<!-- Thanks for sending a pull request!
BEFORE SUBMITTING, PLEASE READ
https://docs.vllm.ai/en/latest/contributing/overview.html
-->
### What this PR does / why we need it?
This PR is used for resolved [issue
1147](https://github.com/vllm-project/vllm-ascend/issues/1147)
1. Move fused_moe code into one file `fused_moe.py`.
2. Integrate branch conditions into function `get_fused_moe_state`.
<!--
- Please clarify what changes you are proposing. The purpose of this
section is to outline the changes and how this PR fixes the issue.
If possible, please consider writing useful notes for better and faster
reviews in your PR.
- Please clarify why the changes are needed. For instance, the use case
and bug description.
- Fixes #
-->
### Does this PR introduce _any_ user-facing change?
1. This PR has removed the env `VLLM_ENABLE_MC2`, because I think this
env is useless, we can make judgments based on the current scenario
without this env, it will only increase complexity.
2. This PR has removed the env `USING_LCCL_COM`, because this env has
already expired.
3. `additional_config.expert_tensor_parallel_size` has already expired,
and now we also use parameter `enable_expert_parallel`, consistent with
the vLLM.
<!--
Note that it means *any* user-facing change including all aspects such
as API, interface or other behavior changes.
Documentation-only updates are not considered user-facing changes.
-->
### How was this patch tested?
<!--
CI passed with new added/existing test.
If it was tested in a way different from regular unit tests, please
clarify how you tested step by step, ideally copy and paste-able, so
that other reviewers can test and check, and descendants can verify in
the future.
If tests were not added, please describe why they were not added and/or
why it was difficult to add.
-->
Signed-off-by: zzzzwwjj <1183291235@qq.com>
### What this PR does / why we need it?
Move all vector operations to a secondary stream, with the expected
overlaping being:
```
| q_rmsnorm | | kv_norm_rope_cache | | q_rope |
| matmul W_DQ | matmul W_DKV | index | index | matmul W_UQ | split | matmul W_KV_T |
```
Currently, the `IndexByTensor` operators introduced by computation of
`cos` and `sin` can't be offloaded to the secondary stream due to a
known bug of graph fusion optimization pass. So we instead keep it in
the main stream, only requires it be computed before `matmul W_UQ` to
avoid hindering later overlapping. The problem may be solved by later
optimization (#993), which hoists the computation of `cos` and `sin` up
to the first layer.
### Does this PR introduce _any_ user-facing change?
Controlled by `torchair_graph_config.enable_multistream_mla`, defaulted
to False.
### How was this patch tested?
Tested on 1x16 910 node, with tailored 2 layer DSKv2.
Signed-off-by: sdmyzlp <lrwei2@petalmail.com>
Contains on #1111 for completeness.
<!-- Thanks for sending a pull request!
BEFORE SUBMITTING, PLEASE READ
https://docs.vllm.ai/en/latest/contributing/overview.html
-->
### What this PR does / why we need it?
Implement multi-stream parallelism for MoE layers with shared experts,
where computation of shared experts will be overlapped with expert token
dispatch and combine. Also, when multi-stream is enabled, weights of
shared experts will be force to replicate across all cards, regardless
of any tensor parallelism configurations, to avoid AllReduce operations.
With the expected overlaping being:
```
| shared gate_up | shared act | | shared down |
| dispatch | routed gate_up, act, down | combine |
```
<!--
- Please clarify what changes you are proposing. The purpose of this
section is to outline the changes and how this PR fixes the issue.
If possible, please consider writing useful notes for better and faster
reviews in your PR.
- Please clarify why the changes are needed. For instance, the use case
and bug description.
- Fixes #
-->
### Does this PR introduce _any_ user-facing change?
No.
<!--
Note that it means *any* user-facing change including all aspects such
as API, interface or other behavior changes.
Documentation-only updates are not considered user-facing changes.
-->
### How was this patch tested?
Tested on 1x16 910 node, with tailored 2 layer DSKv2.
<!--
CI passed with new added/existing test.
If it was tested in a way different from regular unit tests, please
clarify how you tested step by step, ideally copy and paste-able, so
that other reviewers can test and check, and descendants can verify in
the future.
