[SpecDecode] Add spec decode support (#500)
### What this PR does / why we need it?
Backport: https://github.com/vllm-project/vllm-ascend/pull/252
This support speculative decoding in Ascend, including speculating with
a draft model、by matching n-grams in the prompt、using MLP speculators
and using EAGLE based draft models.
Backport: https://github.com/vllm-project/vllm-ascend/pull/423
spec decode MultiStepWorker support TP1DraftModelRunner fully, support
run the draft_model_runner with multi-step prepare on the NPU directly
and support draft_model_runner use MLA.
1. before this pr, `MultiStepWorker` would not step into the branch
using NPU prepare, but only into the branch using CPU prepare (`line 52`
of `vllm_ascend/patch/patch_multi_step_worker.py`). Although this has
`no effect` on the `correct operation` of speculative decoding and the
performance of the two branches is basically the same as of the current
version, I support entering this branch in this PR. In general, there
are two main changes in `patch_multi_step_worker.py`: first, the
`is_cuda_like()` check is removed and the `TP1DraftModelRunner`
rewritten in vllm_ascend is used; second, the
`supports_gpu_multi_step()` function is made to return true on NPU
devices when outer Multi_step_worker could work correct.
3. before this pr, `TP1DraftModelRunner` only supports Attention on NPU,
but not MLA. The relevant adaptation is in
`vllm_ascend/worker/draft_model_runner.py`. Although I don’t know why
the `input_positions` of `model_input.attn_metadata` in vllm-ascend
needs to be added in `execute_model`, it is done in `model_runner.py`,
so I also made corresponding changes. Otherwise, when atten_backend is
MLA, it will prompt that input_positions cannot be found.
4. I commented out two lines in `draft_model_runner.py` in `line118` to
support the scenario of K>1.
```
# lora_mapping=model_input.lora_mapping,
# lora_requests=model_input.lora_requests,
```
I added comments. In the future, when vllm-ascend supports lora feature,
the changes here can be restored.
TODO:
- [ ] revert the patch when the related issues are addressed in vllm
### How was this patch tested?
CI passed with new added test.
- e2e test for medusa proposer:
tests/singlecard/spec_decode/e2e/test_medusa_correctness.py
- e2e test for mlp proposer:
tests/singlecard/spec_decode/e2e/test_mlp_correctness.py
- e2e test for n-gram proposer:
tests/singlecard/spec_decode/e2e/test_ngram_correctness.py
Tests for patched files:
- tests/singlecard/spec_decode/test_dynamic_spec_decode.py
- tests/singlecard/spec_decode/test_multi_step_worker.py
- tests/singlecard/spec_decode/test_ngram_worker.py
- tests/singlecard/spec_decode/test_spec_decode_worker.py
---------
Signed-off-by: MengqingCao <cmq0113@163.com>
Co-authored-by: mengwei805 <mengwei25@huawei.com>
2025-04-17 20:16:32 +08:00
|
|
|
#
|
|
|
|
|
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
|
|
|
|
|
# This file is a part of the vllm-ascend project.
|
|
|
|
|
# Adapted from vllm-project/vllm/tests/utils.py
|
|
|
|
|
# Copyright 2023 The vLLM team.
|
|
|
|
|
#
|
|
|
|
|
# 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 functools
|
|
|
|
|
import os
|
|
|
|
|
import signal
|
[V1][ModelRunner] Support pooling model for v1 engine (#1359)
