### What this PR does / why we need it?
This helps to fix a bug in for pa get_workspace. In earlier
implementation, we use `_npu_paged_attention_get_workspace` in
`_update_pa_attn_params`. However, this might cause some potential
memory problems as it dynamically allocate new memory for workspace when
calling this api. Therefor, we move this back to capturing, and use a
fixed `SEQ_LEN_WITH_MAX_PA_WORKSPACE` to get max workspace.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
Signed-off-by: Angazenn <supperccell@163.com>
### What this PR does / why we need it?
This patch purpose to optimize the lint check term. The main idea is to
reduce unnecessary installation time.
1. The installation of vllm is not must, only append the path of vllm
src to the `PATHONPATH` is effective
2. This installation of `requirements-dev.txt` is not must, we have a
pre-built image `quay.io/ascend-ci/vllm-ascend:lint` with all the
requirements installed in advance.
**NOTE**: the conditions for triggering image builds are: 1).Daily
scheduled build; 2) Build when requirements are modified; 3) Manual
build. This ensures that the dependencies in our image are up-to-date to
the greatest extent possible.
3. The `mypy` was separated from the `pre-commit` hook for performance
reasons; we found that integrating `mypy` into the `pre-commit` hook
resulted in poor performance.
4. Reduce the CPU core consumption from 16 -> 8
### Does this PR introduce _any_ user-facing change?
The end-to-end lint time was optimized from 20min/per PR to 8min/per PR
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
---------
Signed-off-by: wangli <wangli858794774@gmail.com>
### What this PR does / why we need it?
When running the Qwen2.5-Omni-7B model on Ascend NPU, the engine fails
during the profiling/warmup stage with the following error:
`AclNN_Runtime_Error(EZ9903): rtKernelLaunchWithHandleV2 failed: 507035.
The vector core execution is abnormal.`
error log:
https://github.com/vllm-project/vllm-ascend/actions/runs/21144534911/job/60806765393#step:17:6412
This error is specifically triggered by the `triton_mrope` kernel when
handling the unique `mrope_section` configurations of the Omni model.
Other multimodal models with standard sections (e.g., [16, 24, 24]) or
standard LLMs work correctly with Triton.
Modified vllm_ascend/ops/rotary_embedding.py to add a conditional check
before calling forward_triton.
1. For standard LLMs (mrope_interleaved = True ), it continues to use
Triton for acceleration.
2. For complex configurations (like Qwen2.5-Omni mrope_interleaved =
False ), it now falls back to the native super().forward_oot() path,
which uses the stable torch_npu or PyTorch implementation.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
d68209402d
Signed-off-by: hfadzxy <starmoon_zhang@163.com>
### What this PR does / why we need it?
When using the full_decode_only mode, the vllm framework will still use
the torch.fx.passes.split_module.split_module API to process the
corresponding GraphModule of the model.
However, the output of this API may cause the output of the fx graph to
no longer be a tuple, and torch.compile enforces strict checks on this.
Previously, we manually modified the fx graph, which introduced an
abnormality in the model output type.
In this PR, we switched to using PyTorch's native API to modify the FX
graph, and removed the code that was previously added to handle output
type anomalies.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
---------
Signed-off-by: chencangtao <chencangtao@huawei.com>
Co-authored-by: chencangtao <chencangtao@huawei.com>
According to the official documentation, the parameter
"draft_tensor_parallel_size": 1 is supposed to be applied to the Eagle3
model. However, based on actual debugging, it was found that the number
of tensor parallelisms (tp) of the Eagle model is consistent with that
of the target model. The setting of tp for the draft model did not take
effect as expected.
**Note:** This feature has not been superimposed and tested with `sp`
and `dp`. It will be adapted later
No
```python
from vllm import LLM, SamplingParams
def main():
prompts = [
"The future of AI is",
]
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
llm = LLM(
model="meta-llama/Llama-3.1-8B-Instruct",
tensor_parallel_size=4,
gpu_memory_utilization=0.9,
enforce_eager=True,
speculative_config={
"method": "eagle3",
"model": "yuhuili/EAGLE3-LLaMA3.1-Instruct-8B"
"draft_tensor_parallel_size": 1,
"num_speculative_tokens": 3,
},
)
outputs = llm.generate(prompts, sampling_params)
print(f"Outputs: {outputs}")
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```
Fixesvllm-project/vllm#31345
### What this PR does / why we need it?
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
d68209402d
Signed-off-by: zhaomingyu <zhaomingyu13@h-partners.com>
Co-authored-by: drslark <slarksblood@qq.com>
### What this PR does / why we need it?
Operator `DispatchGmmCombineDecode` does not support non-W8A8 scenarios
and cannot share the same communication domain with Operator
`Dispatch`/`Combine`.