If tests were not added, please describe why they were not added and/or
why it was difficult to add.
-->
---------
Signed-off-by: sdmyzlp <lrwei2@petalmail.com>
Add unpadded Qwen2.5-VL for verl scenario.
When using vllm-ascend for verl scenario, set `USE_OPTIMIZED_QWEN2_5_VL`
(default `1`) to `0` to use unpadded Qwen2.5-VL to avoid errors.
This is cherry-picked from 0.7.3-dev
Signed-off-by: shen-shanshan <467638484@qq.com>
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
Co-authored-by: Shanshan Shen <467638484@qq.com>
### What this PR does / why we need it?
Based on the design of dual-batch overlap proposed by Deepseek team and
also the implementation of fused moe in VLLM project, we implement the
multi-stream(also known as dual-batch) overlap for deepseek+mla on
Ascend NPU. We split the input batch of model into two microbatches and
then overlap the comp/comm ops in attention and moe layers using two
streams to improve the performance. Our approach can be easily extended
when adding dispatch/combine communications for moe layer.
Compared with the previously proposed
[draft](https://github.com/vllm-project/vllm-ascend/pull/842), we use
one stream for computation ops and the other for communication ops,
separately. In out opinions, it is beneficial for arranging the order of
executing different ops and thus avoiding the contention of
computation/communication resources.
ref: [overlap for
llama](https://github.com/vllm-project/vllm/pull/15787/files)
ref: [dbo in
sglang](https://github.com/sgl-project/sglang/pull/4068/files#diff-b4937569fc71f6ad215181b633b2f89c7183a2b4ac39e41fc22635599a9be7de)
### Does this PR introduce _any_ user-facing change?
Adding an env variable "VLLM_ENABLE_DBO". Users can enable dbo by
setting "VLLM_ASCEND_ENABLE_DBO=1"
See /examples/offline_dualbatch_overlap_npu.py for more info.
### How was this patch tested?
This patch can be tested with vllm-0.9.0 using its online service with
benchmark tests. We have decoupled the func of dbo from vllm and it
should be able to run without any modification to the code of vllm(some
modifications is better to implement in vllm though).
Any advice/discussion is welcome.
### Performance Benchmark
We have ran the benchmark_serving script of vllm to test the performance
after using dual-batch overlap.
`python -m vllm.entrypoints.openai.api_server \
--model=DeepSeek-R1-W8A8 \
--trust-remote-code \
--distributed-executor-backend=mp \
-tp=16 \
--port 8006 \
--max-num-seqs 390 \
--max-model-len 32768 \
--max-num-batched-tokens 65536 \
--block-size 128 \
--compilation_config 0 \
--gpu-memory-utilization 0.90 \
--disable-log-requests \
--additional-config
'{"expert_tensor_parallel_size":1,"enable_inter_dp_scheduling":true,"init_torchair_graph_batch_sizes":true,"trace_recompiles":true,"ascend_scheduler_config":{},"enable_graph_mode":false}'`
and run benchmark with the parameters of :
`--dataset-name random --random-input-len 4096 --random-output-len 1
--num-prompts 200 --max-concurrency 8 --request-rate 5
--metric-percentiles 90`
1. test with the version using allgather+allreduce in Ascend 910B (tp16
ep16 + deepseek r1 w8a8)
2. test with the version using alltoall:
prefill qps: 0.90 -> 1.01
Mean TTFT：8226->7432ms
The overlap approach when using alltoall communication can be further
optimized by overlapping micro-batch1's moe comp with micro-batch2's
dispatch a2a comm
---------
Signed-off-by: zhuohuan <zxdu1997@gmail.com>
### What this PR does / why we need it?
Support MOE inner Multi-stream for Deepseek.
This feature requires graph mode with mc2 enabled.
---------
Signed-off-by: David9857 <985700846@qq.com>
### What this PR does / why we need it?
Optimize the performance of calculation logic in sampler and deepseekv2.