### What this PR does / why we need it?
Change as little existing code as possible to add v1 pooling task's
support, notice that i move down the `vllm.v1.worker.gpu_input_batch` to
vllm-ascend, Considering the frequent changes in upstream interfaces, in
order to decouple, so i move it here
### How was this patch tested?
CI passed with new added/existing test, and I have a simple test was
first conducted locally which is adapted from
https://www.modelscope.cn/models/Qwen/Qwen3-Embedding-0.6B, just like
bellow:
```python
import os
import torch
from vllm import LLM
os.environ["VLLM_USE_MODELSCOPE"]="True"
def get_detailed_instruct(task_description: str, query: str) -> str:
return f'Instruct: {task_description}\nQuery:{query}'
# Each query must come with a one-sentence instruction that describes the task
task = 'Given a web search query, retrieve relevant passages that answer the query'
queries = [
get_detailed_instruct(task, 'What is the capital of China?'),
get_detailed_instruct(task, 'Explain gravity')
]
# No need to add instruction for retrieval documents
documents = [
"The capital of China is Beijing.",
"Gravity is a force that attracts two bodies towards each other. It gives weight to physical objects and is responsible for the movement of planets around the sun."
]
input_texts = queries + documents
model = LLM(model="Qwen/Qwen3-Embedding-0.6B", task="embed")
outputs = model.embed(input_texts)
embeddings = torch.tensor([o.outputs.embedding for o in outputs])
scores = (embeddings[:2] @ embeddings[2:].T)
print(scores.tolist())
# [[0.7620252966880798, 0.14078938961029053], [0.1358368694782257, 0.6013815999031067]]
```
---------
Signed-off-by: wangli <wangli858794774@gmail.com>
Signed-off-by: wangli <858794774@qq.com>
Co-authored-by: wangli <858794774@qq.com>
2025-06-30 16:31:12 +08:00
|
|
|
from collections.abc import Sequence
|
2025-07-15 00:54:20 +08:00
|
|
|
from typing import Callable
|
[SpecDecode] Add spec decode support (#500)
### What this PR does / why we need it?
Backport: https://github.com/vllm-project/vllm-ascend/pull/252
This support speculative decoding in Ascend, including speculating with
a draft model、by matching n-grams in the prompt、using MLP speculators
and using EAGLE based draft models.
Backport: https://github.com/vllm-project/vllm-ascend/pull/423
spec decode MultiStepWorker support TP1DraftModelRunner fully, support
run the draft_model_runner with multi-step prepare on the NPU directly
and support draft_model_runner use MLA.
1. before this pr, `MultiStepWorker` would not step into the branch
using NPU prepare, but only into the branch using CPU prepare (`line 52`
of `vllm_ascend/patch/patch_multi_step_worker.py`). Although this has
`no effect` on the `correct operation` of speculative decoding and the
performance of the two branches is basically the same as of the current
version, I support entering this branch in this PR. In general, there
are two main changes in `patch_multi_step_worker.py`: first, the
`is_cuda_like()` check is removed and the `TP1DraftModelRunner`
rewritten in vllm_ascend is used; second, the
`supports_gpu_multi_step()` function is made to return true on NPU
devices when outer Multi_step_worker could work correct.
3. before this pr, `TP1DraftModelRunner` only supports Attention on NPU,
but not MLA. The relevant adaptation is in
`vllm_ascend/worker/draft_model_runner.py`. Although I don’t know why
the `input_positions` of `model_input.attn_metadata` in vllm-ascend
needs to be added in `execute_model`, it is done in `model_runner.py`,
so I also made corresponding changes. Otherwise, when atten_backend is
MLA, it will prompt that input_positions cannot be found.
4. I commented out two lines in `draft_model_runner.py` in `line118` to
support the scenario of K>1.
```
# lora_mapping=model_input.lora_mapping,
# lora_requests=model_input.lora_requests,
```
I added comments. In the future, when vllm-ascend supports lora feature,
the changes here can be restored.
TODO:
- [ ] revert the patch when the related issues are addressed in vllm
### How was this patch tested?
CI passed with new added test.