> for instance, when the draft model uses a non-W8A8 MOE architecture
while the main model employs a W8A8 MOE architecture.
Therefore days ago, I implemented an interception that unconditionally
disables Operator `DispatchGmmCombineDecode` whenever the speculative
mode is `EAGLE` or `EAGLE-3`. [PR:
5293](https://github.com/vllm-project/vllm-ascend/pull/5293)
However, this approach was not precise enough.
This PR further refines the logic by specifically identifying the draft
model's configuration: Operator `DispatchGmmCombineDecode` will now be
disabled only when the draft model uses an MOE architecture and is
non-W8A8.
More info about this operator, please refer to RFC: issue
https://github.com/vllm-project/vllm-ascend/issues/5476
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Acc
test qwen3-235b eplb on a single A3 node(ep16),
with dispatch_gmm_combine_decode
```shell
nic_name="xxxx"
local_ip="xxx.xxx.xxx.xxx"
export HCCL_IF_IP=$local_ip
export GLOO_SOCKET_IFNAME=$nic_name
export TP_SOCKET_IFNAME=$nic_name
export HCCL_SOCKET_IFNAME=$nic_name
export VLLM_ASCEND_ENABLE_FUSED_MC2=2
echo "VLLM_ASCEND_ENABLE_FUSED_MC2=${VLLM_ASCEND_ENABLE_FUSED_MC2}"
export HCCL_OP_EXPANSION_MODE="AIV"
export HCCL_BUFFSIZE=512
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
export OMP_PROC_BIND=false
export OMP_NUM_THREADS=10
export LD_PRELOAD=/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD
vllm serve /dataset/Qwen3-235B-A22B-Instruct-2507-w8a8-QuaRot/ \
--served-model-name "qwen" \
--host 0.0.0.0 \
--port 8004 \
--async-scheduling \
--tensor-parallel-size 4 \
--data-parallel-size 4 \
--max-num-seqs 64 \
--max-model-len 40960 \
--max-num-batched-tokens 16384 \
--gpu-memory-utilization 0.9 \
--enable-expert-parallel \
--no-enable-prefix-caching \
--quantization "ascend" \
--trust-remote-code \
--speculative_config \
'{
"method": "eagle3",
"model": "/dataset/Qwen3-235B-A22B-Instruct-2507-speculator-eagle3/",
"num_speculative_tokens": 2
}' \
--compilation-config '{"cudagraph_mode": "FULL_DECODE_ONLY"}' \
2>&1 | tee qwen3_235b_eagle3.log
```
| dataset | version | metric | mode | vllm-api-stream-chat |
|----- | ----- | ----- | ----- | -----|
| aime2024 | 604a78 | accuracy | gen | 80.00 |
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
Signed-off-by: wangqiankun <wangqiankun13@huawei.com>
This PR addresses a request ID mismatch issue in the PD
(Prefill-Decoding) separation deployment scenario for vllm-ascend.
Upstream vLLM recently mitigated request ID collisions by appending a
random suffix to each request_id (e.g., req-123 → req-123-abc), refer to
[PR-27987](https://github.com/vllm-project/vllm/pull/27987 ) &
[PR-29665](https://github.com/vllm-project/vllm/pull/29665). While this
works in single-node deployments, it breaks compatibility in
PD-separated setups: the Producer (Prefill node) and Consumer (Decoding
node) end up with different request_id values, preventing the Consumer
from correctly retrieving the KV cache generated by the Producer.
To resolve this, this PR introduces a new field remote_request_id in the
metadata passed via mooncake_connector. The Producer preserves and
forwards the original (unmodified) request_id as remote_request_id. The
Consumer then uses this remote_request_id—instead of its locally
generated suffixed ID—to fetch the correct KV cache from the Prefill
node.
This ensures consistent request identification across PD nodes while
maintaining compatibility with upstream vLLM’s request ID deduplication
mechanism.
<img width="1279" height="781" alt="image"
src="https://github.com/user-attachments/assets/274238c1-dab6-4d3a-9ee4-6e578679b762"
/>
- vLLM version: v0.13.0
- vLLM main:
d68209402d
Signed-off-by: ghphotoframe <854746559@qq.com>
Co-authored-by: jiangweixiang <jwx02384838@antgroup.com>
Signed-off-by: zzhx1 <zzh_201018@outlook.com>
### What this PR does / why we need it?
> Extracted from PR #5513
Based on the Sharded-CP feature PR:#4702;
RFC:https://github.com/vllm-project/vllm/issues/30055
### Support FULL_DECODE_ONLY Mode under PD-Mixed Scenario:
Extends DSA-CP to handle the FULL_DECODE_ONLY execution mode when
running in a prefill-decode mixed (PD-mixed) serving environment,
improving throughput and resource utilization for decode-intensive
workloads.
