### What this PR does / why we need it?
vllm-ascend support [msMonitor
](https://gitcode.com/Ascend/mstt/tree/master/msmonitor)tool to collect
performance of vllm-ascend
### Does this PR introduce _any_ user-facing change?
1.add env MSMONITOR_USE_DAEMON；
2.user cann enable msMonitor tool by setting MSMONITOR_USE_DAEMON=1
before run vllm-ascend model；
3.MSMONITOR_USE_DAEMON and VLLM_TORCH_PROFILER_DIR cannot both set
### How was this patch tested?
1.run vllm-ascend model while not set MSMONITOR_USE_DAEMON=1 or set
MSMONITOR_USE_DAEMON=0, model will run successfully;
2.run vllm-ascend model while set MSMONITOR_USE_DAEMON=1, run msMonitor
tool to collect profile data;
3.run vllm-ascend model while set MSMONITOR_USE_DAEMON=1 and
VLLM_TORCH_PROFILER_DIR, will raise error
- vLLM version: v0.10.2
- vLLM main:
f225ea7dd9
Signed-off-by: mei-feiyao <1332490378@qq.com>
### What this PR does / why we need it?
Fixes#3096
1. Fix kv cache initialization error of attention layer. There are some
models with layer name like `attn.attn`, instead of `self_attn`, but the
initialization of kv cache tensors only check for `self_attn` and
`attn.attn`, which leding to the error `AssertionError: Some layers are
not correctly initialized`
2. Set the default value of input arg `sampling_metadata` in
`compute_logits` for the modeling files in vllm-ascend. Thus fixing the
error `Qwen3NextForCausalLM.compute_logits() missing 1 required
positional argument: 'sampling_metadata'`
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
test locally with internlm
- vLLM version: v0.10.2
- vLLM main:
5aeb925452
---------
Signed-off-by: MengqingCao <cmq0113@163.com>
What this PR does / why we need it?
there are two sets of sp implementations for moe and dense models. One
is called sequence_parallelism, and the other is flashcomm_v1.
We did the following things：
Merge two sets of code with the same implementation into one.
Remove the implementation of sequence_parallelism, as this solution
cannot support aclgraph.
Does this PR introduce any user-facing change?
No
How was this patch tested?
e2e&ut
- vLLM version: v0.10.2
- vLLM main:
f225ea7dd9
---------
Signed-off-by: weijinqian_v1 <weijinqian@huawei.com>
Co-authored-by: weijinqian_v1 <weijinqian@huawei.com>
### What this PR does / why we need it?
When we copy the sampled valid token ids from device to host, avoid
using tolist which would trigger a CUDA wise stream sync if the source
is on device. We change it to use non-blocking copy followed by an
explicit CUDA event sync.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
Bring up vLLM server
```bash
VLLM_USE_V1=1 vllm serve Qwen/Qwen2.5-14B-Instruct --disable-l
og-requests -tp 8 --max-num-seqs 64 --no-enable-prefix-caching --max_num_batched_tokens=8000
```
## Before：
## After
As shown in the figure, the TTFT decreased
- vLLM version: v0.10.2
- vLLM main:
9607d5eb44
---------
Signed-off-by: jesse <szxfml@gmail.com>
### What this PR does / why we need it?
Updates the `cudagraph_support` attribute to `aclgraph_support` to use
terminology appropriate for the Ascend platform (ACL graphs instead of
CUDA graphs).
This change also explicitly disables graph support for the MLA attention
backend.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
None needed.
- vLLM version: v0.10.2
- vLLM main:
5aeb925452
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
### What this PR does / why we need it?
Fix the impact to LoRA that
https://github.com/vllm-project/vllm/pull/25249 brought.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
pytest -sv tests/e2e/singlecard/test_ilama_lora.py
pytest -sv tests/e2e/multicard/test_ilama_lora_tp2.py
- vLLM version: v0.10.2
- vLLM main:
9607d5eb44
---------
Signed-off-by: paulyu12 <507435917@qq.com>
### What this PR does / why we need it?
This is the follow-up PR of #2128 .
