### What this PR does / why we need it?
Rename offline_inference_npu_v1.py to offline_inference_npu.py to
recover doctest
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
CI passed
- vLLM version: v0.9.2
- vLLM main:
a8593237c0
Signed-off-by: Yikun Jiang <yikunkero@gmail.com>
### What this PR does / why we need it?
When all_reduce_merge is in progress, shared_experts does not do
all_reduce in mlp, but waits until shared_experts+router_experts are
completed before doing all_reduce
In prefill and decode, as long as shared_experts+router_experts are
all_reduce, there will be benefits.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
bash examples/run_dp_attention_etp16.sh
bash examples/run_dp_attention_etp16_benmark.sh
- vLLM version: v0.9.1
- vLLM main:
977180c912
---------
Signed-off-by: ttanzhiqiang <389825161@qq.com>
### What this PR does / why we need it?
Improve
- Keep the same file name format as v1, `offline_inference_npu_v0.py`,
`offline_inference_npu_v1.py`
- Use `VLLM_USE_V1` = 0/1 clearly in py scripts
- Fix some run errors in `offline_inference_npu_v1.py`, e.g.
`deepseekv3-lite-base-latest` not exists in modescope or hf.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
- vLLM version: v0.9.2
- vLLM main:
baed180aa0
Signed-off-by: xleoken <xleoken@163.com>
### What this PR does / why we need it?
This PR implements the DeepSeek Expert Parallel Load Balancing (EPLB)
strategy to optimize expert distribution in vllm-ascend. The
implementation:
- Adapts the expert-map format to work with vllm-ascend's architecture
- Provides DeepSeek-provided mechanism to balance expert workload across
devices
### Does this PR introduce _any_ user-facing change?
This PR adds a new script that allows users to:
- Generate expert map configurations based on workload analysis
- Optimize expert distribution for their specific use case
### How was this patch tested?
To use this feature:
1. First collect expert heat information during model execution
2. Run the provided script to generate the expert map configuration
3. Apply the generated configuration to your vllm-ascend deployment
User example:
```bash
# expert_load_view.pt: dumped expert heat info file
python3 examples/eplb/eplb_strategy.py --exp_name 'deepseek_demo' \
--input_path expert_load_view.pt --output_path examples/eplb/results/demo \
--num_nodes 4
```
---------
Signed-off-by: ZhengWG <zwg0606@gmail.com>
### What this PR does / why we need it?
Fix lint
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
CI passed
Signed-off-by: Yikun Jiang <yikunkero@gmail.com>
### What this PR does / why we need it?
The `os.environ["VLLM_USE_MODELSCOPE"] = "True"` should be placed before
module imports
if not
```
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
File "/xleoken/projects/vllm-ascend/examples/offline_embed.py", line 48, in <module>
model = LLM(model="Qwen/Qwen3-Embedding-0.6B", task="embed")
File "/usr/local/python3.10.17/lib/python3.10/site-packages/vllm/entrypoints/llm.py", line 243, in __init__
self.llm_engine = LLMEngine.from_engine_args(
File "/usr/local/python3.10.17/lib/python3.10/site-packages/vllm/engine/llm_engine.py", line 494, in from_engine_args
vllm_config = engine_args.create_engine_config(usage_context)
File "/usr/local/python3.10.17/lib/python3.10/site-packages/vllm/engine/arg_utils.py", line 1018, in create_engine_config
model_config = self.create_model_config()
File "/usr/local/python3.10.17/lib/python3.10/site-packages/vllm/engine/arg_utils.py", line 910, in create_model_config
return ModelConfig(
File "/usr/local/python3.10.17/lib/python3.10/site-packages/pydantic/_internal/_dataclasses.py", line 120, in __init__
s.__pydantic_validator__.validate_python(ArgsKwargs(args, kwargs), self_instance=s)
File "/usr/local/python3.10.17/lib/python3.10/site-packages/vllm/config.py", line 528, in __post_init__
hf_config = get_config(self.hf_config_path or self.model,
File "/usr/local/python3.10.17/lib/python3.10/site-packages/vllm/transformers_utils/config.py", line 321, in get_config
config_dict, _ = PretrainedConfig.get_config_dict(
File "/usr/local/python3.10.17/lib/python3.10/site-packages/transformers/configuration_utils.py", line 590, in get_config_dict
config_dict, kwargs = cls._get_config_dict(pretrained_model_name_or_path, **kwargs)
File "/usr/local/python3.10.17/lib/python3.10/site-packages/transformers/configuration_utils.py", line 649, in _get_config_dict
resolved_config_file = cached_file(
File "/usr/local/python3.10.17/lib/python3.10/site-packages/transformers/utils/hub.py", line 266, in cached_file
file = cached_files(path_or_repo_id=path_or_repo_id, filenames=[filename], **kwargs)
File "/usr/local/python3.10.17/lib/python3.10/site-packages/transformers/utils/hub.py", line 491, in cached_files
raise OSError(
OSError: We couldn't connect to 'https://huggingface.co' to load the files, and couldn't find them in the cached files.
