### What this PR does / why we need it?
This PR eliminates the simplicit HD synchronization in sfa backend, and
_build_dummy_attn_metadata and dummy_run in mtp_proposer, significantly
improving dsv3.2 performance in low-latency scenarios.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Performance improvements are observed with E2E performance serving (P:
DP4TP8EP32 D: DP8TP4EP32) with `num_speculative_tokens=3`.
DSV3.2-W8A8-EXP:
TPOT: 41.67ms -> 23.36ms
ITL: 85.93ms -> 55.96ms
DSV3.2-W8A8 (relaesed in December):
TPOT: 18.11ms
ITL: 56.13ms
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
Signed-off-by: linfeng-yuan <1102311262@qq.com>
### What this PR does / why we need it?
We introduced the npugraph_ex backend through the vllm's adaptor
dispatch mechanism to accelerate aclgraph. This solution is based on
torch.compile and uses torchair to optimize the fx.graph. The
performance gains are mainly obtained from the static kernel. We
conducted tests on Qwen3-30B and achieved over 5% performance
optimization.
### Does this PR introduce _any_ user-facing change?
Yes, we add a new switch named"enable_npugraph_ex" in additional_config,
default is False.
We also add an example to show how to register custom replacement pass
### More information about this PR
This feature depends on the release of CANN and torch_npu in Q4.
We tested it on a package that has not been publicly released yet and
verified that the functionality works.
This feature is still experimental at the moment; setting the config
true will directly raise error.
Merging into the main branch initially involves some preliminary commits
to facilitate subsequent development and testing of the feature, as well
as to avoid submitting an excessively large PR at once.
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
---------
Signed-off-by: chencangtao <chencangtao@huawei.com>
Signed-off-by: ChenCangtao <50493711+ChenCangtao@users.noreply.github.com>
Co-authored-by: chencangtao <chencangtao@huawei.com>
Co-authored-by: panchao-hub <315134829@qq.com>
Co-authored-by: wbigat <wbigat@163.com>
Co-authored-by: Mengqing Cao <cmq0113@163.com>
…
### What this PR does / why we need it?
When speculative decoding is enabled and temperature > 0, bonus_logits
and target_logits are sampled separately:
1. bonus_logits are sampled using a fused torch_npu.npu_top_k_top_p
operator invoked inside the main sampler,
2. while target_logits are sampled within the rejection sampler using a
less-optimized implementation composed of smaller operators.
Consequently, the cumsum operation in the top-p sampling for
target_logits becomes especially time-consuming, leading to performance
degradation.
<img width="1029" height="623" alt="image"
src="https://github.com/user-attachments/assets/1969f561-6aa5-41b3-9a87-1f64d4321cbf"
/>
Apply the fused operator to the sampling of target_logits as well to
reduce overhead
<img width="1039" height="572" alt="image"
src="https://github.com/user-attachments/assets/1e6563da-3418-405d-b657-7bbe10dd0924"
/>
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
---------
Signed-off-by: funanyang <985619145@qq.com>
Co-authored-by: weijinqian0 <1184188277@qq.com>
### What this PR does / why we need it?
The first commit support `FULL_DECODE_ONLY`:
- Update `AscendSFAMetadataBuilder` to use `num_input_tokens` for
slicing slots and positions, ensuring fixed tensor shapes.
- Implement padding logic for `query_start_loc` in `NPUModelRunner` to
support uniform decode in full graph mode, aligning with GPU runner
behavior.
- Adjust MLA cosine cache allocation to occur independently of graph
mode and switch to using device-resident sequence lengths for attention
metadata.
- Remove redundant slicing of hidden states and outputs in
`AscendSFAImpl` and optimize `sin`/`cos` cache updates.
The second commit take MTP into account:
- Update `AscendSFAMetadataBuilder` to use `num_input_tokens` for
slicing slots and positions, ensuring fixed tensor shapes.
- Implement padding logic for `query_start_loc` in `NPUModelRunner` to
support uniform decode in full graph mode, aligning with GPU runner
behavior.
- Adjust MLA cosine cache allocation to occur independently of graph
mode and switch to using device-resident sequence lengths for attention
metadata.
- Remove redundant slicing of hidden states and outputs in
`AscendSFAImpl` and optimize `sin`/`cos` cache updates.
And the rest of them are just bugfix.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
Test cases needed.
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
---------
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
### What this PR does / why we need it?
The logprobs_tensor was not initialized before accessing its token_id
member, leading to a crash when tokenizer.decode() is called by passing
a negative token_id
### How was this patch tested?
Constructed an inference request with two prompts and set
SamplingParams(prompt_logprobs=<non-None value>) (e.g.,
prompt_logprobs=1).
After applying the fix (proper initialization of logprobs_tensor), the
same request completed successfully without errors, and the returned
logprobs matched expected values.