- e2e test for medusa proposer:
tests/singlecard/spec_decode/e2e/test_medusa_correctness.py
- e2e test for mlp proposer:
tests/singlecard/spec_decode/e2e/test_mlp_correctness.py
- e2e test for n-gram proposer:
tests/singlecard/spec_decode/e2e/test_ngram_correctness.py
Tests for patched files:
- tests/singlecard/spec_decode/test_dynamic_spec_decode.py
- tests/singlecard/spec_decode/test_multi_step_worker.py
- tests/singlecard/spec_decode/test_ngram_worker.py
- tests/singlecard/spec_decode/test_spec_decode_worker.py
---------
Signed-off-by: MengqingCao <cmq0113@163.com>
Co-authored-by: mengwei805 <mengwei25@huawei.com>
2025-04-17 20:16:32 +08:00
|
|
|
|
[V1][ModelRunner] Support pooling model for v1 engine (#1359)
### What this PR does / why we need it?
Change as little existing code as possible to add v1 pooling task's
support, notice that i move down the `vllm.v1.worker.gpu_input_batch` to
vllm-ascend, Considering the frequent changes in upstream interfaces, in
order to decouple, so i move it here
### How was this patch tested?
CI passed with new added/existing test, and I have a simple test was
first conducted locally which is adapted from
https://www.modelscope.cn/models/Qwen/Qwen3-Embedding-0.6B, just like
bellow:
```python
import os
import torch
from vllm import LLM
os.environ["VLLM_USE_MODELSCOPE"]="True"
def get_detailed_instruct(task_description: str, query: str) -> str:
return f'Instruct: {task_description}\nQuery:{query}'
# Each query must come with a one-sentence instruction that describes the task
task = 'Given a web search query, retrieve relevant passages that answer the query'
queries = [
get_detailed_instruct(task, 'What is the capital of China?'),
get_detailed_instruct(task, 'Explain gravity')
]
# No need to add instruction for retrieval documents
documents = [
"The capital of China is Beijing.",
"Gravity is a force that attracts two bodies towards each other. It gives weight to physical objects and is responsible for the movement of planets around the sun."
]
input_texts = queries + documents
model = LLM(model="Qwen/Qwen3-Embedding-0.6B", task="embed")
outputs = model.embed(input_texts)
embeddings = torch.tensor([o.outputs.embedding for o in outputs])
scores = (embeddings[:2] @ embeddings[2:].T)
print(scores.tolist())
# [[0.7620252966880798, 0.14078938961029053], [0.1358368694782257, 0.6013815999031067]]
```
---------
Signed-off-by: wangli <wangli858794774@gmail.com>
Signed-off-by: wangli <858794774@qq.com>
Co-authored-by: wangli <858794774@qq.com>
2025-06-30 16:31:12 +08:00
|
|
|
import torch
|
|
|
|
|
import torch.nn.functional as F
|
[SpecDecode] Add spec decode support (#500)
### What this PR does / why we need it?
Backport: https://github.com/vllm-project/vllm-ascend/pull/252
This support speculative decoding in Ascend, including speculating with
a draft model、by matching n-grams in the prompt、using MLP speculators
and using EAGLE based draft models.
Backport: https://github.com/vllm-project/vllm-ascend/pull/423
spec decode MultiStepWorker support TP1DraftModelRunner fully, support
run the draft_model_runner with multi-step prepare on the NPU directly
and support draft_model_runner use MLA.
1. before this pr, `MultiStepWorker` would not step into the branch
using NPU prepare, but only into the branch using CPU prepare (`line 52`
of `vllm_ascend/patch/patch_multi_step_worker.py`). Although this has
`no effect` on the `correct operation` of speculative decoding and the
performance of the two branches is basically the same as of the current
version, I support entering this branch in this PR. In general, there
are two main changes in `patch_multi_step_worker.py`: first, the
`is_cuda_like()` check is removed and the `TP1DraftModelRunner`
rewritten in vllm_ascend is used; second, the
`supports_gpu_multi_step()` function is made to return true on NPU
devices when outer Multi_step_worker could work correct.