Checkout your internet connection or see how to run the library in offline mode at 'https://huggingface.co/docs/transformers/installation#offline-mode'.
[ERROR] 2025-07-03-15:27:10 (PID:333665, Device:-1, RankID:-1) ERR99999 UNKNOWN applicaiton exception
```
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Local.
Signed-off-by: xleoken <xleoken@163.com>
### 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>
### What this PR does / why we need it?
Add `max_num_tokens_across_dp` to AscendMetadata to fix dp
This pr fixes the bug introduced by
https://github.com/vllm-project/vllm-ascend/pull/1229, which add an arg
`max_num_tokens_across_dp` when dp_size > 1.
Signed-off-by: MengqingCao <cmq0113@163.com>
### What this PR does / why we need it?
fix bug in 1p1d disaggregated_prefill example
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Tested with python find_device_ips.py and run disaggregated_prefill
example
<!-- Thanks for sending a pull request!
BEFORE SUBMITTING, PLEASE READ
https://docs.vllm.ai/en/latest/contributing/overview.html
-->
Signed-off-by: wangyanhui-cmss <wangyanhui_yewu@cmss.chinamobile.com>
### What this PR does / why we need it?
Single machine 16 cards deepseekr1 attention (tp8/dp2) / moe(etp) Best
performance
rely on:
vllm-ascend commit id:da9acfca6053352730fce75fb772e214755d0341
vllm commit id:b124e1085b1bf977e3dac96d99ffd9d8ddfdb6cc
+ https://github.com/vllm-project/vllm-ascend/pull/910 + [Reduce
_npu_flash_attention mask to 128x128 for memory savings]
https://github.com/vllm-project/vllm-ascend/pull/1100+ [Reduce memory
usage by splitting tokens in fused_experts]
---------
Signed-off-by: ttanzhiqiang <389825161@qq.com>
### What this PR does / why we need it?
Based on the design of dual-batch overlap proposed by Deepseek team and
also the implementation of fused moe in VLLM project, we implement the
multi-stream(also known as dual-batch) overlap for deepseek+mla on
Ascend NPU. We split the input batch of model into two microbatches and
then overlap the comp/comm ops in attention and moe layers using two
streams to improve the performance. Our approach can be easily extended
when adding dispatch/combine communications for moe layer.
Compared with the previously proposed
[draft](https://github.com/vllm-project/vllm-ascend/pull/842), we use
one stream for computation ops and the other for communication ops,
separately. In out opinions, it is beneficial for arranging the order of
executing different ops and thus avoiding the contention of
computation/communication resources.
ref: [overlap for
llama](https://github.com/vllm-project/vllm/pull/15787/files)
ref: [dbo in
sglang](https://github.com/sgl-project/sglang/pull/4068/files#diff-b4937569fc71f6ad215181b633b2f89c7183a2b4ac39e41fc22635599a9be7de)
### Does this PR introduce _any_ user-facing change?
Adding an env variable "VLLM_ENABLE_DBO". Users can enable dbo by
setting "VLLM_ASCEND_ENABLE_DBO=1"
See /examples/offline_dualbatch_overlap_npu.py for more info.
### How was this patch tested?
This patch can be tested with vllm-0.9.0 using its online service with
benchmark tests. We have decoupled the func of dbo from vllm and it
should be able to run without any modification to the code of vllm(some
modifications is better to implement in vllm though).
Any advice/discussion is welcome.