Moves graph parameter management components, including `GraphParams`,
`get_graph_params`, and `set_graph_params`, from the generic `utils.py`
to the more specific `compilation/acl_graph.py`.
Additionally, extracts the `update_attn_params` logic from the
`NPUModelRunner` class into a standalone function within the `acl_graph`
module.
This refactoring improves code organization by centralizing ACL
graph-related logic into its own dedicated module, enhancing modularity
and clarity.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
None needed.
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
### What this PR does / why we need it?
A new kv_role "kv_both" is added to run mixed deployment scenarios. The
mixed deployment will involve a decode phase, where with_prefill should
be false.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.10.2
- vLLM main:
c60e6137f0
Signed-off-by: fems14 <1804143737@qq.com>
### What this PR does / why we need it?
Fix issues mentioned in
https://github.com/vllm-project/vllm-ascend/pull/2791 and some minor
refactoring.
1. Use Enum instead of string.
2. Avoid setting a new property to forward_context in
AscendFusedMoE.forward().
3. Enabling TokenDispatcherWithMoge.
4. Remove redundant code.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Qwen3-30B-A3B/Qwen3-30B-A3B-W8A8/DeepSeek-V3-W4A8-Pruing/deepseek-mtp/pangu-pro-moe-pruing:
1. Enable/Disable EP
2. Aclgraph & eager
- vLLM version: v0.10.2
- vLLM main:
9607d5eb44
Signed-off-by: Pr0Wh1teGivee <calvin_zhu0210@outlook.com>
Co-authored-by: weijinqian0 <12153182+weijinqian0@users.noreply.github.com>
Note: This depends on [vLLM
#25161](https://github.com/vllm-project/vllm/pull/25161) and the
torch\_npu release from September 30.
### What this PR does / why we need it?
This pull request adds `FULL_DECODE_ONLY` mode for GQA/MHA models (MLA
models like DeepSeek V3/R1 are not included). Key improvements include:
* **Reduced dispatch latency:** By replaying the entire model execution
graph at once, we cut overhead compared with multiple smaller replays.
* **Stabilized multi-device performance:** Captureing the whole model as
one static graph also mitigates the dispatch fluctuations across
devices.
* **Stream/resource savings:** Consolidating graph captures frees up
streams, allowing more graphs to be captured.
**Known issues:**
1. `_npu_paged_attention` currently manages its own workspace in
`torch_npu`, which can deadlock when synchronizing during graph replay —
we’re working on a fix.
There may be other corner cases. This PR is the first in a planned
series; we’ll continue to iterate and address remaining issues in
follow-ups.
This is essentially a port of #1503 and #1677, but includes two major
changes:
1. Let `graph_dispatcher` decide the graph mode instead of hard-coding
it in the backend, which decouples Full Graph and Piecewise Graph and
could make it possible to remove dynamo.
2. Adapt to the new `attn_group` logic, but leave a small hack in
`update_graph_params`; multi-attention models may or may not be fully
supported yet.
### Does this PR introduce _any_ user-facing change?
```python
compilation_config={
"cudagraph_mode": "FULL_DECODE_ONLY",
},
```
### How was this patch tested?
Tests included.
- vLLM version: v0.10.2
- vLLM main:
9607d5eb44
---------
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
### What this PR does / why we need it?
Follow up `UniformTypeKVCacheSpecs` changes introduced by
https://github.com/vllm-project/vllm/pull/25101, which support different
hidden size in uniform type kvcache specs
This also fix the CI issue about `TypeError: AttentionGroup.__init__()
missing 1 required positional argument: 'kv_cache_spec'`
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
Tests passed with exsiting e2e tests.
- vLLM version: v0.10.2
- vLLM main:
c60e6137f0
---------
Signed-off-by: MengqingCao <cmq0113@163.com>
### What this PR does / why we need it?
Add new npu_fused_infer_attention_score op to improve perfomance in
splitfuse cases and resolve long-seq mask problems .