**In pure prefill nodes:**
- Both q_proj and o_proj are sharded across world ranks, using
**broadcast** for weights distribution.
**In PD-mixed nodes (supporting both prefill and decode):**
- q_proj is fully replicated (not sharded) to avoid communication
overhead during decoding.
- o_proj Using the original TP `RowParallelLinear` method to store
weights
**During prefill execution:**
- o_proj forwards through all_gather to collect weights, reconstructing
the complete o_proj weights on each card.
**During decode (graph replay phase):**
- Additional all_to_all (before o_proj) and reduce_scatter (after
o_proj) are introduced to enable sequence-parallel output aggregation
while maintaining correctness under SFA CP.
### benchmark:
- TTFT increased by **527%**
- TPOT increased by **180%**
<img width="1550" height="938" alt="image"
src="https://github.com/user-attachments/assets/9b7a03d8-a3db-4a99-8923-6e5bfcfecf72"
/>
### Does this PR introduce _any_ user-facing change?
None
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
---------
Signed-off-by: zzhx1 <zzh_201018@outlook.com>
Signed-off-by: zzhxx <zhangzihang23@mails.ucas.ac.cn>
Co-authored-by: clrs97 <524936896@qq.com>
### What this PR does / why we need it?
1) Default enable MLAPO for deepseek MLA Attention W8A8 models on PD
disagregation D Instance, for example: DeepSeekV3-W8A8,
DeepSeek-R1-W8A8.
2) Default enable MLAPO for DeepSeek SFA Attention W8A8 models,
currently is DeepSeek-V3.2-W8A8.
### Does this PR introduce _any_ user-facing change?
Don't need use manully to VLLM_ASCEND_ENABLE_MLAPO=1, to enable MLAPO
feature for deepseek w8a8 model
The effect of enabling MLAPO SFA model deployed on a single A3 Node:
Test
with:tests/e2e/nightly/single_node/models/test_deepseek_v3_2_exp_w8a8.py
dataset: gsm8k-lite，without set MTP, FULL GRAPH, has 19% promote：
未默认开启 MLAPO 时：
├─────────────────────────┤
│ TTFT │ 14055.8836 ms │
├─────────────────────────┤
│ ITL │ 66.8171 ms. │
├─────────────────────────┤
│ Output Token Throughput │ 104.9105 token/s │
├─────────────────────────┤
默认开启 MLAPO 时：
├─────────────────────────┤
│ TTFT │ 3753.1547 ms │
├─────────────────────────┤
│ ITL. │ 61.4236 ms. │
├─────────────────────────┤
│ Output Token Throughput │ 125.2075 token/s│
├─────────────────────────┤
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
---------
Signed-off-by: leo-pony <nengjunma@outlook.com>
### What this PR does / why we need it?
1. Implement a **high-performance Triton custom kernel** for the rotary
position embedding (RoPE) operator on **Ascend NPU** platform
2. Fix critical bugs in the Triton RoPE kernel registration and
invocation process: including incorrect fake impl function name
matching, wrong torch ops namespace for kernel call, missing self
parameter in cos/sin slice fetching, and syntax errors in function type
annotations.
3. Achieve **extreme performance optimization** for the core RoPE
operator: the single inference latency is reduced from **57.1 μs** to
**9 μs**, with **6.34x performance improvement** and **84.24% latency
reduction**.
4. The RoPE operator is a **hot path** that is executed in every
transformer layer during LLM inference, the optimization will directly
reduce the overall inference latency and improve the throughput of LLM
serving on Ascend NPU.
5. Keep full backward compatibility: the Triton kernel is enabled only
when `HAS_TRITON=True`, and automatically fall back to the original
Ascend NPU native implementation if Triton is not available, no
functional regression.
### Does this PR introduce _any_ user-facing change?
**NO**
- No changes to any public APIs, interfaces or inference behaviors of
vLLM.
- No impact on the text generation quality and correctness of the large
model.
- The optimization is transparent to end users, only the inference speed
(latency/throughput) is improved without any functional change.
### How was this patch tested?
1. **Environment Validation**: Tested on Ascend NPU platform with
vLLM-Ascend framework, Triton library installed and enabled
(`HAS_TRITON=True`).
2. **Kernel Registration Test**: Verified the Triton RoPE kernel
(`rope_forward_triton`) is successfully registered to
`torch.ops._C_ascend` namespace without any
`ValueError/NameError/SyntaxError`.
