### What this PR does / why we need it?
1. Move `torchair_attention` to `torchair` dir.
2. Make `AscendAttentionTorchairBackend` extend `AscendAttentionBackend`
to reduce duplicate methods.
3. Make `AscendTorchairMetadata` extend `AscendMetadata` to reduce
duplicate properties.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/0933f9d518ecb18e8427bf42731d2a3d00276883
---------
Signed-off-by: shen-shanshan <467638484@qq.com>
### What this PR does / why we need it?
1. Remove the return statement, it will always skip following logic.
2. Update `deepseek` to `Qwen2.5-Instruct` for OOM in github e2e test
env.
3. Fix the comparison logic
### Does this PR introduce _any_ user-facing change?
NO.
### How was this patch tested?
Local Test.
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/0933f9d518ecb18e8427bf42731d2a3d00276883
Signed-off-by: xleoken <xleoken@163.com>
### What this PR does / why we need it?
Move torchair related model arch into torchair moduel to make the code
clear. Next step we'll remove all torchair related code outside of
torchair moduel.
### Does this PR introduce _any_ user-facing change?
No.
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/08d5f7113a024818b2867782c2539794b7aa162b
Signed-off-by: linfeng-yuan <1102311262@qq.com>
### What this PR does / why we need it?
This PR adopt Mooncake TransferEngine for kv cache register and
pull_blocks style disaggregate prefill implementation.
### Does this PR introduce any user-facing change?
No
### Dependencies
1. Cann Dependencies
Using Mooncake TransferEngine with Ascend Transport requires CANN
version 8.2.RC1 or higher.(see detail
Mooncake[#502](https://github.com/kvcache-ai/Mooncake/pull/502))
2. vllm-ascend
This PR depends on changes introduced by #950 (modifications to
`model_runner_v1`) and #1361 (updates to `schedule`), both of which have
been merged into the `v0.9.1-dev` branch and are expected to land in
`main` shortly.
### How was this patch tested?
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/1c859a1387286cf650c3bc24fdeac706b97999e8
---------
Signed-off-by: leichao.lc <leichao139636@163.com>
Co-authored-by: jianzs <zheng.shoujian@outlook.com>
Co-authored-by: zzy-ContiLearn <1831242919@qq.com>
Co-authored-by: fems14 <1804143737@qq.com>
Co-authored-by: Dreamerleader <2270923832@qq.com>
Co-authored-by: chris668899 <15105191595@126.com>
Co-authored-by: Pz1116 <zpbzpb123123@gmail.com>
### What this PR does / why we need it?
Use function CustomOp.register_oot to achieve the customop registery
```
from vllm.model_executor.custom_op import CustomOp
CustomOp.register_oot(_decorated_op_cls=AscendRMSNorm, name="RMSNorm")
```
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
CI passed with new added/existing test.
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/afa5b7ca0b417abadfa85e32f28969b72e58a885
---------
Signed-off-by: Icey <1790571317@qq.com>
### What this PR does / why we need it?
this pr refactor select_experts of moe module
i merge implementations of quantitative and non-quantitative method in a
new class
use such as vllm like ExpertsSelector.select_experts
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
test in qwen3-moe and all ut.
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/e18859298d109870b22cb5b8672d1078818e268d
Signed-off-by: yangcheng <yangcheng104@huawei.com>
Co-authored-by: yangcheng (AJ) <y00806874@china.huawei.com>
### What this PR does / why we need it?
This PR refactors the MoE (Mixture of Experts) communication logic by
introducing a strategy pattern. It defines an abstract base class,
`MoECommMethod`, which encapsulates different communication strategies
for MoE layers. By decoupling the MoE implementation from any single
communication method, this change makes it simpler to add, replace, or
optimize communication strategies in the future.
Plan / Roadmap
1. Introduce `MoECommMethod`, implement `AllGatherImpl`, and adapt ACL
Graph handling to cover all scenarios (this PR).
2. Implement `MC2CommImpl` and `AllToAllCommImpl` to optimize
performance in specific scenarios.
3. Enable W8A8 / Int8 models to use `unified_fused_experts`.
Other notes
* Data-parallel (DP) communication currently does not work with vLLM's
dispatch/combine mechanisms; an alternative approach is required to
resolve this incompatibility.
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/f7ad6a1eb3deb9ca70a6bce3705dbd16cf9d8b28
---------
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
### What this PR does / why we need it?
1.Shared Expert Sharding Strategy Update: Switched from TP-aligned to
pure DP for shared experts, enabling more efficient execution.