1. The original op's performance is suboptimal in certain scenarios,
necessitating optimization through the _new op_
(npu_fused_infer_attention_score)。
2. For ultra-long sequences (128k), the original operator will allocate
a large attn_mask, which consumes excessive CPU memory. In contrast, the
_new op_ supports a fixed-size compressed mask, effectively resolving
this issue.
NOTE1: The current PR retains the original logic and uses a version
check of the CANN package to determine whether the _new op_ can be
enabled. This ensures no impact on existing users. In future versions,
this version check and the original logic will be deprecated, and the
_new op_ scheduling will be uniformly adopted.
NOTE2: This pr relies on future CANN version, which is not available
now.
NOTE3: To enable the new op in chunked prefill, the parameter
additional_config should be set like `--additional-config
'{"ascend_scheduler_config":
{"enabled":true,"enable_chunked_prefill":true}}' \` at least.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
CI passed
- vLLM version: v0.10.2
- vLLM main:
6c5f82e5aa
---------
Signed-off-by: tangtianyi <tangtianyi4@huawei.com>
Signed-off-by: Angazenn <supperccell@163.com>
Co-authored-by: Angazenn <supperccell@163.com>
### What this PR does / why we need it?
based on the https://github.com/vllm-project/vllm/pull/23770,
fix Async scheduling and PP compatibility with DP, also fixes issue with
finished requests not being processed in async scheduling and PP cases,
and possible worker race conditions.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.10.2
- vLLM main:
544fe76b95
---------
Signed-off-by: jesse <szxfml@gmail.com>
### What this PR does / why we need it?
This pr fixes a few issues on prefill disaggregation:
1. Fix prefill disaggregation kvcache addr alignment issue, llmdatadist
needs the addr of tensors to be aligned with 2M
2. Fix prefill disaggregation kvcache shape error, llmdatadist requires
k/v tensors with shape [num_blocks, ...], however the implentment before
this pr is [2, num_blocks, ...], which will break prefill disaggregation
3. Use hybrid kv cache only when running qwen3_next to fix accuracy
issue on prefill disaggregation.
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
Tested locally by @liziyu179
- vLLM version: v0.10.2
- vLLM main:
4f02b77de4
---------
Signed-off-by: MengqingCao <cmq0113@163.com>
### What this PR does / why we need it?
For sleep mode level 2, we discarded model both weights and kv_cache,
but the problems is: When we discard weights, we also discard some
tensors representing the model state which we called
`model.named_buffers()`, such as: `running_mean / running_var` in
BatchNorm、rope cos-sin cache ... when we update weights, but forgot to
update buffers as well, this will lead to some unknown issue
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.10.2
- vLLM main:
5963b98b46
---------
Signed-off-by: wangli <wangli858794774@gmail.com>
### What this PR does / why we need it?
This PR depends on the merge of #2707 and has adapted the aclgraph
functionality to support MTP.
### How was this patch tested?
- vLLM version: v0.10.2
- vLLM main:
2b85697031
---------
Signed-off-by: xuyexiong <xuyexiong@huawei.com>
### What this PR does / why we need it?
Added a new connector for Mooncake store integration to enable kvcache
reuse in scenarios with system prompts or multi-turn dialogues.
### How was this patch tested?
- vLLM version: v0.10.2
- vLLM main:
5963b98b46
---------
Signed-off-by: LCAIZJ <leichao139636@163.com>
Signed-off-by: fems14 <1804143737@qq.com>
Co-authored-by: fems14 <1804143737@qq.com>
Co-authored-by: Dreamerleader <2270923832@qq.com>
Co-authored-by: Pz1116 <zpbzpb123123@gmail.com>
Co-authored-by: lizy124 <1950471827@qq.com>
Co-authored-by: zouyida2052 <zouyida2002@gmail.com>
### Motivation
Currently dynamically experts balancing would stop-the-world.
Asynchronously expert load balancing would be better without flowing
problems:
Host-bound latency:
There are many cpu operations during EPLB such as
eplb-algorithm、creating p2p ops、and log2phy expert converting would
spend long cpu time, as ~1s.
