### 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:
ad32e3e19c
---------
Signed-off-by: zhoux77899 <zhouxiang100@huawei.com>
Signed-off-by: Ruri <33858552+zhoux77899@users.noreply.github.com>
Signed-off-by: Ruri <zhouxiang100@huawei.com>
aclgraph is stable and fast now. Let's drop torchair graph mode now.
TODO: some logic to adapt torchair should be cleaned up as well. We'll
do it in the following PR.
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
Co-authored-by: Mengqing Cao <cmq0113@163.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:
ad32e3e19c
Signed-off-by: lhp-deep <liuhaopeng1@huawei.com>
Co-authored-by: weijinqian0 <1184188277@qq.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?
Remove unused vanilla attn code.
### Does this PR introduce _any_ user-facing change?
NA
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c
Signed-off-by: zzzzwwjj <1183291235@qq.com>
Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>
### What this PR does / why we need it?
This PR introduces the Ascend implementation of the
`dispatch_ffn_combine` kernel and wires it into the vLLM-Ascend runtime,
together with follow‑up fixes to ensure the kernel builds and runs
correctly in CI.
- Add full host and device implementation of the `dispatch_ffn_combine`
kernel under `csrc/dispatch_ffn_combine`, including tiling logic, MOE
routing helpers, and kernel utilities for quantized FFN dispatch.
- Integrate the new kernel with the PyTorch binding
(csrc/torch_binding.cpp, csrc/torch_binding_meta.cpp) and the Ascend
runtime (vllm_ascend/ascend_forward_context.py,
vllm_ascend/worker/model_runner_v1.py).
- Extend fused MoE communication and token dispatch support in
`vllm_ascend/ops/fused_moe`, adding methods/utilities needed by the new
dispatch path.
- Update quantization logic in vllm_ascend/quantization/w8a8_dynamic.py
to support the new FFN dispatch flow.
- Fix kernel build issues by adjusting `csrc/build_aclnn.sh`, CMake
configuration, and include/namespace usage in the new kernel files.
- Add an end‑to‑end nightly test
`tests/e2e/nightly/ops/test_dispatch_ffn_combine.py` and helper
utilities in `vllm_ascend/utils.py` to validate the new kernel.
### 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/v0.12.0
---------
Signed-off-by: mojave2 <chenchen145@huawei.com>
Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>
### What this PR does / why we need it?
This pr https://github.com/vllm-project/vllm-ascend/pull/2958 is
supporting gatingtopk operator generalization, but caused nightly ci
error.
Now we add check logits for ops gatingtopk, and fix nightly ci.
- vLLM version: v0.12.0
Signed-off-by: 1092626063 <1092626063@qq.com>
### What this PR does / why we need it?
The main goal of this PR to alleviate the high maintenance burden from
model duplication when we are going to do the model optimization. Some
of our optimized models diverges a little from the vllm's modeling, but
needs to rewrite several part of original one, brings negligible
maintenance bruden to the vllm-ascend.In order to solve that, we propose
to leverage `torch.compile` and `inductor pattern matcher`,
automatically fuse the pattern we want to merge. For more details can
refer to the RFC https://github.com/vllm-project/vllm-ascend/issues/4239
This pr integrates `AddRMSNorm` and the `Quant` operator, which can
improve the inference speed of models using `w8a8 `quantization.
### Does this PR introduce _any_ user-facing change?
Yes, add new additional_config
### How was this patch tested?
```python
def main():
prompts = [
"The president of the United States is Mr.",
]
# Create a sampling params object.
sampling_params = SamplingParams(max_tokens=100, temperature=0.6, top_k=40, top_p=0.95)
# Create an LLM.
llm = LLM(
model="/root/.cache/modelscope/hub/models/vllm-ascend/Qwen3-8B-W8A8",
# enforce_eager=True,
tensor_parallel_size=1,
trust_remote_code=True,
gpu_memory_utilization=0.7,
quantization="ascend",
)
# Generate texts from the prompts.
outputs = llm.generate(prompts, sampling_params)
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```
```text
Prompt: 'The president of the United States is Mr.', Generated text: ' Trump. The president of the United States is Mr. Biden. Which of the following statements is correct? \n\nA. Mr. Trump is Mr. Biden. \nB. Mr. Trump is not Mr. Biden. \nC. The president of the United States is not Mr. Trump. \nD. The president of the United States is not Mr. Biden.\n\nThe question presents a contradiction: it states that "The president of the United States is Mr. Trump" and "The president of'
```
- vLLM version: 86e178f7c4d8c3b0eaf3c8e3f810a83f63b90e24
- vLLM main:
86e178f7c4
---------
Signed-off-by: Icey <1790571317@qq.com>
Signed-off-by: wxsIcey <1790571317@qq.com>
### What this PR does / why we need it?
Add Qwen3Next support in main
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2
---------
Signed-off-by: SunnyLee219 <3294305115@qq.com>
### What this PR does / why we need it?
This PR adds a triton rope kernel witch supports scenarios of `rope_dim
!= head_dim`. This can save the split op before rope and the concat op
after rope. Profiling shows improvement.
