### What this PR does / why we need it?
This PR fix bugs and refactor cached mask generation logic. Now just
pre-construct and use the cached mask on cpu instead of device on npu.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
CI passed with new added/existing test.
- vLLM version: v0.10.1.1
- vLLM main:
https://github.com/vllm-project/vllm/commit/9b5f64238fbd0f98928587b3426cbf69eea96ae7
Signed-off-by: rjg-lyh <1318825571@qq.com>
### What this PR does / why we need it?
convert the format of gmm to nz
### Does this PR introduce _any_ user-facing change?
not involved
### How was this patch tested?
ut: test_fused_ops.py and e2e: test_fused_moe.py
**performance**:
(qwen3 30B, 2k->20k)
base:
Total Token throughput (tok/s): 719.93
gmm nz:
Total Token throughput (tok/s): 728.52
- vLLM version: v0.10.1.1
- vLLM main:
https://github.com/vllm-project/vllm/commit/bfc1edc9f5bde581e0eec5c830a5a4a7b710fe6a
Signed-off-by: huangxialu <huangxialu1@huawei.com>
### What this PR does / why we need it?
This pr updates compilation config in `check_and_update_config`, we use
`compilation_config.level` to update `compilation_config.cudagraph_mode`
to ensure the config is correct.
Add `compilation_config.cudagraph_num_of_warmups = 1` when V1 is
enabled, cause this is also used in torchair graph mode. and this fixes
https://github.com/vllm-project/vllm-ascend/issues/2523
fix the bug that the `aclgraphmode` always be `NONE` while running
forward in aclgraph mode
### How was this patch tested?
CI passed with new added/existing test.
- vLLM version: v0.10.1.1
- vLLM main:
https://github.com/vllm-project/vllm/commit/f58675bfb36b67cdbca4d2356a2f580e7a706ec3
---------
Signed-off-by: MengqingCao <cmq0113@163.com>
### What this PR does / why we need it?
This method replaces the previous all-gather approach for small numbers
of tokens.
The key changes include:
- A new `AscendFusedMoE` layer that handles token splitting, local
computation, and final aggregation via all-gather.
- Logic in the model runner to dynamically select between the new MC2
method and the existing all-gather method based on the number of input
tokens.
- Sharding the MoE communication mask across tensor-parallel ranks.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
Test case fixed.
- vLLM version: v0.10.1.1
- vLLM main:
https://github.com/vllm-project/vllm/commit/b00e69f8ca55f4a82847d39466f57ceb748324c1
---------
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
[Bugfix]Support Qwen3-MOE on aclgraph mode in sizes capture and add new
ut
What this PR does / why we need it?
This PR solves the problem of sizes capture and stream error caused by
using ACLgraph on the Qwen3-30B MOE model.
Add new ut.
Does this PR introduce any user-facing change?
no
How was this patch tested?
ut
- vLLM version: v0.10.1.1
- vLLM main:
https://github.com/vllm-project/vllm/commit/6fad29b11b3680c44782cd6e5fe555779d620d6c
Signed-off-by: lilinsiman <lilinsiman@gmail.com>
### What this PR does / why we need it?
Refactor all2all-related fused_experts (both quantized/unquantized) into
TokenDispatcherWithAll2AllV, including dispatch & combine calculation.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
E2E & UT
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/65197a5fb37ef4d8b93e0b99ecc8b902fe948e97
Signed-off-by: Pr0Wh1teGivee <calvin_zhu0210@outlook.com>
### What this PR does / why we need it?
Add configuration check logic for ascend scheduler: if chunked_prefill
is disabled, max_num_batched_tokens couldn't be less than max_model_len,
following vLLM;
### Does this PR introduce _any_ user-facing change?
users cannot set max_num_batched_tokens smaller than max_model_len with
ascend scheduler
### How was this patch tested?
CI and vllm serving passed
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/f77a0802b758a32c5b9f7bc04e9498d77e8d99e0
Signed-off-by: linfeng-yuan <1102311262@qq.com>
### What this PR does / why we need it?
Skip failed ut to recover CI quickly
related ut:
- `test_embed_models_correctness`: revert me when pooler is adapted with
the latest vllm main
- `test_check_and_update_config_enforce_eager_mode`: revert me when the
occasional failed is fixed
- vLLM version: v0.10.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/8896eb72ebe90be6f1b83f32b3d3d0c379db4f82
Signed-off-by: MengqingCao <cmq0113@163.com>
### What this PR does / why we need it?
For dense models, by not applying tensor parallelism (TP) to the
attention module and applying TP to the MLP module, the allreduce
operations in the attention module can be eliminated, thereby reducing
computational overhead. However, this approach increases memory usage,
so the environment variable VLLM_ASCEND_ENABLE_MLP_OPTIMZE is used to
control this optimization.
- vLLM main:
https://github.com/vllm-project/vllm/commit/b17109beeafbf9577c319ab61530810943a7fc4b
Signed-off-by: wangxiaoxin-sherie <wangxiaoxin7@huawei.com>
Co-authored-by: wangxiaoxin-sherie <wangxiaoxin7@huawei.com>
### What this PR does / why we need it?
The deepseek w4a8 weights we supported before were in mindie-format
format. It uses int8 to represent int4, so the weight size is similar to
w8a8, and we need to do a few extra steps to make vllm-ascend load it
normally.
Now we can directly use the new weight format, which uses two int4 packs
to save the weight, the weight size is reduced, and there is no need to
do many extra operations to directly use it on vllm-ascend, but we are
also compatible with the weights of the previous mindie format.
The weight changes in the new version:
1. The weight is packed (2 int4 pack to int8)
2. The bias required in the apply method is directly generated by
modelslim
### 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`
#### 1.How to get weights using Modelslim
##### Installation steps
we can use the branch br_release_MindStudio_8.1.RC2_TR5_20260624
git clone -b br_release_MindStudio_8.1.RC2_TR5_20260624
https://gitee.com/ascend/msit.git
cd msit/msmodelslim
bash install.sh
##### Generate w4a8 weights
cd /example/DeepSeek
Command reference: msmodelslim/example/DeepSeek/README.md Execute the
[pre-check](https://gitee.com/ascend/msit/blob/br_release_MindStudio_8.1.RC2_TR5_20260624/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/br_release_MindStudio_8.1.RC2_TR5_20260624/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}
##### Adapt to vllm-ascend
Modification in `config.json`:`"model_type":deepseekv2` is changed to
`"model_type":deepseek_v3`;
#### 2.How to run w4a8
##### a.How to run eager mode
export VLLM_ASCEND_MLA_PA=1
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 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/103f1ec8d348a5f336f11d972d6285c4fb4736d4
---------
Signed-off-by: Wang Kunpeng <1289706727@qq.com>
### 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?
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?
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?
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?
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>
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>