Communication latency: The transfer time would cost much in the
situation without nvlink. As the weight of an expert maybe transfer to
multiple new positions, thus N times send/recv for one expert, with
result long latency. We had tested that batch_isend_irecv cost more
100ms for 16 experts weight transmission in A2 server of ascend.
SwiftBalancer would not stop-the-world anymore, in out test on NPU 1~2ms
cost for each layer while benefit 5ms-8ms decode latency with ep_size =
64.
The following updates have been made:
1、expert distribution recording with lower cost.
2、async cpu computing for eplb algo and other python operator.
3、new eplb algo with less expert rebalancing while almost the same
effect.
### Proposed Change
We will gradually migrate the EPLB logic to the VLLM community and
implement a generalized design. Relevant RFC:
https://github.com/vllm-project/vllm/issues/22246
The overall workflow involves:
<img width="801" height="302"
alt="474430541-23b06f58-23bc-44a3-a1be-00f268aeb15c"
src="https://github.com/user-attachments/assets/1d73a459-1b23-4b0a-812a-bf0a75debfed"
/>
1. Record experts distribution during forward. We using expert_token_num
after disptach instead of topk_ids, thus we got much smaller tensor
shape to reduce cost of hbm recording and add-operator.
2. Do all-gather for experts distribution. Using all-gather instead of
all-reduce as less traffic volume.
3. Wake up eplb worker process with experts distribution when
num_iterations comes. Run eplb algorithm in eplb worker.
4. Generate p2p send/recv ops and other operator such as log2phy would
cost long cpu time.
5. Lanch ibatch_send_recv in async_stream before forward.
6. After forward, wait for the ibatch_send_recv finish, then do uapte
expert map and expert weights.
### Co-author
Co-authored-by: raindaywhu raindaywhu@raindaywhu@ 163.con
Co-authored-by: njuyuan yuanjl19@smail.nju.edu.cn
Co-authored-by: qmkakaxi wjh1594260677@qq.com
Co-authored-by: Skywalker-EP 173723846@qq.com
- vLLM version: v0.10.2
- vLLM main:
567939953b
---------
Signed-off-by: offline0806 <z00858301@china.huawei.com>
Co-authored-by: offline0806 <z00858301@china.huawei.com>
### What this PR does / why we need it?
In memory of #677 , a long overdue milestone. Now DeepSeek V3/R1 should
be OK with ACL Graph.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
Working on it.
- vLLM version: v0.10.2
- vLLM main:
68dbde5dbb
---------
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
### What this PR does / why we need it?
1. Replace prepare/finalize operation in fused_moe.py by
moe_comm_method.prepare()/finalize()
2. Replace unified_fused_experts by moe_comm_method.fused_experts() in
fused_moe.py/w8a8_dynamic.py/w4a8_dynamic.py
3. Add calling _select_moe_comm_method in spec-decode proposers.
4. Currently, w4a8_dynamic does not support gatherep, use all2allv
instead.
5. Remove redundant code.
### Does this PR introduce _any_ user-facing change?
AllgatherEP switch is disabled in aclgraph/eager mode, just follow the
rules in modelrunner_v1._select_moe_comm_method()
### How was this patch tested?
e2e & ut
- vLLM version: v0.10.2
- vLLM main:
7f6f2c1182
Signed-off-by: Pr0Wh1teGivee <calvin_zhu0210@outlook.com>
Co-authored-by: weijinqian0 <12153182+weijinqian0@users.noreply.github.com>
### What this PR does / why we need it?
modelslim will generate self.bias for rms norm in quantization, since
RMSNorm in vllm has no this parameter, so its nesscesary
to create a AscendQuantRmsNorm.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
tested by deepseek-v3.1-w8a8
<img width="2496" height="592" alt="image"
src="https://github.com/user-attachments/assets/004c6e76-3d7a-4a1f-b59f-a14304012663"
/>
- vLLM version: main
- vLLM main:
d6249d0699
Signed-off-by: 22dimensions <waitingwind@foxmail.com>
### What this PR does / why we need it?
This PR prefetchs the weight of mlp layers in Qwen Dense Models to
optimize the performance in Decode phase mainly.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
CI passed with new added/existing test.