### Does this PR introduce _any_ user-facing change?
None
### How was this patch tested?
I will add related ut after ci integrated with triton.
- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2
---------
Signed-off-by: whx-sjtu <2952154980@qq.com>
### What this PR does / why we need it?
Fix eplb enable when using mtp float weights. It will be remove when
eplb supporting mtp and float weights.
### How was this patch tested?
Deepseek-V3 + MTP + EPLB in A3.
- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2
---------
Signed-off-by: offline0806 <3337230449@qq.com>
Signed-off-by: offline893 <158537145+offline893@users.noreply.github.com>
Co-authored-by: offline0806 <3337230449@qq.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
The M4 quantization method in ModelSlim adds bias to model weights that
originally do not have a linear bias. PR #4235 supported PD-MIX
quantization and M4 quantization, adding bias to `w8a8.py` and
`w8a8_dynamic.py`, and implementing adaptations in `ops/linear.py` to
prevent it from being reset to `None` by
`self.register_parameter("bias", None)`. However, this modification
introduced an issue where the bias was still being reset to `None` in
certain scenarios, causing errors during service startup. Therefore,
support for M4 quantization is temporarily being reverted in this PR.
___
- vLLM version: v0.11.2
Signed-off-by: SlightwindSec <slightwindsec@gmail.com>
### What this PR does / why we need it?
Support shared expert DP for deepseek_mtp feature.
`shared_expert_dp` requires `SP==True`, with corresponding parameter
restrictions.
Previously, due to the coupling between `shared_expert_dp` and torchair,
and the removal of `deepseek_mtp` in vllm_ascend, shared expert dp of
deepseek_mtp was temporarily removed.
Currently, by performing the `reduce_scatter` on the input of
deepssek_mtp in `mtp_proposer.py`, we ensure that it matches the
dimensions of `input_embedding`, and then perform the `all_gather` on
the output of mtp.
### How was this patch tested?
baseline:
<img width="1184" height="692" alt="image"
src="https://github.com/user-attachments/assets/9680d53a-7b1d-481a-accc-b8f3dae2b9e3"
/>
enable shared_expert_dp and multistream_overlap_shared_expert:
<img width="1167" height="687" alt="image"
src="https://github.com/user-attachments/assets/2531d06b-dfda-4e24-8628-6f4b0f677ddc"
/>
TPOT: 48ms -> 45.4ms
Average TPS per rank: 117.6 -> 126.1
- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2
---------
Signed-off-by: chenmenglong <chenmenglong1@huawei.com>
Signed-off-by: zengran <zengran2@huawei.com>
Co-authored-by: zengran <zengran2@huawei.com>
### What this PR does / why we need it?
In `torchair_mla.py`, the `self.oproj` function includes an additional
parameter `is_force_scatter`, while the `AscendRowParallelLinear`
function in `linear.py` does not add this parameter.
- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2
Signed-off-by: zzhx1 <zzh_201018@outlook.com>
In PD-separated deployment scenarios:
* MoE layers use dynamic quantization exclusively.
* For the Attention module, Prefill (P) nodes use **dynamic**
quantization, while Decode (D) nodes use **static** quantization.
In PD-mixed deployment scenarios:
* **All components fall back to dynamic quantization**, as it is
difficult to distinguish between Prefill and Decode tokens.
___
- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2
---------
Signed-off-by: SlightwindSec <slightwindsec@gmail.com>
Signed-off-by: Slightwind <slightwindsec@gmail.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Repair the problem of moe model accuracy caused by version upgrade.
Reason:
The new version adds the "reduce_output" operation after "forward_impl".
Then we have fully taken over the implementation of the FusedMoe module.
- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2
---------
Signed-off-by: weijinqian_v1 <weijinqian@huawei.com>
Co-authored-by: weijinqian_v1 <weijinqian@huawei.com>
### What this PR does / why we need it?
The expert mapping table and weights of the dynamic EPLB were not
updated, causing the accuracy to be correct but not effective. This bug
has now been fixed.
- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2
---------
Signed-off-by: shenchuxiaofugui <1311027364@qq.com>
### What this PR does / why we need it?
qwen3-next suppot triton chunk_gated_delta_rule ops
### co-owners
@OsirisDuan
- vLLM version: v0.11.2
Signed-off-by: shiyuan680 <917935075@qq.com>
### What this PR does / why we need it?
Currently, there are two paths to judge the chip type in code,
`get_ascend_soc_version` use `get_soc_version` api in torch_npu, and
`is_310p` `use _build_info.__soc_version__`, which generate when
install. We need to unify the two paths.