3. **Functional Correctness Test**: Run large model (GLM4/MoE) inference
on the Ascend NPU platform, the generated text content is **completely
correct** (no garbled text, no logical errors), consistent with the
original implementation.
4. **Performance Benchmark Test**: Measure the single execution latency
of the RoPE operator before/after optimization, confirm the latency is
stably reduced from 57.1 μs to 9 μs, the performance gain is valid and
stable.
5. **Fallback Mechanism Test**: Manually disable Triton
(`HAS_TRITON=False`), verify the code correctly falls back to the
original Ascend NPU native RoPE implementation, no service crash and
normal inference.
6. **Compatibility Test**: Test with different tensor shapes/sizes of
query/key, all cases work correctly with the Triton kernel, no shape
mismatch error.
- operator supply by Hexiang Wang
- vLLM version: v0.13.0
- vLLM main:
11b6af5280
---------
Signed-off-by: ZCG12345 <2097562023@qq.com>
### What this PR does / why we need it?
Update causal_conv1d_update ops for better perf.
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
---------
Signed-off-by: SunnyLee219 <3294305115@qq.com>
### What this PR does / why we need it?
In long-sequence scenarios, the chunked-prefill component may encounter
dimension misalignment issues, which previously occurred during
precision testing on the code_generate_lite dataset. This PR removes
redundant computations and instead derives the value using existing
results and straightforward calculations.
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
Signed-off-by: QiuChunshuo <qiuchunshuo@huawei.com>
### What this PR does / why we need it?
This PR makes `AscendMLAMetadataBuilder` and `AscendSFAMetadataBuilder`
properly inherit from the base class `MLACommonMetadataBuilder` in vllm
by adding `super().__init__()` calls.
**Changes:**
- Add `super().__init__()` call in `AscendMLAMetadataBuilder.__init__()`
- Add `super().__init__()` call in `AscendSFAMetadataBuilder.__init__()`
- Extract `ascend_chunked_prefill_workspace_size()` to
`vllm_ascend/attention/utils.py` to avoid code duplication
- Override `determine_chunked_prefill_workspace_size()` to support
Ascend-specific 128k tokens workspace size (vs 64k in parent class)
- Update unit tests to mock parent class `__init__` for proper isolation
**Why we need it:**
- Follow proper Python inheritance patterns by calling
`super().__init__()`
- Reduce code duplication by reusing parent class initialization logic
- Better maintainability as parent class changes will be automatically
inherited
Part of issue #5463 item 10
### Does this PR introduce _any_ user-facing change?
No, this is an internal refactoring that does not change any user-facing
behavior.
Signed-off-by: lico67373 <918688502@qq.com>
### What this PR does / why we need it?
In [PR 5040](https://github.com/vllm-project/vllm-ascend/pull/5040), the
`dispatch_gmm_combine_decode` operator was configured with an incorrect
global_bs parameter. This PR is to fix the bug.
The global_bs provided as input should have the same meaning as in the
`moe_distributed_dispatch` operator, specifically: (the maximum batch
size across all cards) * (expert parallel world size).
However, the implementation incorrectly used the variable
max_num_tokens, which does not account for tensor parallelism. This
error likely resulted in an unnecessarily large (overestimated) value.
More info about this operator, please refer to RFC: issue
https://github.com/vllm-project/vllm-ascend/issues/5476
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Acc
test qwen3-235b eplb on a single A3 node(ep16),
with dispatch_gmm_combine_decode
| dataset | version | metric | mode | vllm-api-stream-chat |
|----- | ----- | ----- | ----- | -----|
| aime2024 | 604a78 | accuracy | gen | 80.00 |
- vLLM version: v0.13.0
- vLLM main:
11b6af5280
Signed-off-by: wangqiankun <wangqiankun13@huawei.com>
### What this PR does / why we need it?
This PR aims to fix setting of `speculative_config.enforce_eager` in
deepseek v3.2 mtp. The point is that, vllm sets
`speculative_config.enforce_eager` as True if using deepseek_v32 with
mtp. Since we support graph mode, we simply ignore it here. However,
this fix will also implicitly ignore user setting of
`speculative_config.enforce_eager`, we need to take care and remove it
once vllm supports this feature.
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
by ci
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
Signed-off-by: Zetong Li <slippersss@126.com>
### What this PR does / why we need it?
adapt to: https://github.com/vllm-project/vllm/pull/30475.
just change get_num_encoder_tokens() to get_num_encoder_embeds() in
recompute_schedule.py, which seems that it is currently not in use. The
get_num_encoder_tokens() function in VLLM no longer exists.