3. before this pr, `TP1DraftModelRunner` only supports Attention on NPU,
but not MLA. The relevant adaptation is in
`vllm_ascend/worker/draft_model_runner.py`. Although I don’t know why
the `input_positions` of `model_input.attn_metadata` in vllm-ascend
needs to be added in `execute_model`, it is done in `model_runner.py`,
so I also made corresponding changes. Otherwise, when atten_backend is
MLA, it will prompt that input_positions cannot be found.
4. I commented out two lines in `draft_model_runner.py` in `line118` to
support the scenario of K>1.
```
# lora_mapping=model_input.lora_mapping,
# lora_requests=model_input.lora_requests,
```
I added comments. In the future, when vllm-ascend supports lora feature,
the changes here can be restored.
TODO:
- [ ] revert the patch when the related issues are addressed in vllm
### How was this patch tested?
CI passed with new added test.
- e2e test for medusa proposer:
tests/singlecard/spec_decode/e2e/test_medusa_correctness.py
- e2e test for mlp proposer:
tests/singlecard/spec_decode/e2e/test_mlp_correctness.py
- e2e test for n-gram proposer:
tests/singlecard/spec_decode/e2e/test_ngram_correctness.py
Tests for patched files:
- tests/singlecard/spec_decode/test_dynamic_spec_decode.py
- tests/singlecard/spec_decode/test_multi_step_worker.py
- tests/singlecard/spec_decode/test_ngram_worker.py
- tests/singlecard/spec_decode/test_spec_decode_worker.py
---------
Signed-off-by: MengqingCao <cmq0113@163.com>
Co-authored-by: mengwei805 <mengwei25@huawei.com>
2025-04-17 20:16:32 +08:00
|
|
|
from typing_extensions import ParamSpec
|
|
|
|
|
|
|
|
|
|
_P = ParamSpec("_P")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def fork_new_process_for_each_test(
|
|
|
|
|
f: Callable[_P, None]) -> Callable[_P, None]:
|
|
|
|
|
"""Decorator to fork a new process for each test function.
|
|
|
|
|
See https://github.com/vllm-project/vllm/issues/7053 for more details.
|
|
|
|
|
"""
|
|
|
|
|
|
|
|
|
|
@functools.wraps(f)
|
|
|
|
|
def wrapper(*args: _P.args, **kwargs: _P.kwargs) -> None:
|
|
|
|
|
# Make the process the leader of its own process group
|
|
|
|
|
# to avoid sending SIGTERM to the parent process
|
|
|
|
|
os.setpgrp()
|
|
|
|
|
from _pytest.outcomes import Skipped
|
|
|
|
|
pid = os.fork()
|
|
|
|
|
print(f"Fork a new process to run a test {pid}")
|
|
|
|
|
if pid == 0:
|
|
|
|
|
try:
|
|
|
|
|
f(*args, **kwargs)
|
|
|
|
|
except Skipped as e:
|
|
|
|
|
# convert Skipped to exit code 0
|
|
|
|
|
print(str(e))
|
|
|
|
|
os._exit(0)
|
|
|
|
|
except Exception:
|
|
|
|
|
import traceback
|
|
|
|
|
traceback.print_exc()
|
|
|
|
|
os._exit(1)
|
|
|
|
|
else:
|
|
|
|
|
os._exit(0)
|
|
|
|
|
else:
|
|
|
|
|
pgid = os.getpgid(pid)
|
|
|
|
|
_pid, _exitcode = os.waitpid(pid, 0)
|
|
|
|
|
# ignore SIGTERM signal itself
|
|
|
|
|
old_signal_handler = signal.signal(signal.SIGTERM, signal.SIG_IGN)
|
|
|
|
|
# kill all child processes
|
|
|
|
|
os.killpg(pgid, signal.SIGTERM)
|
|
|
|
|
# restore the signal handler
|
|
|
|
|
signal.signal(signal.SIGTERM, old_signal_handler)
|
|
|
|
|
assert _exitcode == 0, (f"function {f} failed when called with"
|
|
|
|
|
f" args {args} and kwargs {kwargs}")
|
|
|
|
|
|
|
|
|
|
return wrapper
|
[V1][ModelRunner] Support pooling model for v1 engine (#1359)