We need to unify these codes based on the following points:
1. We need to ensure consistency in chip type judgment between compiling
and running states;
2. In compiling state, we need chip type to complete op's compilation,
but in running state, we only need device
type(910B/910_93/310P/910_95/etc) to make code branch judgement;
3. In compiling state, torch_npu may not have been installed yet, so we
can't use torch_npu's api.
Based on the above points, we have made the following changes:
1. When user set env `SOC_VERSION`, use it; when not set, query
soc_version by `npu-smi`;
2. generate device_type based on soc_version when compiling, and write
`__device_type__` instead of `__soc_version__` in `_build_info.py`;
3. In running state, use `__device_type__` to judge code branch.
### Does this PR introduce _any_ user-facing change?
When not set env `SOC_VERSION`, it will not be `ASCEND910B1` by default,
we will query soc_version by `npu-smi`. And env `SOC_VERSION` must be in
the list `soc_to_device` in `setup.py`.
- vLLM version: v0.11.0
- vLLM main:
2918c1b49c
Signed-off-by: zzzzwwjj <1183291235@qq.com>
### What this PR does / why we need it?
mkdir triton package and move triton files
- vLLM version: v0.11.0
- vLLM main:
2918c1b49c
Signed-off-by: shiyuan680 <917935075@qq.com>
### What this PR does / why we need it?
Temporarily fix the oom issue, will update to vllm's plan later.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
e2e&ut
- vLLM version: v0.11.0
- vLLM main:
2918c1b49c
Signed-off-by: Pr0Wh1teGivee <calvin_zhu0210@outlook.com>
The main purposes of this PR are as follows:
1. Remove the multicast-related code;
Reason:
1. In the scenario like a2 Dual-System Back-to-Back Networking，the
performance is worse than all_gather. Before the modification, in e2e
test, it was 3 tps; after the modification, it is 10 tps.
2. At the same time, we usually enable the SP feature，it is consistent
with the current logic.
3. The advantage of broadcast communication lies in the fact that it
does not suffer from uneven DP load and does not require the prefill ACL
graph to be enabled. But we support prefill Acl graph recently.
So we think there is no need to maintain the multicast as one choice in
moe communication.
Performance benefits are as follows:
When not enable_flashcomm1, TTFT remains relatively stable at around
43000ms, which is approximately 15000ms faster than before the
modification.
When enable_flashcomm1, there is no diffenence, TTFT remains relatively
stable at around 29000ms.
- vLLM version: v0.11.0
- vLLM main:
2918c1b49c
---------
Signed-off-by: weijinqian_v1 <weijinqian@huawei.com>
Signed-off-by: weijinqian0 <1184188277@qq.com>
Co-authored-by: weijinqian_v1 <weijinqian@huawei.com>
There is a lot hack code for v0.11.0, which makes the code hard to
upgrade to newer vLLM version. Since v0.11.0 will release soon. Let's
drop v0.11.0 support first. Then we'll upgrade to v0.11.2 soon.
- vLLM version: v0.11.0
- vLLM main:
2918c1b49c
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
### What this PR does / why we need it?
Eplb Verify Fix
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
- vLLM version: v0.11.0
- vLLM main:
2918c1b49c
---------
Signed-off-by: shenchuxiaofugui <1311027364@qq.com>
Signed-off-by: LI SHENGYONG <49200266+shenchuxiaofugui@users.noreply.github.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
### What this PR does / why we need it?
Add information on the scope of EPLB support.
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
- vLLM version: v0.11.0
- vLLM main:
2918c1b49c
---------
Signed-off-by: shenchuxiaofugui <1311027364@qq.com>
### What this PR does / why we need it?
Redundant experts bugfix
### Does this PR introduce _any_ user-facing change?
After configuring the path for experts_map, users do not need to
configure iinit_redundancy_expert.
### How was this patch tested?
The accuracy of EPLB was tested with and without the use of redundant
experts.