- vLLM version: v0.13.0
- vLLM main:
ad32e3e19c
Signed-off-by: 01267596 <xiongkai123@cmbchina.com>
Co-authored-by: 01267596 <xiongkai123@cmbchina.com>
### What this PR does / why we need it?
In the pcp full graph Qwen model scenario, the inconsistency between the
Q shape and actual q len of the FIA operator is fixed.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
Signed-off-by: weiguihua2 <weiguihua2@huawei.com>
### What this PR does / why we need it?
Resolved a double-free memory vulnerability in the pooling layer under
re-computation scenarios.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
d68209402d
Signed-off-by: fems14 <1804143737@qq.com>
### What this PR does / why we need it?
Cancel the embeddings sharing when the embeddings of main model and the
embeddings of eagle model are different.
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
Cause i don't have `Meta-Llama-3.1-8B-Instruc`t locally, i commented it
and run:
```shell
pytest -s tests/e2e/singlecard/spec_decode/test_v1_spec_decode.py::test_llama_qwen_eagle_acceptance
```
The output is fine:
```text
.
======================================================================================================================== warnings summary =========================================================================================================================
<frozen importlib._bootstrap>:241
<frozen importlib._bootstrap>:241: DeprecationWarning: builtin type SwigPyPacked has no __module__ attribute
<frozen importlib._bootstrap>:241
<frozen importlib._bootstrap>:241: DeprecationWarning: builtin type SwigPyObject has no __module__ attribute
-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
====================================================================================================== 3 passed, 1 skipped, 2 warnings in 196.19s (0:03:16) =======================================================================================================
```
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
Signed-off-by: drslark <slarksblood@qq.com>
### What this PR does / why we need it?
This is a part of
https://github.com/vllm-project/vllm-ascend/issues/4715#issue-3694310762
1. refactor the npugraph_ex config，modified the default configuration of
the static kernel, new default value of static kernel is false
2. support online-infer with static kernel
3. fixed the issue where manually modifying FX graphs caused an abnormal
model return type, and removed the related redundant code.
### Does this PR introduce _any_ user-facing change?
yes，the new config of npugraph_ex is as follow:
```
additional_config={
"npugraph_ex_config": {
"enable": True,
"enable_static_kernel": False
}
}
```
### How was this patch tested?
```
vllm serve /data/DeepSeek-V3.1-Terminus-w4a8 \
--host 0.0.0.0 \
--port 8004 \
--data-parallel-size 4 \
--tensor-parallel-size 4 \
--quantization ascend \
--seed 1024 \
--served-model-name deepseek_v3 \
--enable-expert-parallel \
--max-num-seqs 48 \
--max-model-len 40000 \
--async-scheduling \
--max-num-batched-tokens 9000 \
--trust-remote-code \
--no-enable-prefix-caching \
--speculative-config '{"num_speculative_tokens": 3, "method":"deepseek_mtp","disable_padded_drafter_batch": false}' \
--gpu-memory-utilization 0.9 \
--compilation-config '{"cudagraph_capture_sizes":[4,32,64,112,160,176,192], "cudagraph_mode": "FULL_DECODE_ONLY"}' \
--additional-config \
'{"enable_shared_expert_dp": true,"multistream_overlap_shared_expert": true,"npugraph_ex_config":{"enable":true}}'
```
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
---------
Signed-off-by: chencangtao <chencangtao@huawei.com>
Signed-off-by: ChenCangtao <50493711+ChenCangtao@users.noreply.github.com>
Co-authored-by: chencangtao <chencangtao@huawei.com>
### What this PR does / why we need it?
Optimized operator performance and add ut test
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
test in qwen2.5 7b vl, ops time approved 90%
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
this pr is for
# https://github.com/vllm-project/vllm-ascend/issues/5208
Signed-off-by: shiyuan680 <917935075@qq.com>
### What this PR does / why we need it?
Fixed the issue where the PCP and MTP services could not be started due
to asynchronous scheduling.
After the pcp, mtp, and asynchronous scheduling functions are enabled,
the service is suspended because of a shape mismatch after a curl
request is sent. This PR resolves this issue.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
---------
Signed-off-by: weiguihua2 <weiguihua2@huawei.com>
### What this PR does / why we need it?
Add layernormFn triton op for qwen3Next model for better performance.
<img width="248" height="526" alt="image"
src="https://github.com/user-attachments/assets/27b47157-5df5-4db1-aa88-1dae799b2bf6"
/>
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
---------
Signed-off-by: SunnyLee219 <3294305115@qq.com>
### What this PR does / why we need it?
4/N EAGLE refactor plan devided into many parts, this PR is the first
change, which modifies the attn_metadata update method by modifying
common_metadata and then rebuilding the code.
