### What this PR does / why we need it?
torch_npu.npu_grouped_matmul:
https://www.hiascend.com/document/detail/zh/Pytorch/710/apiref/torchnpuCustomsapi/context/torch_npu-npu_grouped_matmul.md
According to the document, when `split_item` is 2 or 3,
`torch_npu.npu_grouped_matmul` will return a list which has one element.
Therefore, the `torch.cat` after `torch_npu.npu_grouped_matmul` is
unnecessary.
### Does this PR introduce _any_ user-facing change?
not involved
### How was this patch tested?
ut and e2e covered: `tests/ut/ops/test_fused_ops.py`,
`tests/e2e/singlecard/ops/test_fused_moe.py`
**performance**:
(qwen3 30B, 2k->20k)
base:
Total Token throughput (tok/s): 667.76
remove cat:
Total Token throughput (tok/s): 680.82
- vLLM version: v0.10.0
- vLLM main:
fa00c5d75b
Signed-off-by: huangxialu <huangxialu1@huawei.com>
### What this PR does / why we need it?
This PR is to add e2e test for using npu_mm_all_reduce_base fusion
kernel.
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
not involved
- vLLM version: v0.10.0
- vLLM main:
5d5d419ca6
Signed-off-by: Ronald1995 <ronaldautomobile@163.com>
### What this PR does / why we need it?
Qwen3 MoE supports SP. In scenarios like AlltoAll, AlltoAllv, and MC2,
replacing AllReduce with Reduce-Scatter and AllGather achieves
computational benefits in norm operations while saving one AllGather
communication. This feature is enabled during the P-phase and delivers
notable gains in long-sequence scenarios (e.g., 16k–25k), with
performance improvements reaching 5%–10%.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
```
compilation_config={
"pass_config":{
"enable_sequence_parallelism": True
}
},
enable_expert_parallel=True,
```
- vLLM version: v0.10.0
- vLLM main:
9edd1db02b
---------
Signed-off-by: libaokui <libaokui@huawei.com>
Co-authored-by: libaokui <libaokui@huawei.com>
### What this PR does / why we need it?
Fix `ascend_env has no attr VLLM_ASCEND_ENABLE_CHUNK_MC2`, remove
useless lines
- vLLM version: v0.10.0
- vLLM main:
9edd1db02b
---------
Signed-off-by: wangli <wangli858794774@gmail.com>
### What this PR does / why we need it?
Support MTP with：
- [x] V0 Scheduler
- [x] TorchAir
- [x] Single DP
- [x] Multi DP
- [x] Disaggregate PD
Known issues：
- [ ] Not support V1 Scheduler (chunked prefill), will be supported in a
few weeks
- [ ] vllm v0.10.0 does not support metrics with `DP > 1` right now,
need to comment out the line 171-175 in file
`vllm/vllm/v1/metrics/loggers.py`
```
if (len(self.engine_indexes) > 1
and vllm_config.speculative_config is not None):
raise NotImplementedError("Prometheus metrics with Spec Decoding "
"with >1 EngineCore per AsyncLLM is not "
"supported yet.")
```
To start an online server with torchair enabled, here is an example:
```
python -m vllm.entrypoints.openai.api_server \
--model="/weights/DeepSeek-R1_w8a8/" \
--trust-remote-code \
--max-model-len 40000 \
--tensor-parallel-size 4 \
--data_parallel_size 4 \
--max-num-seqs 16 \
--no-enable-prefix-caching \
--enable_expert_parallel \
--served-model-name deepseekr1 \
--speculative-config '{"num_speculative_tokens": 1, "method":"deepseek_mtp"}' \
--quantization ascend \
--host 0.0.0.0 \
--port 1234 \
--additional-config '{"ascend_scheduler_config":{"enabled":true,"enable_chunked_prefill":false},"torchair_graph_config":{"enabled":true,"graph_batch_sizes":[16]},"enable_weight_nz_layout":true}' \
--gpu_memory_utilization 0.9
```
offline example with torchair enabled
```
from vllm import LLM, SamplingParams
prompts = [
"Hello, my name is",
"The president of the United States is",
"The capital of France is",
"The future of AI is",
]
# Create a sampling params object.
sampling_params = SamplingParams(max_tokens=16, temperature=0)
# Create an LLM.
llm = LLM(
model="/home/data/DeepSeek-R1_w8a8/",
tensor_parallel_size=16,
max_num_seqs=16,
gpu_memory_utilization=0.9,
distributed_executor_backend="mp",
enable_expert_parallel=True,
speculative_config={
"method": "deepseek_mtp",
"num_speculative_tokens": 1,
},
trust_remote_code=True,
enforce_eager=False,
max_model_len=2000,
additional_config = {
'torchair_graph_config': {
'enabled': True,
"graph_batch_sizes": [16],
'enable_multistream_shared_expert': False,
},
"ascend_scheduler_config": {
"enabled": True
},
# 'expert_tensor_parallel_size': 16,
}
)