### Does this PR introduce _any_ user-facing change?
Added VLLM_ENABLE_TOPK_OPTIMZE config in sampler
### How was this patch tested?
pytest test_sampler.py
Signed-off-by: wangxiaoxin (A) <wangxiaoxin7@huawei.com>
Co-authored-by: wangxiaoxin (A) <wangxiaoxin7@huawei.com>
Co-authored-by: ZhengWG <zwg0606@gmail.com>
More and more config options are added to additional_config. This PR
provide a new AscendConfig to manage these config options by an easier
way to make code cleaner and readable.
This PR also added the `additional_config` doc for users.
Added the test_ascend_config.py to make sure the new AscendConfig works
as expect.
TODO: Add e2e test with torchair and deepseek once the CI resource is
available.
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
### What this PR does / why we need it?
the interface of qwen2.5vl changes from column linear to qkv linear,
this makes our weight pad func become abnormal, thus we optimize
split_qkv func to fix this bug.
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
with CI
Signed-off-by: zouyida2052 <zouyida2002@gmail.com>
### What this PR does / why we need it?
1. In previous PRs https://github.com/vllm-project/vllm-ascend/pull/580https://github.com/vllm-project/vllm-ascend/pull/784, I saved GPU memory
by promptly deleting unnecessary tensors. For tensors passed from
upper-layer functions, I used a list container to transfer the parameter
and then popped the tensor from the list within the inner function to
achieve deletion. Recently, I discovered a better implementation in
sglang—the `dispose_tensor` function and I recommend adopting this
approach.
2. Dispose `hidden_states` and `residual` from the previous layer once
they're no longer used.
3. Avoid to generate `self.inputs_embeds` in `ModelRunnerV1` in
non-multimodal scenarios.
With the aforementioned optimizations, using the DeepSeek-R1-W8A8 model
under the conditions of `TP=16` and `max-model-len=32768`, we can save
1.3GB of npu memory.
**Reference**: https://github.com/sgl-project/sglang/pull/6147
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
---------
Signed-off-by: ApsarasX <apsarax@outlook.com>
Tweak packed_modules_mapping to support W8A8 weights.
<!-- Thanks for sending a pull request!
BEFORE SUBMITTING, PLEASE READ
https://docs.vllm.ai/en/latest/contributing/overview.html
-->
### What this PR does / why we need it?
<!--
- Please clarify what changes you are proposing. The purpose of this
section is to outline the changes and how this PR fixes the issue.
If possible, please consider writing useful notes for better and faster
reviews in your PR.
- Please clarify why the changes are needed. For instance, the use case
and bug description.
- Fixes #
-->
### Does this PR introduce _any_ user-facing change?
<!--
Note that it means *any* user-facing change including all aspects such
as API, interface or other behavior changes.
Documentation-only updates are not considered user-facing changes.
-->
### How was this patch tested?
<!--
CI passed with new added/existing test.
If it was tested in a way different from regular unit tests, please
clarify how you tested step by step, ideally copy and paste-able, so
that other reviewers can test and check, and descendants can verify in
the future.
If tests were not added, please describe why they were not added and/or
why it was difficult to add.
-->
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
The [vllm
commit](67da5720d4)
changed the input and rotary position embedding for qwen 2.5 vl which
break CI. This PR fix the CI failure for qwen2.5 vl in quick
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
<!-- Thanks for sending a pull request!
BEFORE SUBMITTING, PLEASE READ
https://docs.vllm.ai/en/latest/contributing/overview.html
-->
### What this PR does / why we need it?
1. This PR introduces native `all_to_all` communication operator to fix
`allgather` bugs when dp_size > 1. Besides, it adds a naive
implementation of force-load-balance when doing profile runs.
2. The operator `npu_dequant_swiglu_quant` only supports input
hidden_states with dtype `torch.int32`. This tensor occupies space of
`global_bs * seq_len * topk * hidden_size`, which might be very large as
`ep_size` grows. Therefore we need to disable this operator and use
original `swiglu` && `quantize`.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
By performing offline inference:
---------
Signed-off-by: angazenn <zengyanjia@huawei.com>
Co-authored-by: angazenn <zengyanjia@huawei.com>
### What this PR does / why we need it?
Fix the method of importing environment variables in DeepSeek model to
support successful compilation via aclgraph.