### Does this PR introduce _any_ user-facing change?
ut
### How was this patch tested?
no
- vLLM version: v0.13.0
- vLLM main:
bde38c11df
---------
Signed-off-by: lilinsiman <lilinsiman@gmail.com>
Signed-off-by: Zetong Li <slippersss@126.com>
Co-authored-by: Zetong Li <slippersss@126.com>
### What this PR does / why we need it?
Set a additional config parameter to control whether the gmmswigluequant
fuseion operator is enabled; it is enabled by True. / When enabled with
a small number of GPUs, the gmmswigluquant fused operator can cause some
performance degradation.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
#### Perf
test model: GLM 4.6(w8a8)
- single A3 node(ep16, tp16), async-scheduling, mtp, FULL_DECODE_ONLY
- bs=1, input_lens=32000, ouput_lens=1024
Without this PR: TPOT 32.22.ms
With this PR: TPOT 30.23ms
---------
Signed-off-by: zjks98 <zhangjiakang4@huawei.com>
Co-authored-by: zjks98 <zhangjiakang4@huawei.com>
### What this PR does / why we need it?
When enable VLLM_ASCEND_FLASHCOMM2_PARALLEL_SIZE>1, we need index
operation to reorganize the batch, because that we need ensure the
correct batch-id for each rank after the reduce-scatter op in
VLLM_ASCEND_FLASHCOMM2_PARALLEL_SIZE>1. But we do not need it when
VLLM_ASCEND_FLASHCOMM2_PARALLEL_SIZE=1, which dose not need
reduce-scatter.
Signed-off-by: Levi-JQ <yujinqi2@huawei.com>
Co-authored-by: Levi-JQ <yujinqi2@huawei.com>
### What this PR does / why we need it?
In PD disaggregation case, when P has multi nodes, mooncake fails to
send data. Fix the issue in this PR.
The details:
If a P rank does not need to transfer kv cache to any one D rank, D node
should send a message to P node to release the kv
cache in P node. If P has multi nodes, D node should know the
corresponding IP in each P node, then D node can send message to the
right P node. Otherwise, send data error will happen. This PR fix this
issue by providing P nodes IP to D node through Parameter
`remote_port_send_num`.
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
---------
Signed-off-by: wangxiaochao <w00642655@china.huawei.com>
Co-authored-by: wangxiaochao <w00642655@china.huawei.com>
### What this PR does / why we need it?
[Feature] Adapt DispathGmmCombineDecode opertor to align with weight
scale dtype of small operators.
- **Before**: weight scale must be float32
- **After**: weight scale can be float32/float16 when x is float16,
float32/bfloat16 when x is float32/bfloat16. And w1 scale can use
different dtype with w2 scale.
More info about this operator, please refer to RFC: issue
https://github.com/vllm-project/vllm-ascend/issues/5476
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
#### Perf
> When scale is of type fp16 or bf16, it will be cast to fp32 internally
within the operator, while the subsequent computations remain unchanged.
Therefore, this PR will introduce an additional cast operation but halve
the memory copy operations for scale . Furthermore, since the scale data
is only a few KB in size and participates in relatively few
computations, its impact is almost negligible compared to major
operations like matrix multiplication. Thus, the theoretical performance
change should be minimal.
test single operator cases from qwen3-235b,
- single A3 node(ep16), 64 moe experts, 4 experts / die (like qwen3-235b
ep32)
- batch=18/32, token_hidden_size 4096, moe_intermediate_size 1536
The test was conducted for 100 rounds, and the average of the last 95
rounds was taken.
| | bs18(us)| bs32(us)|
| -----| -----| -----|
|Without this PR|96.28|108.83|
|With this PR|96.06|107.90|
Note: Single-operator benchmarks represent an ideal scenario. They are
usually only useful for referencing relative changes and may not fully
align with performance data observed within the full model.
#### Acc
test qwen3-235b eplb on a single A3 node(ep16),
with dispatch_gmm_combine_decode
| dataset | version | metric | mode | vllm-api-stream-chat |
|----- | ----- | ----- | ----- | -----|
| aime2024 | 604a78 | accuracy | gen | 83.33 |
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
Signed-off-by: wangqiankun <wangqiankun13@huawei.com>
### What this PR does / why we need it?
This PR refactors `get_kv_cache_spec` method to delegate AttentionSpec
creation to each attention module's own `get_kv_cache_spec()` method,
aligning with the vllm source code structure.
**Changes:**
- Simplify `get_kv_cache_spec` in `model_runner_v1.py` and
`cpu_offload_connector.py`
- Remove manual `AttentionType` checks for `Attention` modules
- Delegate spec creation to each attention module's `get_kv_cache_spec`
method directly
- Let `MambaBase` layers use their own `get_kv_cache_spec` method
- Keep `use_sparse` hack for `MLAAttention` (DeepSeek DSA mode) as
Ascend-specific handling
This change follows RFC #5463 item 12: move AttentionSpec to Attention
module.