# Generate texts from the prompts.
# llm.start_profile()
outputs = llm.generate(prompts, sampling_params)
# llm.stop_profile()
for output in outputs:
prompt = output.prompt
generated_text = output.outputs[0].text
print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
```
- vLLM version: v0.10.0
- vLLM main:
302962e806
---------
Signed-off-by: xuyexiong <xuyexiong@huawei.com>
### What this PR does / why we need it?
In fact, the kimi-k2 model is similar to the deepseek model, and we only
need to make a few changes to support it. what does this pr do:
1. Add kimi-k2-w8a8 deployment doc
2. Update quantization doc
3. Upgrade torchair support list
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.10.0
- vLLM main:
9edd1db02b
---------
Signed-off-by: wangli <wangli858794774@gmail.com>
### What this PR does / why we need it?
Supports Deepseek-R1 w4a8 quantization.
Since R1 w4a8 uses mixed quantization, only the MOE layer uses
w4a8_dynamic quantization, so we added the w4a8_dynamic.py file, which
includes the AscendW4A8DynamicFusedMoEMethod class.
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
Adding ut case in `tests/ut/quantization/test_w4a8_dynamic.py` and
`tests/ut/quantization/test_quantizer.py`
Adding e2e case in
`tests/e2e/multicard/test_offline_inference_distributed.py::test_models_distributed_DeepSeek_W4A8DYNAMIC`
to test deepseek w4a8_dynamic quantized model
#### 1.How to get weights using Modelslim
##### Installation steps
Use the branch master, the commit id is:
298e175d69b3b855111a1e09bbe2fcd12fdb4e24
git clone https://gitee.com/ascend/msit.git
cd msit/msmodelslim
bash install.sh
##### The required transformers environment
transformers>=4.48.2
##### Generate w4a8 weights
cd /example/DeepSeek
Command reference: msmodelslim/example/DeepSeek/README.md Execute the
[pre-check](https://gitee.com/ascend/msit/blob/master/msmodelslim/example/DeepSeek/README.md#%E8%BF%90%E8%A1%8C%E5%89%8D%E5%BF%85%E6%A3%80)
and [DeepSeek-R1 w4a8 mix
quantization](https://gitee.com/ascend/msit/blob/master/msmodelslim/example/DeepSeek/README.md#deepseek-r1-w4a8-%E6%B7%B7%E5%90%88%E9%87%8F%E5%8C%96%E5%89%8D%E4%B8%89%E5%B1%82-mlpw8a8-dynamic-%E9%87%8F%E5%8C%96mla%E5%85%B1%E4%BA%AB%E4%B8%93%E5%AE%B6w8a8%E9%87%8F%E5%8C%96%E8%B7%AF%E7%94%B1%E4%B8%93%E5%AE%B6w4a8-dynamic%E9%87%8F%E5%8C%96)
chapter
Reference command：python3 quant_deepseek_w4a8.py --model_path {Original
weight path} --save_path {Generate weight path} --mindie_format
##### Adapt to vllm-ascend
Since mindie_format generates mindie format, some adaptation
modifications are needed for vllm-ascend to use it:
`quant_model_description_w8a8_dynamic.json` rename to
`quant_model_description.json`, and add `"group_size": 256`
Modification in `config.json`：`"model_type":deepseekv2` is changed to
`"model_type":deepseek_v3`; `quantization_config` is removed;
tips:The group_size and weights match. If the w4a8 weights are not
generated using msmodelslim, you can check the group_size in
quantization_config in config.json.