Signed-off-by: rjg-lyh <1318825571@qq.com>
### What this PR does / why we need it?
In the w8a8 quantization code of `fused_experts`, the output of almost
every operator is assigned a new variable name. If we want to save NPU
memory, we manually `del` these variables to end their lifecycle, which
fills the code with `del` statements and looks inelegant.
Therefore, I plan to names the output of most operators as
`hidden_states`, thereby ending the lifecycle of the previous
`hidden_states`.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Signed-off-by: ApsarasX <apsarax@outlook.com>
-->
### What this PR does / why we need it?
<!--
- Please clarify what changes you are proposing. The purpose of this
section is to outline the changes and how this PR fixes the issue.
If possible, please consider writing useful notes for better and faster
reviews in your PR.
- Please clarify why the changes are needed. For instance, the use case
and bug description.
- Fixes #
-->
1. Improve inference speed and usability for deepsek models with NPU
graph mode.
2. Modify some codes to adapt to CANN 8.1.RC1.beta1.
3. Add a switch for NPU graph mode and its cache.
### Does this PR introduce _any_ user-facing change?
<!--
Note that it means *any* user-facing change including all aspects such
as API, interface or other behavior changes.
Documentation-only updates are not considered user-facing changes.
-->
This PR provides an experimental configuration to enable NPU graph mode
for Deepseek models. User can set
additional_config={'enable_graph_mode': True} to try this feature. Note
that this feature currently only supports for V0 engine.
### How was this patch tested?
<!--
CI passed with new added/existing test.
If it was tested in a way different from regular unit tests, please
clarify how you tested step by step, ideally copy and paste-able, so
that other reviewers can test and check, and descendants can verify in
the future.
If tests were not added, please describe why they were not added and/or
why it was difficult to add.
-->
This patch was tested with the newest torch_npu 2.5.1
(https://pypi.org/project/torch-npu/#files) and CANN 8.1.RC1.beta1
toolkit&nnal&kernels
(https://www.hiascend.com/developer/download/community/result?module=cann)
released in 25/30 April.
Signed-off-by: linfeng-yuan <1102311262@qq.com>
### What this PR does / why we need it?
Optimize qwen2_vl and qwen2_5_vl.
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
Testing this PR on 1080p picture with tp=1, bs=1 on Qwen2-VL and
Qwen2.5-VL, every fa op's during time lasting from 11ms to 9ms, got
roughly 22% perf boost.
---------
Signed-off-by: zouyida2052 <zouyida@huawei.com>
Signed-off-by: zouyida2052 <zouyida2002@gmail.com>
Co-authored-by: zouyida2052 <zouyida@huawei.com>
### What this PR does / why we need it?
Fix#674 to avoild KVCache overallocation and OOM risks.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Signed-off-by: ApsarasX <apsarax@outlook.com>
<!-- Thanks for sending a pull request!
BEFORE SUBMITTING, PLEASE READ
https://docs.vllm.ai/en/latest/contributing/overview.html
-->
### What this PR does / why we need it?
As custom deepseek modeling do some changes to support graph mode in
https://github.com/vllm-project/vllm-ascend/pull/585, so i follow it to
change custom deepseek_mtp modeling.
And some modifications for k>1 were not carried over by the
https://github.com/vllm-project/vllm-ascend/pull/429, now i add it.
In order to better take care of the MTP feature in the vllm-ascend
repository, I added cases related to graph mode(torchair), but i skip it
since torchair can not correctly clean up memory in vllmrunner.
Also i add some case for MTP quantization weights, but test weight is
not ready, so i skip it and i will open it when test quant weights is
ready.
https://github.com/vllm-project/vllm-ascend/pull/648 did not completely
fix the sample
change(https://github.com/vllm-project/vllm-ascend/issues/660) issue, I
added the relevant changes.
### Does this PR introduce _any_ user-facing change?
now, u can use following method to use mtp in deepseek v3/r1 float or
quant weights with eager mode.