- vLLM version: main
- vLLM main:
a1213fae5f
Signed-off-by: rjg-lyh <1318825571@qq.com>
Co-authored-by: Shuming19 <313093131@qq.com>
This PR is based on top of
[#23569](https://github.com/vllm-project/vllm/pull/23569) and
[#24219](https://github.com/vllm-project/vllm/pull/24219).
### What this PR does / why we need it?
This PR allows the model runner to function asynchronously when using
async scheduling. This allows full overlap of the cpu operations
(including prepare_inputs) and the model forward pass. This diff is
functional and does not support speculative decoding, PP, or guided
decoding.
Expected speedup is 5-10% over the current async scheduling.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
server
```
python -m vllm.entrypoints.openai.api_server --model=Qwen3-32B\
--trust-remote-code --enforce-eager \
--distributed-executor-backend=mp \
-tp=4 \
--port 8006 \
--max-model-len 32000 \
--block-size 128 \
--gpu-memory-utilization 0.99
```
client
```
python $TEST_PY --backend vllm --trust-remote-code --model Qwen3-32B \
--dataset-name random --random-input-len 2048 --random-output-len 2048 \
--ignore-eos\
--num-prompts 48 --max-concurrency 48 --request-rate inf --temperature 0 \
--metric-percentiles 90 --base-url http://localhost:8006 --save-result \
--result-dir $PROFILER_DIR
```
benchmark test based on Qwen3-32B TPOT result:
||forward async| scheduler async |sync|
|-|-|-|-|
|avg|41.73|41.86|44.20|
|improve0|0.3%|0|0|
|improve1|5.58%|0|0|
benchmark test based on Qwen2___5-VL-7B-Instruct TPOT result:
||forward async|sync|
|-|-|-|
|avg|23.22|29.16|
|improve|20.3%|0|
- vLLM version: main
- vLLM main:
e93f4cc9e3
Signed-off-by: jiangpeng36 <jiangpeng36@huawei.com>
Signed-off-by: Ronald1995 <ronaldautomobile@163.com>
Co-authored-by: jiangpeng36 <jiangpeng36@huawei.com>
Co-authored-by: Ronald1995 <ronaldautomobile@163.com>
### What this PR does / why we need it?
Adjust the position of warm_up_atb.
### Does this PR introduce _any_ user-facing change?
not involved
### How was this patch tested?
CI passed with existing test.
- vLLM version: main
- vLLM main:
b23fb78623
Signed-off-by: huangxialu <huangxialu1@huawei.com>
### What this PR does / why we need it?
Remove compatibility maintenance for vllm v0.10.1 and v0.10.1.1
### Does this PR introduce _any_ user-facing change?
branch main of vllm-ascend will not be compatible with vllm v0.10.1 and
v0.10.1.1
### How was this patch tested?
CI passed with existing test.
- vLLM version: v0.10.1.1
- vLLM main:
6fb2788163
---------
Signed-off-by: MengqingCao <cmq0113@163.com>
### What this PR does / why we need it?
Flashcomm_v1 optim in Qwen Dense Models.
### 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:
5e537f45b4
Co-authored-by: 1024daniel <xxltju324@gmail.com>
### What this PR does / why we need it?
fix https://github.com/vllm-project/vllm-ascend/issues/2702
- A2: skip graph_size update that makes it to tp_size because
dispatch/combine op support different batch size across EP ranks
- A3: add `max_num_reqs = max(new_graph_batch_sizes)` to fix graph_size
and max_num_reqs mismatch
### Does this PR introduce _any_ user-facing change?
Nope
### How was this patch tested?
- vLLM version: v0.10.1.1
- vLLM main:
e599e2c65e
---------
Signed-off-by: realliujiaxu <realliujiaxu@163.com>
### What this PR does / why we need it?
Refector prepare_inputs in model_runner_v1.py for more easy read.
### Does this PR introduce _any_ user-facing change?
NO
### How was this patch tested?
PASS CI
- vLLM version: v0.10.1.1
- vLLM main:
e599e2c65e
---------
Signed-off-by: ChenTaoyu-SJTU <ctynb@qq.com>
### What this PR does / why we need it?
Removes the condition that skips metadata synchronization when
`enforce_eager` is enabled.