#### 2.How to run w4a8
##### a.How to run eager mode
export VLLM_USE_V1=1 # v1
python -m vllm.entrypoints.openai.api_server --model=$1
--trust-remote-code -tp $2 -dp $3 --enable_expert_parallel
--quantization ascend --port $4 --max-model-len $5 --max-num-seqs $6
--enforce-eager
eg: python -m vllm.entrypoints.openai.api_server
--model=/weightpath/w4a8_4_layer --trust-remote-code -tp 4 -dp 4
--enable_expert_parallel --quantization ascend --port 8002
--max-model-len 5120 --max-num-seqs 128 --enforce-eager
##### b.How to run graph mode
export VLLM_USE_V1=1 # v1
export HCCL_BUFFSIZE=1024
python -m vllm.entrypoints.openai.api_server --model=$1
--trust-remote-code -tp $2 -dp $3 --enable_expert_parallel
--quantization ascend --port $4 --max-model-len $5
--additional_config='{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":true}}'
eg: python -m vllm.entrypoints.openai.api_server
--model=/weight/dsr1_w4a8_vllm --trust-remote-code -tp 4 -dp 4
--enable_expert_parallel --quantization ascend --port 8002
--max-model-len 5120
--additional_config='{"ascend_scheduler_config":{"enabled":true},"torchair_graph_config":{"enabled":true}}'
- vLLM version: v0.10.0
- vLLM main:
c494f96fbc
---------
Signed-off-by: Wang Kunpeng <1289706727@qq.com>
### What this PR does / why we need it?
Fixes unable to load `qwen3_moe` quantized weights issue due to #1994
### Does this PR introduce _any_ user-facing change?
None
### How was this patch tested?
Add a `qwen3_moe` W8A8 quantized model in
`tests/e2e/multicard/test_qwen3_moe.py`
- vLLM version: v0.10.0
- vLLM main:
c494f96fbc
---------
Signed-off-by: zhoux77899 <zhouxiang100@huawei.com>
This PR significantly optimizes performance for quantized Mixture of
Experts (MoE) layers by changing the order of quantization and
communication operations.
In the previous implementation, the `all2all` operation was performed on
unquantized `hidden_states` (in FP16/BF16) *before* quantization,
resulting in substantial communication overhead. By performing
quantization on each EP rank **first** and then sending the much smaller
quantized data, we reduce the communication volume by nearly 50%.
Additionally, this PR includes a minor optimization to cast `int` inputs
to `float` for the `argsort` operation, forcing it to run on a faster
NPU core instead of the AICPU.
These changes lead to a clear and significant performance gain in MoE
quantization scenarios.
- vLLM version: v0.10.0
- vLLM main:
7175817637
---------
Signed-off-by: SlightwindSec <slightwindsec@gmail.com>
There is lot of torchair code in model runner leading the code hard for
maintenance. We'll create new torchair_model_runner to split torchair
related logic. Following the workflow #2203, this is the first PR.
What this PR does:
create the new torchair model runner, more function will be added later
- vLLM version: v0.10.0
- vLLM main:
586f286789
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
What's the PR does:
1. Move AscendSparseMoeBlock to qwen3 model, since it's only used by
qwen3 model.
2. Disable AscendSparseMoeBlock if aclgraph is enabled,
AscendSparseMoeBlock doesn't work with aclgraph currently.
- vLLM version: v0.10.0
- vLLM main:
cdfd6871a5
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
Before refactoring cross-DP decoding metadata aggregation, clean up the
token‐padding logic .
### What this PR does：
1. First checks whether any DP instance is in the prefill phase.
2. If in the `decode` phase and `torchair_graph_enabled `is true, pads
each DP instance’s token count up to the global maximum.
3. If in the `prefill` phase, or in decode phase with graph mode
**disabled**, returns each DP instance’s original token count without
padding.
This reordering removes the previous two‐step padding/unpadding flow and
ensures padding only occurs when strictly necessary.
- vLLM version: v0.10.0
- vLLM main:
bd3db7f469
Signed-off-by: yx0716 <jinyx1007@foxmail.com>
Signed-off-by: MengqingCao <cmq0113@163.com>
we recently added disaggregated_prefill and ascend_forward_context
feature by
ba3dfbd59e
and
df0ec55162.
This PR fix some nit introduced by them to make the code clear.
1. drop `current_platform` usage. It'll lead unknown circular import
error in some case
2. update `set_ascend_forward_context` function to make the logic clear.
for example, remove V0 support in this function.