2.O_Proj AllReduce → ReduceScatter: Reduced communication overhead by
using ReduceScatter, made possible by pure DP sharding.
3.AllGather Postponed: Delayed to after QKV down projection to reduce
synchronization impact during prefill.
### How was this patch tested?
Adding ut case in `tests/ut/attention/test_mla_v1.py`
#### How to run
use parameter `--additional_config='{"enable_shared_expert_dp": true}'`
##### a.How to run eager mode
eg:
python -m vllm.entrypoints.openai.api_server --model=/model_path
--trust-remote-code -tp 8 -dp 2 --enable_expert_parallel --port 8002
--max-model-len 5120 --max-num-batched-tokens 16384 --enforce-eager
--disable-log-requests
--additional_config='{"ascend_scheduler_config":{"enabled":true},"enable_shared_expert_dp":
true,"chunked_prefill_for_mla":true}'
##### b.How to run graph mode
eg:
python -m vllm.entrypoints.openai.api_server --model=/model_path
--trust-remote-code -tp 8 -dp 2 --enable_expert_parallel --port 8002
--max-model-len 5120 --max-num-batched-tokens 16384
--disable-log-requests
--additional_config='{"ascend_scheduler_config":{"enabled":true},"enable_shared_expert_dp":
true,"chunked_prefill_for_mla":true,"torchair_graph_config":{"enabled":true}}'
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/9edd1db02bc6dce6da503503a373657f3466a78b
---------
Signed-off-by: Wang Kunpeng <1289706727@qq.com>
Signed-off-by: SlightwindSec <slightwindsec@gmail.com>
Co-authored-by: SlightwindSec <slightwindsec@gmail.com>
### What this PR does / why we need it?
Thanks to the PR https://github.com/vllm-project/vllm-ascend/pull/426
make vllm-ascend support the aclgraph inference to reduce the host
overhead. However, the capability of aclgraph strongly relies on the
functionality provided by `torch.compile`, which is the key feature
supported in torch 2.x . Therefore, capture custom op into aclgraph is
only possible when it can be recognize and captured by `torch.compile`.
In this PR, we register the meta implementation of current custom ops to
enable the fx graph capture. And by doing that, insert those custom ops
into aclgraph become a natural thing to the ascend runtime.
### Does this PR introduce _any_ user-facing change?
No user face change.
### How was this patch tested?
Tested in unittest, we will integrate the `rotary_embedding` op into a
small custom model and use `torch.compile` and aclgraph to capture and
replay it to verify its functionality.
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/1b9902806915040ac9b3029f2ab7522ec505afc3
---------
Signed-off-by: ganyi <pleaplusone.gy@gmail.com>
### What this PR does / why we need it?
Fix AscendFusedMoE init error. Use `super().__init__()` instead of
`super(FusedMoE, self).__init__()` to ensure the member variables in
base class could be called by the children class
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
CI passed with new existing test.
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/766bc8162cb37ad32605eee051d4f049ec325926
---------
Signed-off-by: MengqingCao <cmq0113@163.com>
### What this PR does / why we need it?
Remove transformer pins for v0.9.1-dev, because we already release the
v0.9.1rc2 with right transformer version
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
doctest CI passed
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/7e6544c7978364fcb8178f4ab8b1325e45880aa9
Signed-off-by: Yikun Jiang <yikunkero@gmail.com>
### What this PR does / why we need it?
Qwen3 MoE supports SP. In scenarios like AlltoAll, AlltoAllv, and MC2,
replacing AllReduce with Reduce-Scatter and AllGather achieves
computational benefits in norm operations while saving one AllGather
communication. This feature is enabled during the P-phase and delivers
notable gains in long-sequence scenarios (e.g., 16k–25k), with
performance improvements reaching 5%–10%.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
```
compilation_config={
"pass_config":{
"enable_sequence_parallelism": True
}
},
enable_expert_parallel=True,
```
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/9edd1db02bc6dce6da503503a373657f3466a78b
---------
Signed-off-by: libaokui <libaokui@huawei.com>
Co-authored-by: libaokui <libaokui@huawei.com>
### What this PR does / why we need it?
Support MTP with:
- [x] V0 Scheduler
- [x] TorchAir
- [x] Single DP
- [x] Multi DP
- [x] Disaggregate PD
Known issues:
- [ ] Not support V1 Scheduler (chunked prefill), will be supported in a
few weeks
- [ ] vllm v0.10.0 does not support metrics with `DP > 1` right now,
need to comment out the line 171-175 in file
`vllm/vllm/v1/metrics/loggers.py`
```
if (len(self.engine_indexes) > 1
and vllm_config.speculative_config is not None):
raise NotImplementedError("Prometheus metrics with Spec Decoding "
"with >1 EngineCore per AsyncLLM is not "
"supported yet.")