### Performance Benchmark
We have ran the benchmark_serving script of vllm to test the performance
after using dual-batch overlap.
`python -m vllm.entrypoints.openai.api_server \
--model=DeepSeek-R1-W8A8 \
--trust-remote-code \
--distributed-executor-backend=mp \
-tp=16 \
--port 8006 \
--max-num-seqs 390 \
--max-model-len 32768 \
--max-num-batched-tokens 65536 \
--block-size 128 \
--compilation_config 0 \
--gpu-memory-utilization 0.90 \
--disable-log-requests \
--additional-config
'{"expert_tensor_parallel_size":1,"enable_inter_dp_scheduling":true,"init_torchair_graph_batch_sizes":true,"trace_recompiles":true,"ascend_scheduler_config":{},"enable_graph_mode":false}'`
and run benchmark with the parameters of :
`--dataset-name random --random-input-len 4096 --random-output-len 1
--num-prompts 200 --max-concurrency 8 --request-rate 5
--metric-percentiles 90`
1. test with the version using allgather+allreduce in Ascend 910B (tp16
ep16 + deepseek r1 w8a8)
2. test with the version using alltoall:
prefill qps: 0.90 -> 1.01
Mean TTFT：8226->7432ms
The overlap approach when using alltoall communication can be further
optimized by overlapping micro-batch1's moe comp with micro-batch2's
dispatch a2a comm
---------
Signed-off-by: zhuohuan <zxdu1997@gmail.com>
### What this PR does / why we need it?
- Adds support for passing prompt_embeds to LLM.generate as
```bash
llm.generate({"prompt_embeds": input_embeds}, sampling_params)
```
or
```bash
llm.generate(
[{"prompt_embeds": input_embeds} for input_embeds in inputs_embeds], sampling_params
)
```
- Add `prompt_embeds` to examples
### How was this patch tested?
CI passed with new added/existing test.
and I have test with the example script in this pr, and the output seems
looks good:
```bash
[Single Inference Output]
------------------------------
The capital of France is Paris. Paris is the largest city in France and is
------------------------------
Adding requests: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 3966.87it/s]
Processed prompts: 100%|█████████████████████████████████████████████████████████████████████████| 3/3 [00:00<00:00, 3.99it/s, est. speed input: 177.08 toks/s, output: 63.91 toks/s]
[Batch Inference Outputs]
------------------------------
Q1: Please tell me about the capital of France.
A1: The capital of France is Paris. It is located in the northern part of the
Q2: When is the day longest during the year?
A2: The day is longest during the year at the summer solstice. This typically occurs
Q3: Where is bigger, the moon or the sun?
A3: The sun is significantly bigger than the moon.
The sun has a diameter of
------------------------------
```
---------
Signed-off-by: wangli <wangli858794774@gmail.com>
More and more config options are added to additional_config. This PR
provide a new AscendConfig to manage these config options by an easier
way to make code cleaner and readable.
This PR also added the `additional_config` doc for users.
Added the test_ascend_config.py to make sure the new AscendConfig works
as expect.
TODO: Add e2e test with torchair and deepseek once the CI resource is
available.
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
### What this PR does / why we need it?
- According to https://github.com/vllm-project/vllm-ascend/issues/807,
we pull request for customer ascendc kernel of multi-step.
- also a bug we found in multi_step_runner.py is fixed when we use
multi-step on V0 Engine.
### Does this PR introduce _any_ user-facing change?
no user-facing change
### How was this patch tested?
we add Unit Test file and offline inference file to test the custom
ascendc kernel. See test/ops/test_multi_step.py and
examples/offline_multi_step.py
---------
Signed-off-by: wan_danfeng <wonderful199082@126.com>
1. Fix format check error to make format.sh work
2. Add codespell check CI
3. Add the missing required package for vllm-ascend.
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
### What this PR does / why we need it?
- This PR proposes a P2P version of Disaggregated Prefill based on
llm_datadist which manages data transfer.
- This solution reconstructs previous offline single-node Disaggregated
Prefill solution, and supports multi-node and online serveing now.
- Currently this solution supports 1P1D situation of Deepseek hybrid
parallelism (P: TP+EP, D: DP+EP). Note that xPyD situation is considered
in the solution design, and will be supported soon within v1 engine.
