^^^^^^^^^^^
E TypeError: unsupported operand type(s) for |: 'NoneType' and 'NoneType'
```
**1. Root Cause Analysis:**
The project uses a common pattern to break circular dependencies:
```python
if TYPE_CHECKING:
from vllm.config import VllmConfig
else:
VllmConfig = None # Placeholder assigned at runtime
```
When Python parses the function definition `def
register_ascend_customop(vllm_config: VllmConfig | None)`, it attempts
to evaluate the expression `VllmConfig | None`.
Since `VllmConfig` is assigned `None` at runtime, the expression
effectively becomes `None | None`. In Python, `None` is an instance of
`NoneType`. While the `|` operator is implemented for Type objects
(classes), it is not supported for `NoneType` instances, leading to the
`TypeError` shown above.
**2. Solution:**
To maintain the modern `|` syntax required by our new linting standards
while preserving our dependency management strategy, I have introduced:
```python
from __future__ import annotations
```
at the top of the affected files. This enables **Postponed Evaluation of
Annotations (PEP 563)**.
**3. Impact and Benefits:**
- By enabling `annotations`, Python no longer executes the `VllmConfig |
None` operation during module load. Instead, it stores the annotation as
a string literal, completely avoiding the `None | None` calculation.
- We can keep the `VllmConfig = None` placeholders. This ensures that
other modules can still import these symbols without triggering an
`ImportError`, maintaining a stable dependency graph.
- IDEs and static type checkers (MyPy/Pyright) continue to resolve the
types correctly. This allows us to use modern syntax without sacrificing
type safety or runtime stability.
- The only side effect is that `__annotations__` will now return strings
instead of type objects. Since this module does not use runtime type
enforcement or reflection, this change has zero negative impact on
existing functionality.
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/11b6af5280d6d6dfb8953af16e67b25f819b3be9
---------
Signed-off-by: MrZ20 <2609716663@qq.com>
2026-01-16 20:57:46 +08:00
|
|
|
|
2025-12-17 08:53:44 +08:00
|
|
|
self.passes.append(QKNormRopeFusionPass(config))
|
2026-01-19 09:28:07 +08:00
|
|
|
|
|
|
|
|
if self.ascend_compilation_config.get("fuse_allreduce_rms", True):
|
|
|
|
|
from .passes.allreduce_rmsnorm_fusion_pass import MatmulAllReduceAddRMSNormPass
|
|
|
|
|
|
|
|
|
|
self.passes.append(MatmulAllReduceAddRMSNormPass(config))
|
2026-02-27 08:27:41 +08:00
|
|
|
|
2026-03-02 17:54:25 +08:00
|
|
|
if self.ascend_compilation_config.get("fuse_muls_add", True):
|
|
|
|
|
from .passes.muls_add_pass import MulsAddFusionPass
|
|
|
|
|
|
|
|
|
|
self.passes.append(MulsAddFusionPass(config))
|
|
|
|
|
|
2026-02-27 08:27:41 +08:00
|
|
|
if config.compilation_config.pass_config.enable_sp:
|
2026-03-24 17:16:00 +08:00
|
|
|
from .passes.sequence_parallelism import SequenceParallelismPass
|
|
|
|
|
from .passes.sequence_parallelism_moe import SequenceParallelismMoePass
|
2026-02-27 08:27:41 +08:00
|
|
|
|
2026-03-24 17:16:00 +08:00
|
|
|
self.passes.append(SequenceParallelismPass(config))
|
|
|
|
|
self.passes.append(SequenceParallelismMoePass(config))
|