### What this PR does / why we need it?
Change as little existing code as possible to add v1 pooling task's
support, notice that i move down the `vllm.v1.worker.gpu_input_batch` to
vllm-ascend, Considering the frequent changes in upstream interfaces, in
order to decouple, so i move it here
### How was this patch tested?
CI passed with new added/existing test, and I have a simple test was
first conducted locally which is adapted from
https://www.modelscope.cn/models/Qwen/Qwen3-Embedding-0.6B, just like
bellow:
```python
import os
import torch
from vllm import LLM
os.environ["VLLM_USE_MODELSCOPE"]="True"
def get_detailed_instruct(task_description: str, query: str) -> str:
return f'Instruct: {task_description}\nQuery:{query}'
# Each query must come with a one-sentence instruction that describes the task
task = 'Given a web search query, retrieve relevant passages that answer the query'
queries = [
get_detailed_instruct(task, 'What is the capital of China?'),
get_detailed_instruct(task, 'Explain gravity')
]
# No need to add instruction for retrieval documents
documents = [
"The capital of China is Beijing.",
"Gravity is a force that attracts two bodies towards each other. It gives weight to physical objects and is responsible for the movement of planets around the sun."
]
input_texts = queries + documents
model = LLM(model="Qwen/Qwen3-Embedding-0.6B", task="embed")
outputs = model.embed(input_texts)
embeddings = torch.tensor([o.outputs.embedding for o in outputs])
scores = (embeddings[:2] @ embeddings[2:].T)
print(scores.tolist())
# [[0.7620252966880798, 0.14078938961029053], [0.1358368694782257, 0.6013815999031067]]
```
---------
Signed-off-by: wangli <wangli858794774@gmail.com>
Signed-off-by: wangli <858794774@qq.com>
Co-authored-by: wangli <858794774@qq.com>
2025-06-30 16:31:12 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
def matryoshka_fy(tensor: torch.Tensor, dimensions: int):
|
|
|
|
|
tensor = torch.tensor(tensor)
|
|
|
|
|
tensor = tensor[..., :dimensions]
|
|
|
|
|
tensor = F.normalize(tensor, p=2, dim=1)
|
|
|
|
|
return tensor
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
def check_embeddings_close(
|
|
|
|
|
*,
|
|
|
|
|
embeddings_0_lst: Sequence[list[float]],
|
|
|
|
|
embeddings_1_lst: Sequence[list[float]],
|
|
|
|
|
name_0: str,
|
|
|
|
|
name_1: str,
|
|
|
|
|
tol: float = 1e-3,
|
|
|
|
|
) -> None:
|
|
|
|
|
assert len(embeddings_0_lst) == len(embeddings_1_lst)
|
|
|
|
|
|
|
|
|
|
for prompt_idx, (embeddings_0, embeddings_1) in enumerate(
|
|
|
|
|
zip(embeddings_0_lst, embeddings_1_lst)):
|
|
|
|
|
assert len(embeddings_0) == len(embeddings_1), (
|
|
|
|
|
f"Length mismatch: {len(embeddings_0)} vs. {len(embeddings_1)}")
|
|
|
|
|
|
|
|
|
|
sim = F.cosine_similarity(torch.tensor(embeddings_0),
|
|
|
|
|
torch.tensor(embeddings_1),
|
|
|
|
|
dim=0)
|
|
|
|
|
|
|
|
|
|
fail_msg = (f"Test{prompt_idx}:"
|
|
|
|
|
f"\nCosine similarity: \t{sim:.4f}"
|
|
|
|
|
f"\n{name_0}:\t{embeddings_0[:16]!r}"
|
|
|
|
|
f"\n{name_1}:\t{embeddings_1[:16]!r}")
|
|
|
|
|
|
|
|
|
|
assert sim >= 1 - tol, fail_msg
|