```python
llm = LLM(
model="wemaster/deepseek_mtp_main_random_bf16",
tensor_parallel_size=2,
speculative_config={
"num_speculative_tokens": 1,
},
enforce_eager=True,
trust_remote_code=True,
disable_log_stats=False,
gpu_memory_utilization=0.8,
max_model_len=64,
)
```
or use mtp in deepseek v3/r1 float or quant weights with graph
mode（torchair）
```python
llm = LLM(
model="wemaster/deepseek_mtp_main_random_bf16",
tensor_parallel_size=2,
speculative_config={
"num_speculative_tokens": 1,
},
trust_remote_code=True,
additional_config={
'enable_graph_mode': True,
},
disable_log_stats=False,
gpu_memory_utilization=0.8,
max_model_len=64,
)
```
add notes:
1. now, we support k>1, so u can set num_speculative_tokens > 1 if there
is sufficient redundant computing power;
2. MTP is not supported in V1, we will support it when vLLM does it in
https://github.com/vllm-project/vllm/issues/13500.
3. if u run MTP failed by `segmentation fault`, u can follow v0.7.3
patch https://github.com/vllm-project/vllm-ascend/pull/236 file
`vllm_ascend/patch/patch_metrics.py` method
`__npu_async_metrics_collector_init__`
### How was this patch tested?
local tested passed and test by CI
Signed-off-by: mengwei805 <mengwei25@huawei.com>
### What this PR does / why we need it?
The torch-npu 2.5.1 are published:
https://pypi.org/project/torch-npu/2.5.1/
It's time to remove all torch-npu dev version from vllm-ascend code base
### Does this PR introduce _any_ user-facing change?
Yes, using torch-npu 2.5.1
### How was this patch tested?
- [ ] CI passed
- [ ] Manually test
- [ ] Grep all `dev2025`
---------
Signed-off-by: Yikun Jiang <yikunkero@gmail.com>
### What this PR does / why we need it?
1. support deepseek with w8a8 quant;
2. support deepseek with mix-parallel(multi-DP, EP+TP);
3. support deepseek with graphmode.
---------
Signed-off-by: wen-jie666 <wenjie39@huawei.com>
Signed-off-by: Yizhou Liu <liuyizhou5@h-partners.com>
Signed-off-by: libaokui <libaokui@huawei.com>
Signed-off-by: linfeng-yuan <1102311262@qq.com>
Co-authored-by: wen-jie666 <wenjie39@huawei.com>
### What this PR does / why we need it?
Add notes for deepseek's patch and remove some of the unnecessary
comments
---------
Signed-off-by: ganyi <pleaplusone.gy@gmail.com>
### What this PR does / why we need it?
The pr will fix some bug about spec decode / MTP
The pr add a mtp e2e UT `test_mtp_correctness.py`
**vllm_ascend/attention/attention.py**
1. add support `self.attn_mask_cache` only has 1 element to cover scene
in which both spec docode and chunked prefill are enabled.
**vllm_ascend/distributed/parallel_state.py**
1. remove 2 assert because spec decode worker would use init_worker
twice
**vllm_ascend/models/deepseek_mtp.py**
1. remove unused params;
2. add support w8a8 in `CustomDeepSeekMTP`
**vllm_ascend/quantization/quant_config.py**
1. use `AscendUnquantizedFusedMoEMethod` instead of
`UnquantizedFusedMoEMethod`
**other**
1. replace `from vllm.logger import init_logger` to `from vllm.logger
import logger` all of the vllm-ascend project
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
Signed-off-by: mengwei805 <mengwei25@huawei.com>
### What this PR does / why we need it?
Fix api in DeepSeekV2, aligning with the latest code of the main branch
in vllm.
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
Test locally with deepseek-v2-lite, and will add CI by @Potabk.
Plz update the model UT after this pr is merged, thx! cc @Potabk
Signed-off-by: MengqingCao <cmq0113@163.com>
### What this PR does / why we need it?
We proposed the FastPatch method, which optimized patch embedding
(Conv3D) for Qwen2VL.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
We've tested it on benchmark, it meets our satisfaction and is better
than original patch_embed layer.
---------
Signed-off-by: baifanxxx <baifanxxx@gmail.com>
Signed-off-by: zouyida <zouyida@huawei.com>
Co-authored-by: zouyida <zouyida@huawei.com>
### What this PR does / why we need it?
this pr fixes the error while inferring Qwen2_VL.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
We've tested it on benchmark, it meets our satisfaction and is equal to
gpu.
---------
Signed-off-by: zouyida <zouyida@huawei.com>