```
To start an online server with torchair enabled, here is an example:
```
python -m vllm.entrypoints.openai.api_server \
--model="/weights/DeepSeek-R1_w8a8/" \
--trust-remote-code \
--max-model-len 40000 \
--tensor-parallel-size 4 \
--data_parallel_size 4 \
--max-num-seqs 16 \
--no-enable-prefix-caching \
--enable_expert_parallel \
--served-model-name deepseekr1 \
--speculative-config '{"num_speculative_tokens": 1, "method":"deepseek_mtp"}' \
--quantization ascend \
--host 0.0.0.0 \
--port 1234 \
--additional-config '{"ascend_scheduler_config":{"enabled":true,"enable_chunked_prefill":false},"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]},"enable_weight_nz_layout":true}' \
--gpu_memory_utilization 0.9
```
offline example with torchair enabled
```
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=16, temperature=0)
# Create an LLM.
llm = LLM(
model="/home/data/DeepSeek-R1_w8a8/",
tensor_parallel_size=16,
max_num_seqs=16,
gpu_memory_utilization=0.9,
distributed_executor_backend="mp",
enable_expert_parallel=True,
speculative_config={
"method": "deepseek_mtp",
"num_speculative_tokens": 1,
},
trust_remote_code=True,
enforce_eager=False,
max_model_len=2000,
additional_config = {
'torchair_graph_config': {
'enabled': True,
"graph_batch_sizes": [16],
'enable_multistream_shared_expert': False,
},
"ascend_scheduler_config": {
"enabled": True
},
# 'expert_tensor_parallel_size': 16,
}
)
# Generate texts from the prompts.
# llm.start_profile()
outputs = llm.generate(prompts, sampling_params)
# llm.stop_profile()
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/302962e806e9820643ae25987e8e38ed035e05d3
---------
Signed-off-by: xuyexiong <xuyexiong@huawei.com>
### What this PR does / why we need it?
Supports Deepseek-R1 w4a8 quantization.
Since R1 w4a8 uses mixed quantization, only the MOE layer uses
w4a8_dynamic quantization, so we added the w4a8_dynamic.py file, which
includes the AscendW4A8DynamicFusedMoEMethod class.
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
Adding ut case in `tests/ut/quantization/test_w4a8_dynamic.py` and
`tests/ut/quantization/test_quantizer.py`
Adding e2e case in
`tests/e2e/multicard/test_offline_inference_distributed.py::test_models_distributed_DeepSeek_W4A8DYNAMIC`
to test deepseek w4a8_dynamic quantized model
#### 1.How to get weights using Modelslim
##### Installation steps
Use the branch master, the commit id is:
298e175d69b3b855111a1e09bbe2fcd12fdb4e24
git clone https://gitee.com/ascend/msit.git
cd msit/msmodelslim
bash install.sh
##### The required transformers environment
transformers>=4.48.2
##### Generate w4a8 weights
cd /example/DeepSeek
Command reference: msmodelslim/example/DeepSeek/README.md Execute the
[pre-check](https://gitee.com/ascend/msit/blob/master/msmodelslim/example/DeepSeek/README.md#%E8%BF%90%E8%A1%8C%E5%89%8D%E5%BF%85%E6%A3%80)
and [DeepSeek-R1 w4a8 mix
quantization](https://gitee.com/ascend/msit/blob/master/msmodelslim/example/DeepSeek/README.md#deepseek-r1-w4a8-%E6%B7%B7%E5%90%88%E9%87%8F%E5%8C%96%E5%89%8D%E4%B8%89%E5%B1%82-mlpw8a8-dynamic-%E9%87%8F%E5%8C%96mla%E5%85%B1%E4%BA%AB%E4%B8%93%E5%AE%B6w8a8%E9%87%8F%E5%8C%96%E8%B7%AF%E7%94%B1%E4%B8%93%E5%AE%B6w4a8-dynamic%E9%87%8F%E5%8C%96)
chapter
Reference command:python3 quant_deepseek_w4a8.py --model_path {Original
weight path} --save_path {Generate weight path} --mindie_format
##### Adapt to vllm-ascend
Since mindie_format generates mindie format, some adaptation
modifications are needed for vllm-ascend to use it:
`quant_model_description_w8a8_dynamic.json` rename to
`quant_model_description.json`, and add `"group_size": 256`
Modification in `config.json`:`"model_type":deepseekv2` is changed to
`"model_type":deepseek_v3`; `quantization_config` is removed;
tips:The group_size and weights match. If the w4a8 weights are not
generated using msmodelslim, you can check the group_size in
quantization_config in config.json.