---------
Signed-off-by: hw_whx <wanghexiang7@huawei.com>
Signed-off-by: ganyi <pleaplusone.gy@gmail.com>
Co-authored-by: hw_whx <wanghexiang7@huawei.com>
Co-authored-by: ganyi <pleaplusone.gy@gmail.com>
### What this PR does / why we need it?
1. support deepseek with w8a8 quant;
2. support deepseek with mix-parallel(multi-DP, EP+TP);
3. support deepseek with graphmode.
---------
Signed-off-by: wen-jie666 <wenjie39@huawei.com>
Signed-off-by: Yizhou Liu <liuyizhou5@h-partners.com>
Signed-off-by: libaokui <libaokui@huawei.com>
Signed-off-by: linfeng-yuan <1102311262@qq.com>
Co-authored-by: wen-jie666 <wenjie39@huawei.com>
### What this PR does / why we need it?
Adapt Disaggregated Prefill feature onto Ascend device
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
The test usage has been provided alongwith the PR, in
examples/offline_disaggregated_prefill_npu.py
To run it, do this
```
export PROMPT_DEVICE_ID=0,1
export DECODE_DEVICE_ID=2,3
python examples/offline_disaggregated_prefill_npu.py
```
---------
Signed-off-by: ZihuiQian <qianzihui@huawei.com>
Co-authored-by: ZihuiQian <qianzihui@huawei.com>
- Set default model to Qwen2.5-0.5B-Instruct in example
- Remove Ultravox 0.3 because it is not tested currently
Signed-off-by: MengqingCao <cmq0113@163.com>
### What this PR does / why we need it?
Fix device of tensors created in `AscendAttentionBackendImpl`.
While specifying device to cards except card-0, there'll cause an
**device conflict** because the tensors (such as `attn_mask`) will be
put on card-0 by default.
This pr creates these tensors on the correct card corresponding to the
input.
### Does this PR introduce _any_ user-facing change?
User could specify device with local rank by this pr, and a modify on
vLLM is also needed, will related to this pr when created.
### How was this patch tested?
This is tested by the following code locally. Will add a test case when
the modify in vLLM is also completed.
```python
from vllm import LLM, SamplingParams
prompts = [
"Hello, my name is",
"The president of the United States is",
"The capital of France is",
"The future of AI is",
]
# Create a sampling params object.
sampling_params = SamplingParams(max_tokens=100, temperature=0.0)
# Create an LLM.
llm = LLM(model="~/.cache/modelscope/hub/Qwen/Qwen2___5-7B-Instruct", device="npu:1")
# 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}")
```
Signed-off-by: MengqingCao <cmq0113@163.com>
### What this PR does / why we need it?
vLLM Ascend plugin (vllm-ascend) is a backend plugin for running vLLM on
the Ascend NPU.
This plugin is the recommended approach for supporting the Ascend
backend within the vLLM community. It adheres to the principles outlined
in the [RFC]: Hardware pluggable, providing a hardware-pluggable
interface that decouples the integration of the Ascend NPU with vLLM.
This patch also include changes to make CI work and use cache speed up
e2e test, including:
1. Change push (post merge ci) and pull_request (pr ci) trigger branch
to main
2. Make mypy work by ignore base_communicator and clear unused deps
3. Several improvements for vllm_ascend_test:
- use cache (pip, ms, hf) speed up e2e test (25mins --> 5mins)
- switch `git clone` command to `action/checkout` to speedup checkout
and
- Enable sv for pytest for better info dump
- Remove network host to resole `docker: conflicting ontions: cannot
attach both user-defined and non-user-definednetwork-modes`, which is a
problem on docker 1.45 but not on 1.39.
4. Adapt MLA decode optimizations:
cabaf4eff3
### Does this PR introduce _any_ user-facing change?
Yes, init the PR.
### How was this patch tested?
- This is the first PR to make ascend NPU work on vLLM. All code is
tested on ascend with vLLM V0 Engine.
- CI passed
---------
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
Signed-off-by: Yikun Jiang <yikunkero@gmail.com>
Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>
Co-authored-by: MengqingCao <cmq0113@163.com>
Co-authored-by: wangshuai09 <391746016@qq.com>
Co-authored-by: Shanshan Shen <467638484@qq.com>
Co-authored-by: wangli <wangli858794774@gmail.com>