- vLLM version: v0.11.0
- vLLM main:
2918c1b49c
---------
Signed-off-by: shenchuxiaofugui <1311027364@qq.com>
### What this PR does / why we need it?
Support the Qwen3-Next-80B-A3B-Instruct quantization model and Fix the
NZ issue. Triton kernel doesn't support data format nz, thus we skip
converting weight to nz on layer `conv1d`
- vLLM version: v0.11.0
- vLLM main:
2918c1b49c
---------
Signed-off-by: IncSec <1790766300@qq.com>
### What this PR does / why we need it?
avoid mrope fusion op when running qwen2.5-vl on a+x machine
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
Test text VQA accuracy on G8600 with aisbench
- vLLM version: v0.11.0
- vLLM main:
2918c1b49c
---------
Signed-off-by: 李少鹏 <lishaopeng21@huawei.com>
### What this PR does / why we need it?
Past：
npu_moe_gating_top_k can only support 'group_count=256' pattern
Now：
1、npu_moe_gating_top_k support all size of group_count
2、the functionality of `torch_npu.npu_moe_gating_top_k_softmax` are
included in `torch_npu.npu_moe_gating_top_k`
CANN: depends on 8.3.RC1
Performance：
1. GLM4.5-w8a8, TPS improve 6%
2. Qwen3, the same as before
- vLLM version: v0.11.0
- vLLM main:
2918c1b49c
Signed-off-by: 1092626063 <1092626063@qq.com>
### What this PR does / why we need it?
In PR https://github.com/vllm-project/vllm-ascend/pull/3420, we
initially placed the quantization type (quant_type) in the MoECommMethod
class. However, since MoECommMethod follows a singleton pattern, it
couldn't accommodate scenarios where different layers in the model might
use different quantization approaches (e.g., MTP modules using
floating-point computation while the main model employs quantized
computation).
In this PR, we've moved the quantization type to the AscendFusedMoe
class and pass it as a parameter to MoECommMethod.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
```bash
export HCCL_BUFFSIZE=1024
export VLLM_VERSION=0.11.0
vllm serve /home/data/DeepSeek-R1_w8a8/ \
--data-parallel-size 2 \
--tensor-parallel-size 8 \
--enable-expert-parallel \
--served-model-name dsv3 \
--max-model-len 32768 \
--max-num-batched-tokens 4096 \
--max-num-seqs 16 \
--quantization ascend \
--trust-remote-code \
--gpu-memory-utilization 0.9 \
--speculative-config '{"num_speculative_tokens": 2, "method":"deepseek_mtp"}'
```
- vLLM version: v0.11.0
- vLLM main:
83f478bb19
---------
Signed-off-by: realliujiaxu <realliujiaxu@163.com>
### What this PR does / why we need it?
Supports generalized FlashComm2 optimization, which reduces
communication overhead, decreases RmsNorm computation, and saves one
AllGather step by replacing Allreduce operations in the Attention module
with pre-AlltoAll and post-AllGather operations (used in combination
with FlashComm1). This feature is enabled during the Prefill phase and
is recommended to be used together with FlashComm1, delivering broad
performance improvements, especially in long sequence scenarios with
large tensor parallelism (TP) configurations. Benchmark tests show that
under TP16DP1 configuration, it can improve the prefill performance of
the DeepSeek model by 8% on top of FlashComm1.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.11.0
- vLLM main:
83f478bb19
---------
Signed-off-by: zzhxx <2783294813@qq.com>
Signed-off-by: Levi-JQ <yujinqi2@huawei.com>
Co-authored-by: Levi-JQ <yujinqi2@huawei.com>
Co-authored-by: zzhxx <2783294813@qq.com>
### What this PR does / why we need it?
Adapts mtp function to Qwen3-next.
- vLLM version: v0.11.0
- vLLM main:
83f478bb19
Signed-off-by: drslark <slarksblood@qq.com>
### What this PR does / why we need it?
1、update prepare_finalize.py：fix A2 accuracy problem when pcp and dcp
- vLLM version: v0.11.0
- vLLM main:
83f478bb19
Signed-off-by: LookAround <lixushi@huawei.com>
### What this PR does / why we need it?
move quant before allgather in Allgather EP, rely on
https://github.com/vllm-project/vllm-ascend/pull/3334
Deepseek R1 W8A8 performance on A2 with
`HCCL_ALGO="level0:NA;level1:pipeline"`:
| Seq length | Mean TTFT (ms) main | Mean TTFT (ms) this PR |
|----------|----------|----------|
| 4k | 375.21 | 364.99 |
| 16k | 1465.23 | 1421.75 |
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.11.0
- vLLM main:
83f478bb19
---------
Signed-off-by: realliujiaxu <realliujiaxu@163.com>
### What this PR does / why we need it?
This PR upgrade CANN from 8.2rc1 to 8.3rc1 and remove the CANN version
check logic.
TODO: we notice that UT runs failed with CANN 8.3 image. So the base
image for UT is still 8.2. We'll fix it later.
- vLLM version: v0.11.0
- vLLM main:
83f478bb19
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
### What this PR does / why we need it?
Upgrade torch-npu to the official release version 2.7.1
- vLLM version: v0.11.0
- vLLM main:
83f478bb19
---------
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