- Fixes#5463 (item 12)
### Does this PR introduce _any_ user-facing change?
No. This is an internal refactoring that simplifies code structure
without changing any external behavior.
### How was this patch tested?
- Syntax validation passed via `python -m py_compile`
- CI tests will verify the changes work correctly with existing test
cases
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
Signed-off-by: lico67373 <918688502@qq.com>
### What this PR does / why we need it?
1. Rename dynamic_ep to default_eplb.
2. Rename dynamic_ep_v2 to swift_balancer
3. Discard func compose_expert_update_info_bipartite.
- vLLM version: v0.13.0
- vLLM main:
bde38c11df
Signed-off-by: shenchuxiaofugui <1311027364@qq.com>
This PR add `MatmulAllreduceRmsnorm` operator and introduces a graph
fusion pass for `matmul_allreduce_rmsnorm` operations. The
implementation includes a new configuration flag, a pattern matching
pass using `torch._inductor.pattern_matcher`.
Co-authored-by: Trunrain [270250579@qq.com](mailto:270250579@qq.com)
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c
---------
Signed-off-by: wxsIcey <1790571317@qq.com>
Signed-off-by: tongrunze <t00574058@china.huawei.com>
### What this PR does / why we need it?
Migrate the torch profiler configuration from deprecated environment
variables (`VLLM_TORCH_PROFILER_DIR`, `VLLM_TORCH_PROFILER_WITH_STACK`,
`VLLM_TORCH_PROFILER_WITH_PROFILE_MEMORY`) to the explicit
`ProfilerConfig` object, aligning with vLLM's configuration best
practices.
The profiler environment variable approach is deprecated in vLLM and
will be removed in v0.14.0 or v1.0.0.
### Does this PR introduce _any_ user-facing change?
yes, for deverlopers who want to fetch profiler, he should use `--profiler-config` instead of `VLLM_TORCH_PROFILER_DIR`
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
11b6af5280
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
### What this PR does / why we need it?
1. If the model has dense layers, the current code will attempt to
obtain the routing experts of the dense layers, which will cause an
error. This should be fixed by modifying the code to skip the dense
layers when obtaining the routing experts.
2. The global_expert_map that the function directly outputs a affects
the performance of dsv3.2.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
DeepSeek V3.1 conversation is normal.
#### aime precision test (dsv3.1)
baseline without eplb
| dataset | version | metric | mode | vllm-api-general-chat |
|----- | ----- | ----- | ----- | -----|
| aime2024 | 604a78 | accuracy | gen | 66.67 |
eplb
| dataset | version | metric | mode | vllm-api-general-chat |
|----- | ----- | ----- | ----- | -----|
| aime2024 | 604a78 | accuracy | gen | 70.00 |
- vLLM version: v0.13.0
- vLLM main:
11b6af5280
Signed-off-by: shenchuxiaofugui <1311027364@qq.com>
### What this PR does / why we need it?
As mentioned in https://github.com/vllm-project/vllm-ascend/issues/5339,
multi-modal inference on vllm-ascend may lead to OOM issues in some
scenarios.
After our analysis, this is due to the memory fragmentation caused by
frequent dynamic memory size adjustments during runtime. During the
inference, the figure for non-torch memory see a gradual increase from
around 1G to over 5G until the OOM issue occurs.
We find that this problem can be resolved by just directly setting
`PYTORCH_NPU_ALLOC_CONF=expandable_segments:True`. Find more details at
https://docs.vllm.ai/projects/ascend/en/latest/faqs.html#how-to-handle-the-out-of-memory-issue.
Thus, we decide to set this value by default, except RL (sleep mode)
scenarios.
It's also worthy to note that this environment variable may have more
than one key-value pairs. We should append `",expandable_segments:True"`
to the current configs.
For example:
```python
PYTORCH_NPU_ALLOC_CONF = "page_size:1g" + ",expandable_segments:True".
```
> [!NOTE]
> `max_split_size_mb` or `garbage_collection_threshold` cannot be
enabled together with `expandable_segments=True`.
### Does this PR introduce _any_ user-facing change?
Users do not need to set
`PYTORCH_NPU_ALLOC_CONF=expandable_segments:True` manually any more.
### How was this patch tested?
I have build a dataset consisting of my own photographs, which can
stably reproduce this OOM issue on Qwen3-VL serie models.
After apply this PR, this problem has been resolved and the amount of
non-torch memory will keep stable at around 1G throughout the whole
inference.