#### 2.How to run w4a8
##### a.How to run eager mode
export VLLM_USE_V1=1 # v1
python -m vllm.entrypoints.openai.api_server --model=$1
--trust-remote-code -tp $2 -dp $3 --enable_expert_parallel
--quantization ascend --port $4 --max-model-len $5 --max-num-seqs $6
--enforce-eager
eg: python -m vllm.entrypoints.openai.api_server
--model=/weightpath/w4a8_4_layer --trust-remote-code -tp 4 -dp 4
--enable_expert_parallel --quantization ascend --port 8002
--max-model-len 5120 --max-num-seqs 128 --enforce-eager
##### b.How to run graph mode
export VLLM_USE_V1=1 # v1
export HCCL_BUFFSIZE=1024
python -m vllm.entrypoints.openai.api_server --model=$1
--trust-remote-code -tp $2 -dp $3 --enable_expert_parallel
--quantization ascend --port $4 --max-model-len $5
--additional_config='{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":true}}'
eg: python -m vllm.entrypoints.openai.api_server
--model=/weight/dsr1_w4a8_vllm --trust-remote-code -tp 4 -dp 4
--enable_expert_parallel --quantization ascend --port 8002
--max-model-len 5120
--additional_config='{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":true}}'
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/c494f96fbcf5e9f19f59e3dea6c2780aeb6c567f
---------
Signed-off-by: Wang Kunpeng <1289706727@qq.com>
### What this PR does / why we need it?
Fixes unable to load `qwen3_moe` quantized weights issue due to #1994
### Does this PR introduce _any_ user-facing change?
None
### How was this patch tested?
Add a `qwen3_moe` W8A8 quantized model in
`tests/e2e/multicard/test_qwen3_moe.py`
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/c494f96fbcf5e9f19f59e3dea6c2780aeb6c567f
---------
Signed-off-by: zhoux77899 <zhouxiang100@huawei.com>
This PR significantly optimizes performance for quantized Mixture of
Experts (MoE) layers by changing the order of quantization and
communication operations.
In the previous implementation, the `all2all` operation was performed on
unquantized `hidden_states` (in FP16/BF16) *before* quantization,
resulting in substantial communication overhead. By performing
quantization on each EP rank **first** and then sending the much smaller
quantized data, we reduce the communication volume by nearly 50%.
Additionally, this PR includes a minor optimization to cast `int` inputs
to `float` for the `argsort` operation, forcing it to run on a faster
NPU core instead of the AICPU.
These changes lead to a clear and significant performance gain in MoE
quantization scenarios.
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/7175817637bde6c668b75cce91c022e3a33b3684
---------
Signed-off-by: SlightwindSec <slightwindsec@gmail.com>
(cherry picked from commit 816375e0c1071d0696dfab1a1ce35674f9f37aa0)
### What this PR does / why we need it?
Suppose that you want to start a prefiller instance with npus `2,3`
only. So you start the instance with `ASCEND_RT_VISIBLE_DEVICES=2,3`.
The current programming will start two workers, whose ranks are `0` and
`1` respectedly. And they will pick the first and second ip addresses of
npus in the ranktable instead of the thirdth and forth ones. But
actually they are using card `2,3` and therefore they can not link with
remote instances when they attempt to transfer the KVCache.
Hence, at most 1 prefiller instance and at most 1 decoder instance can
work on a single machine since they always pick the first npu ip address
in the ranktable currently.
This pull request is proposed to fix the problem. This fix pick ips of
only those devices that are in `ASCEND_RT_VISIBLE_DEVICES` from the
ranktable.
### Does this PR introduce _any_ user-facing change?
If the user use ranktable generated by `gen_ranktable.sh`, they should
not face any change.