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
Signed-off-by: jiangweixiang <jwx02384838@antgroup.com>
Co-authored-by: jiangweixiang <jwx02384838@antgroup.com>
Co-authored-by: Mengqing Cao <cmq0113@163.com>
### What this PR does / why we need it?
1.Add the implementation of normal Aclnn operators: MoeCombineNormal,
MoeDispatchNormal, NotifyDispatch,and DispatchLayout.
- MoeCombineNormal: Implements the combine logic within MoE operations.
- MoeDispatchNormal: Implements the dispatch logic within MoE
operations.
- NotifyDispatch: Exchanges topk_idx information among different ranks
to calculate the device memory required for the dispatch stage.
- DispatchLayout: Used to calculate information related to the device
memory layout for the dispatch stage.
2.Provide PyTorch interfaces for normal operators—get_dispatch_layout,
dispatch_prefill, and combine_prefill—to be used for MoE communication
during the prefill stage in vLLM.
- get_dispatch_layout: Calculates information related to the device
memory layout for the dispatch operator, and is called before
dispatch_prefill.
- dispatch_prefill: Initiates the dispatch operation.
- combine_prefill: Initiates the combine operation.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
The functionality has already been validated using the local Qwen model.
Test cases will be added after support for multi-NPU use cases in the CI
pipeline is finalized.
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
Signed-off-by: shiro-zzzz <zhangdianhao@huawei.com>
### What this PR does / why we need it?
To support pipeline parallel with PD disaggregation, this PR support PP
in mooncake connector and fix other bugs when enable pp with other
optimization params, including following changes:
- mooncake connector support pp in prefill, we do not support decode pp
currently
- fix bugs when enable both pp and flashcomm1
- optimize ascend-scheduler to support full batch in multiple pipeline
stages, original implementation would cause all pipeline stages
batch_size total summed to max_num_seq, which makes pipeline is not
full, this optimization can make all stages running with full batch_size
= max_num_seq, the same changes will contribute to vllm scheduler too.
### Does this PR introduce _any_ user-facing change?
add `pp_size` in mooncake connector kv_connector_extra_config
```
"kv_connector_extra_config": {
"use_ascend_direct": true,
"prefill": {
"dp_size": 1,
"tp_size": 4,
"pp_size": 4
},
"decode": {
"dp_size": 16,
"tp_size": 1
}
}
```
### How was this patch tested?
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
---------
Signed-off-by: chenxiao <Jaychou1620@Gmail.com>
Signed-off-by: Kurumi5210 <Jaychou1620@Gmail.com>
Signed-off-by: Kurumi5210 <jaychou1620@gmail.com>
Signed-off-by: 秋刀鱼 <jaychou1620@Gmail.com>
Co-authored-by: chenxiao <Jaychou1620@Gmail.com>
Co-authored-by: zss <zss@qq.com>
Co-authored-by: zss <3265779424@qq.com>
### What this PR does / why we need it?
Adds W4A16 quantization method for the Kimi-K2-Thinking model and
updates relevant modules to support the new quantization method.
- Implements complete W4A16 quantization method including weight
packing/unpacking, per-group quantization parameter generation,
post-processing logic and MoE method application.
- Adds parameters `use_int4_w4a16`, `w1_offset` and `w2_offset`, adjusts
`with_quant` conditional logic to support W4A16 matrix multiplication.
- Adds `packed_modules_model_mapping` for Kimi-K2-Thinking model and
processing logic for `weight_packed` field.
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
---------
Signed-off-by: zhoux77899 <zhouxiang100@huawei.com>
Signed-off-by: Ruri <33858552+zhoux77899@users.noreply.github.com>
Signed-off-by: Ruri <zhouxiang100@huawei.com>
### What this PR does / why we need it?
Set the global env `TRANSFORMERS_OFFLINE: 1`, which will avoid
downloading the file and return the path to the
local cached file if it exists when using modelscope's
`snapshot_download` api
Signed-off-by: wangli <wangli858794774@gmail.com>
### What this PR does / why we need it?
This patch do some tiny optimization for nightly ci:
1. Polling the frequency with which the service prints logs when it
starts up in order to obtain useful information more quickly.
2. Shorten the timeout for waiting server
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
---------
Signed-off-by: wangli <wangli858794774@gmail.com>
### What this PR does / why we need it?
When `hf_quant_cfg` is not None and `hf_quant_cfg.quant_method == ""`,
func `override_quantization_method` will return None and raise
ValidationError.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
Signed-off-by: zzzzwwjj <1183291235@qq.com>
### What this PR does / why we need it?
As there is not accuracy test for qwen3-235B-A22B model
Test result:
dataset version metric mode vllm-api-general-chat
--------- --------- -------- ------ -----------------------
gsm8k 7cd45e accuracy gen 96.29
Times long for test case running: 30mintues
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
Signed-off-by: leo-pony <nengjunma@outlook.com>
Co-authored-by: Mengqing Cao <cmq0113@163.com>
### What this PR does / why we need it?
We didn’t account for this earlier because we didn’t have A3 in CI, but
now that we do, this test case needs a few extra tweaks — please take a
look at `profile_run`.