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
---------
Signed-off-by: shen-shanshan <467638484@qq.com>
### What this PR does / why we need it?
The force-free secondary release request causes the node to crash. When
requests are pulled too quickly, they should not be added to the
delay-free queue.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
By ci
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
Signed-off-by: wangxiaoteng <wangxiaoteng@huawei.com>
### What this PR does / why we need it?
As issue #5948 reported，when using cpu_offload_connector with TP=1, the
server will hang on starting, we found several bugs here to fix.
1. some crash error encountered because of code changed with vllm
version updating, some of them can be fixed as #5948, and this PR fixed
all of them.
2. hang problem described in #5948, the direct reason is that in
cpu_offload_connector, RPC client using the same client id in scheduler
and worker when tensor_parrallel_size is 1, this PR force the client id
to be different, then it is fixed.
- Why we didn't find this hang problem before?
Because we using --distributed-executor-backend mp or
tensor_parrallel_size > 1 in our test, in our old test case, the
scheduler and workers are different procceses, then client ids build by
`worker-{os.getpid()}` are not the same. But when using
tensor_parrallel_size=1, vllm will use uniproc as
distributed-executor-backend by default, the scheduler and worker will
by in the same proccess, then client ids are the same and hang.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
Signed-off-by: lidenghui <lidenghui1110@gmail.com>
### What this PR does / why we need it?
Add basic 310p support. Only dense models work with eager mode now.
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
---------
Signed-off-by: Tflowers-0129 <2906339855@qq.com>
Signed-off-by: Shaoxu Cheng <2906339855@qq.com>
### What this PR does / why we need it?
reverts #5761 to fix accuracy issues when using piecewise graph mode.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
Signed-off-by: Angazenn <supperccell@163.com>
### What this PR does / why we need it?
Convert `vllm-ascend/compilation` to ruff format.
### Does this PR introduce _any_ user-facing change?
During this migration, we encountered some **errors** in our CI and
testing environments, such as:
```
vllm_ascend/utils.py:653: in <module>
def register_ascend_customop(vllm_config: VllmConfig | None = None):
^^^^^^^^^^^^^^^^^
E TypeError: unsupported operand type(s) for |: 'NoneType' and 'NoneType'
```
**1. Root Cause Analysis:**
The project uses a common pattern to break circular dependencies:
```python
if TYPE_CHECKING:
from vllm.config import VllmConfig
else:
VllmConfig = None # Placeholder assigned at runtime
```
When Python parses the function definition `def
register_ascend_customop(vllm_config: VllmConfig | None)`, it attempts
to evaluate the expression `VllmConfig | None`.
Since `VllmConfig` is assigned `None` at runtime, the expression
effectively becomes `None | None`. In Python, `None` is an instance of
`NoneType`. While the `|` operator is implemented for Type objects
(classes), it is not supported for `NoneType` instances, leading to the
`TypeError` shown above.
**2. Solution:**
To maintain the modern `|` syntax required by our new linting standards
while preserving our dependency management strategy, I have introduced:
```python
from __future__ import annotations
```
at the top of the affected files. This enables **Postponed Evaluation of
Annotations (PEP 563)**.
**3. Impact and Benefits:**
- By enabling `annotations`, Python no longer executes the `VllmConfig |
None` operation during module load. Instead, it stores the annotation as
a string literal, completely avoiding the `None | None` calculation.
- We can keep the `VllmConfig = None` placeholders. This ensures that
other modules can still import these symbols without triggering an
`ImportError`, maintaining a stable dependency graph.
- IDEs and static type checkers (MyPy/Pyright) continue to resolve the
types correctly. This allows us to use modern syntax without sacrificing
type safety or runtime stability.
- The only side effect is that `__annotations__` will now return strings
instead of type objects. Since this module does not use runtime type
enforcement or reflection, this change has zero negative impact on
existing functionality.
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
11b6af5280
---------
Signed-off-by: MrZ20 <2609716663@qq.com>
### What this PR does / why we need it?
This PR fix the input constraints checks for the mlapo and bmm_transpose
operators.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
CI passed with new added/existing test.
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
### Perf
64K/3K，1P1D，bs=32
before this pr:
TPOT 29ms, TTFT 47s，TPS 606 token/s
after this pr:
TPOT 29ms, TTFT 48s，TPS 636 token/s
Signed-off-by: rjg-lyh <1318825571@qq.com>
### What this PR does / why we need it?
1. Fix DeepSeek-V3.2-W8A8-Pruning mtp
2. Add DeepSeek-V3.2-W8A8-Pruning e2e test
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
11b6af5280
Signed-off-by: hfadzxy <starmoon_zhang@163.com>