This change is necessary to correctly sync the `with_prefill` and
`enable_dbo` flags across all data parallel ranks, which is not required
in the base implementation. Forcing the sync operation prevents
potential inconsistencies, albeit with a minor performance impact.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
Add a E2E online test case?
- vLLM version: v0.10.1.1
- vLLM main:
e599e2c65e
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
### What this PR does / why we need it?
Refactors the Mixture-of-Experts (MoE) communication method selection
logic. The choice between all-gather, all-to-all, and mc2 is now
determined by expert parallel configuration, SoC version (A2/A3), and
token count for better performance.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
Added.
- vLLM version: v0.10.1.1
- vLLM main:
eafa8dcde6
---------
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
### What this PR does / why we need it?
Refactor spec decode
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
CI passed with new added/existing test.
- vLLM version: v0.10.1.1
- vLLM main:
6997a25ac6
---------
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
Signed-off-by: Icey <1790571317@qq.com>
Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>
### What this PR does / why we need it?
This PR ports #2312#2506#2531 to main branch.
Original implementation of torchair caching forces users to make
everything prepared, fix all the configuration and enable
`use_cached_npu_graph`, and it might cause some problems confusing to
understand and tackle for users. It is better to compile the graph twice
instead of reusing the old kvcaches and cached torchair graph. And the
extra duration time is acceptable. Additionally, this pr fixes a
recompilation problem of torchair graph mode caused by
`running_in_graph` variable in `AscendMLATorchairImpl`.
### Does this PR introduce _any_ user-facing change?
If users want to enabling torchair.cache_compile with high compilation
speed, it is recommended to enable both `use_cached_kv_cache_bytes` and
`use_cached_graph` in `torchair_graph_config`. Without
`use_cached_kv_cache_bytes`, we'll compile torchair computation graph
twice to avoid runtime error caused by configuration mismtaches (the
second compilation will be much faster). Additionally, we've made a
change to how the TORCHAIR_CACHE_HOME enviroment variable is utilized to
enhance safety and prevent accidental file deletion by adding a suffix
directory.
### How was this patch tested?
CI and e2e vllm serving pass.
- vLLM version: v0.10.1.1
- vLLM main:
70549c1245
---------
Signed-off-by: linfeng-yuan <1102311262@qq.com>
### What this PR does / why we need it?
Fix MTP torchair bug caused by torchair refactor and moe refactor
Depends on PRs:
fused moe fix: https://github.com/vllm-project/vllm-ascend/pull/2627
torchair multi DP fix:
https://github.com/vllm-project/vllm-ascend/pull/2626
### Does this PR introduce _any_ user-facing change?
when dp is enabled, to run mtp online server, need to disable server log
due to the current metrics does not support multi dp
`--disable-log-stats`
### How was this patch tested?
- vLLM version: v0.10.1.1
- vLLM main:
7c8271cd1e
Signed-off-by: xuyexiong <xuyexiong@huawei.com>
### What this PR does / why we need it?
* **Unify execution paths:** Consolidates the quantized and
non-quantized execution paths into a single `fused_experts` function,
removing duplicated logic and making the control flow clearer and easier
to maintain.
* **W8A8 dynamic quantization:** Adds support for W8A8 dynamic
quantization inside the unified MoE kernel. Communication routines are
updated to correctly handle dynamic quantization scales for activations.
* **Weight pre-processing:** Prae-transpose the `w13` and `w2` weight
matrices (as implemented in PR #2025) so that quantized and
non-quantized models follow the same code path for the MoE gating,
up-projection, and down-projection operations.
* **All-to-all communication:** Adds an `all-to-all` collective
communication pattern. For large token counts on modern hardware,
`all-to-all` is more efficient than the previous `all-gather` strategy.
However, `all-to-all` is not really captured and replayed due to
multiple D2H operations which will trigger synchronization, and thus
raise error when capture graphs. We only use `all-to-all` when fallback
to `compiled_graph_for_general_shape`.