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
Co-authored-by: Mengqing Cao <cmq0113@163.com>
What this PR does / why we need it?
Improve usability,local_buffer_size support for units: GB, MB, KB, B,
For example, "2GB"
{
"local_hostname": "XXX.XXX.XXX.XXX",
"metadata_server": "P2PHANDSHAKE",
"protocol": "ascend",
"device_name": "",
"use_ascend_direct": true,
"master_server_address": "XXX.XXX.XXX.XXX:50088",
"global_segment_size": 60000000000,
"local_buffer_size": "2GB"
}
Does this PR introduce any user-facing change?
local_buffer_size support for units: GB, MB, KB, B
How was this patch tested?
Mooncake configures local_buffer_size as GB, MB, KB, B
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
Signed-off-by: lty <linxianchong1@huawei.com>
### What this PR does / why we need it?
This PR adapt DispatchGmmCombineDecode operator to parameters of small
operators.
1. This operator no longer requires permuting the weights and scales of
GMM1.
2. This operator no longer requires transposing the weights of GMM2.
Therefore, this operator and the small operator can use the same
parameters (weights and scales), which is beneficial for model
adaptation.
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
Signed-off-by: wangqiankun <wangqiankun13@huawei.com>
### What this PR does / why we need it?
Fix configuration error in our documentations.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
NA.
Signed-off-by: linfeng-yuan <1102311262@qq.com>
### What this PR does / why we need it?
In reinforcement learning scenarios, the current inference applies a
transpose operation to the weights. For a cleaner architecture, the
weight transpose module was moved to wakeup.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
Signed-off-by: lhp-deep <liuhaopeng1@huawei.com>
Co-authored-by: weijinqian0 <1184188277@qq.com>
### What this PR does / why we need it?
Fix dp padding logic in dummyrun. After
https://github.com/vllm-project/vllm/pull/28579, `num_tokens` will be
padded in `CudagraphDispatcher`, thus we also need to do the pad in the
dummy_run.
### How was this patch tested?
Test locally with the following scripts
```bash
VLLM_USE_MODELSCOPE=true python3 -m vllm.entrypoints.openai.api_server \
--model wemaster/deepseek_mtp_main_random_bf16 \
--trust-remote-code \
--data-parallel-size 4 \
--tensor-parallel-size 1 \
--compilation-config '{"cudagraph_capture_sizes":[96],"cudagraph_mode":"FULL_DECODE_ONLY"}' \
--enable-expert-parallel
```
```bash
vllm bench serve --model wemaster/deepseek_mtp_main_random_bf16 --endpoint /v1/completions --dataset-name random --random-input 512 --random-output 100 --num-prompts 48 --request-rate 1 --ready-check-timeout-sec 0
```
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
Signed-off-by: MengqingCao <cmq0113@163.com>
### What this PR does / why we need it?
Qwen2.5-VL mrope precision problem would been solved once this pr is
merged
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Test on G8600 with textVQA dataset
- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2
---------
Signed-off-by: 李少鹏 <lishaopeng21@huawei.com>
Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>
### What this PR does / why we need it?
1.Add the implementation of normal Aclnn operators: MoeCombineNormal,
MoeDispatchNormal, NotifyDispatch,and DispatchLayout.
- MoeCombineNormal: Implements the combine logic within MoE operations.
- MoeDispatchNormal: Implements the dispatch logic within MoE
operations.
- NotifyDispatch: Exchanges topk_idx information among different ranks
to calculate the device memory required for the dispatch stage.
- DispatchLayout: Used to calculate information related to the device
memory layout for the dispatch stage.
2.Provide PyTorch interfaces for normal operators—get_dispatch_layout,
dispatch_prefill, and combine_prefill—to be used for MoE communication
during the prefill stage in vLLM.
- get_dispatch_layout: Calculates information related to the device
memory layout for the dispatch operator, and is called before
dispatch_prefill.
- dispatch_prefill: Initiates the dispatch operation.
- combine_prefill: Initiates the combine operation.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
The functionality has already been validated using the local Qwen model.
Test cases will be added after support for multi-NPU use cases in the CI
pipeline is finalized.
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
Signed-off-by: shiro-zzzz <zhangdianhao@huawei.com>
### What this PR does / why we need it?
After enabling Mlapo and DCP, since Mlapo has its own mla_preprocess
logic and does not perform additional all_gather operations on the DCP
group, this will lead to dimension mismatch during the subsequent
forward proces
### Does this PR introduce _any_ user-facing change?
N/A
- vLLM version: v0.12.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/ad32e3e19ccf0526cb6744a5fed09a138a5fb2f9
Signed-off-by: zengran <zengran2@huawei.com>
Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>