### How was this patch tested?
Qwen-0.6B 1P 1D, dp=2, `ASCEND_RT_VISIBLE_DEVICES=2,3` for prefiller and
`ASCEND_RT_VISIBLE_DEVICES=4,5` for decoder.
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad57f23f6a528ab01066998b41796a44340fd43d
Signed-off-by: CaveNightingale <cavenightingale@foxmail.com>
### What this PR does / why we need it?
Fix Qwen MoE accuracy bug in DP scenario.
Now the implentment of `FusedMoE` in vLLM use `All2AllManager` to
manager different all2all algorithm branch. And the default branch use
`Multicast` in `dispatch` phase and `all_reduce` in `combine` phase,
which are not implented in vLLM-Ascend. This leading to invoking into a
default implentment in `base_communicator`, with empty `dispatch` and
`combine` operations, thus causing the accuracy issue on it.
This pr is a temporary workaround, refacting all2all in vLLM-Ascend
could be a better way.
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad57f23f6a528ab01066998b41796a44340fd43d
---------
Signed-off-by: MengqingCao <cmq0113@163.com>
### What this PR does / why we need it?
This PR enabled pytest and yaml style accuracy test, users now can
enable accuracy test by running:
```bash
cd ~/vllm-ascend
pytest -sv ./tests/e2e/singlecard/models/test_lm_eval_correctness.py \
--config ./tests/e2e/singlecard/models/configs/Qwen3-8B-Base.yaml \
--report_output ./benchmarks/accuracy/Qwen3-8B-Base.md
pytest -sv ./tests/e2e/singlecard/models/test_lm_eval_correctness.py \
--config-list-file ./tests/e2e/singlecard/models/configs/accuracy.txt
```
Closes: https://github.com/vllm-project/vllm-ascend/issues/1970
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/2836dd73f13015ee386c544760ca0d16888203f3
---------
Signed-off-by: Icey <1790571317@qq.com>
bugfix cherry-pick from v0.9.1-dev
https://github.com/vllm-project/vllm-ascend/pull/2007
### What this PR does / why we need it?
Minimum reproducing code:
```python
# test.py
from vllm import LLM, SamplingParams
prompts = [
"Hello, my name is",
"The future of AI is",
]
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
llm = LLM(model="Qwen2.5-VL-7B-Instruct", max_model_len=26240)
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}")
```
```bash
export USE_OPTIMIZED_MODEL=0
python test.py
```
exception as follow:
```
[rank0]: File "/home/xxx/vllm_ascend/models/qwen2_5_vl_without_padding.py", line 84, in forward
[rank0]: q = torch_npu.npu_rotary_mul(q, cos, sin)
[rank0]: File "/home/anaconda3/envs/xxx/lib/python3.10/site-packages/torch/_ops.py", line 1116, in __call__
[rank0]: return self._op(*args, **(kwargs or {}))
[rank0]: RuntimeError: Expected all tensors to be on the same device, but found at least two devices, npu:0 and cpu! (when checking argument for argument r1 in method wrapper__npu_rotary_mul)
```
In `AscendQwen2_5_VisionAttention_Without_Padding`,
`torch_npu.npu_rotary_mul(q, cos, sin)`, `cos`/`sin` on cpu, but `q` on
npu, so there will be an error.
`qwen2_5_vl_without_padding.py` need this bugfix, because
`AscendQwen2_5_VisionTransformer_Without_Padding.rot_pos_emb` in
wen2_5_vl_without_padding.py is from vllm and `inv_freq` will create on
cpu.
https://github.com/vllm-project/vllm/blob/40d86ee412eeeca93e0c37432db6b96829cb64e2/vllm/model_executor/models/qwen2_5_vl.py#L482
```python
inv_freq = 1.0 / (theta**(torch.arange(0, dim, 2, dtype=torch.float, device='cpu') / dim))
```
`qwen2_5_vl.py` do not need, because
`AscendQwen2_5_VisionRotaryEmbedding` in qwen2_5_vl.py rewrite
`AscendQwen2_5_VisionRotaryEmbedding` and `inv_freq` will create on
device.
```python
inv_freq = 1.0 / (theta**(torch.arange(0, dim, 2, dtype=torch.float) / dim))
```
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
CI passed with new added/existing test.
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/18cc33dd603cce058867f5730919041fa3617eb7
Signed-off-by: pjgao <gaopengju3@huawei.com>
Co-authored-by: pjgao <gaopengju3@huawei.com>