* **Dynamic communication selection:** The model runner now selects the
optimal MoE communication method (`mc2`, `allgather`, or `alltoall`) at
runtime based on token count and the Ascend SoC version.
* **Limitation:** `all-gather` is not yet supported for quantized
models, which means there is still something left to do on A2.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
No further test cases needed.
- vLLM version: v0.10.1.1
- vLLM main:
d660c98c1b
---------
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
### What this PR does / why we need it?
It is confirmed that `num_input_tokens` must be assigned the value of
`maybe_padded_num_tokens` under all circumstances.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
Waiting for daily test for TorchAir.
- vLLM version: v0.10.1.1
- vLLM main:
006477e60b
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
### What this PR does / why we need it?
This PR introduces LMhead tensor model parallel to achieve decreasing of
memory consumption, and TPOT performance improvement. It support both
eager mode and graph mode.
In deepseek r1 w8a8 PD disagregated Decode instance, using pure DP, with
lmhead_tensor_parallel_size = 8, we have 1 ms TPOT optimization, saved
1.48 GB NPU memory per RANK.
performance data:
<img width="1444" height="438" alt="image"
src="https://github.com/user-attachments/assets/3c5ef0d3-a7c7-46fd-9797-4de728eb0cb0"
/>
### Does this PR introduce _any_ user-facing change?
This PR introduces one new config in `additional_config`.
| Name | Effect | Required | Type | Constraints |
| :---------------------------- |
:--------------------------------------- | :------- | :--- |
:----------------- |
| lmhead_tensor_parallel_size | Split the lm_head matrix along the
column dimension (vocab_size) into lmhead_tensor_parallel_size pieces |
No | int | default value is None, once this value is set, the feature
will be enabled, vocab_size must be divisible by this value. |
example
`--additional_config={"lmhead_tensor_parallel_size": 8}`
### How was this patch tested?
- vLLM version: v0.10.1.1
- vLLM main:
de533ab2a1
---------
Signed-off-by: zzhx1 <zzh_201018@outlook.com>
Co-authored-by: zhangzihang <zzh_201018@outlook.com>
### What this PR does / why we need it?
The determination of attention state, padding, and other forward
metadata has been moved to an earlier stage within the input preparation
process. This change enables us to utilize a single all-reduce
operation, maximizing synchronization efficiency as early as possible.
The logic for synchronizing metadata—such as the number of tokens,
prefill status, and DBO status—across data parallel (DP) ranks has now
been unified and simplified.
For performance improvements, the all-reduce operation has been switched
from the `gloo` backend to the `npu` backend, which results in an
reduction of several milliseconds per step (**approximately 10%
performance gain for TPOT!**).
Additionally, the multi-DP server hang issue has been resolved, ensuring
no more hangs occur when `num_requests < dp_size`. Alas, a relief.
Finally, the miscalculated memory usage issue has been addressed by
removing the unnecessary `DummyCommImpl`, allowing the system to use the
real communication method when determining available memory.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
Maybe we should add an test case for multi-DP online server?
@MengqingCao
- vLLM version: v0.10.1.1
- vLLM main:
c5d004aaaf
---------
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
### What this PR does / why we need it?
This patch also supports v0.10.1
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
- CI passed
- test 0.10.1: https://github.com/vllm-project/vllm-ascend/pull/2583
- vLLM version: v0.10.1.1
- vLLM main:
321938e9ac
Signed-off-by: Yikun Jiang <yikunkero@gmail.com>
### 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:
9b5f64238f
Signed-off-by: rjg-lyh <1318825571@qq.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:
f58675bfb3
---------
Signed-off-by: MengqingCao <cmq0113@163.com>
### What this PR does / why we need it?
This PR fixes the bug on some machines where quantmatmul failed to run
with the NZ format. The change ensures proper execution under the
expected data layout.
### How was this patch tested?
CI passed with existing test.
- vLLM version: v0.10.1.1
- vLLM main:
b5d34af328
Signed-off-by: rjg-lyh <1318825571@qq.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:
b00e69f8ca
---------
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