3. Remove useless `self.local_rank_across_dp` in worker
4. Remove `soc_info.py` to use `get_ascend_soc_version` instead.
- vLLM version: v0.10.0
- vLLM main:
02f82fe438
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
### What this PR does / why we need it?
This pr fix broken CI:
1. Fix the
ee2eb6ecd8
changes, in this commit, they fused the gate and up projections in the
vision MLP, This can improve performance by reducing one matrix
multiplication. so, this pr do the following things:
- Specify that the two linear layers are fused as `mlp.gate_up_proj`
when loading the weights.
- Use a SiluAndMul activation function.
2. Fix
aefeea0fde,
Update ModelRunnerOutput parameters to adapt to its changes
3. Fix
[vllm-commit](https://github.com/vllm-project/vllm/pull/20815/files#diff-3ffb829a39ab2b3e4706aa28f5e476815f36c3a87b98d6a66514ebedc8f3ffb4R354-R356),
fix qwen moe
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.10.0
- vLLM main:
fed5849d3f
---------
Signed-off-by: wangli <wangli858794774@gmail.com>
(cherry picked from commit 816375e0c1071d0696dfab1a1ce35674f9f37aa0)
### What this PR does / why we need it?
Suppose that you want to start a prefiller instance with npus `2,3`
only. So you start the instance with `ASCEND_RT_VISIBLE_DEVICES=2,3`.
The current programming will start two workers, whose ranks are `0` and
`1` respectedly. And they will pick the first and second ip addresses of
npus in the ranktable instead of the thirdth and forth ones. But
actually they are using card `2,3` and therefore they can not link with
remote instances when they attempt to transfer the KVCache.
Hence, at most 1 prefiller instance and at most 1 decoder instance can
work on a single machine since they always pick the first npu ip address
in the ranktable currently.
This pull request is proposed to fix the problem. This fix pick ips of
only those devices that are in `ASCEND_RT_VISIBLE_DEVICES` from the
ranktable.
### Does this PR introduce _any_ user-facing change?
If the user use ranktable generated by `gen_ranktable.sh`, they should
not face any change.
### How was this patch tested?
Qwen-0.6B 1P 1D, dp=2, `ASCEND_RT_VISIBLE_DEVICES=2,3` for prefiller and
`ASCEND_RT_VISIBLE_DEVICES=4,5` for decoder.
- vLLM version: v0.10.0
- vLLM main:
ad57f23f6a
Signed-off-by: CaveNightingale <cavenightingale@foxmail.com>
### What this PR does / why we need it?
Updates the FusedMoE method to determine whether to use ACL Graph based
on the `torchair_graph_config`
This is equivalent to #2154 on v0.9.1-dev.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
None needed.
- vLLM version: v0.10.0
- vLLM main:
ad57f23f6a
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
When run the inference of ds-w8a8-mtp, it reported 'ParamllelLMhead has
no attribute 'params_dtype''.
1. add wrapper of vocab_parallel_embedding, fixed the bugs when running
deepseek-w8a8-mtp
Signed-off-by: curryliu <120010041@link.cuhk.edu.cn>
- vLLM version: v0.10.0
- vLLM main:
ad57f23f6a
---------
Signed-off-by: curryliu <120010041@link.cuhk.edu.cn>
### What this PR does / why we need it?
Fix Qwen MoE accuracy bug in DP scenario.
Now the implentment of `FusedMoE` in vLLM use `All2AllManager` to
manager different all2all algorithm branch. And the default branch use
`Multicast` in `dispatch` phase and `all_reduce` in `combine` phase,
which are not implented in vLLM-Ascend. This leading to invoking into a
default implentment in `base_communicator`, with empty `dispatch` and
`combine` operations, thus causing the accuracy issue on it.
This pr is a temporary workaround, refacting all2all in vLLM-Ascend
could be a better way.
- vLLM version: v0.10.0
- vLLM main:
ad57f23f6a
---------
Signed-off-by: MengqingCao <cmq0113@163.com>
### What this PR does / why we need it?
We notice that vllm's main branch merged the PR
https://github.com/vllm-project/vllm/pull/21072 and
https://github.com/vllm-project/vllm/pull/21473 to support ray backend
and fix some rebase bug from previous change. Those changes makes the
disaggregate pd in vllm ascend breaks in some scenario.
In this PR, we adopt those changes to make sure the
`llmdatddist_c_mgr_connector` works fine on the newest vllm main branch.
### Does this PR introduce _any_ user-facing change?
No user face change.
### How was this patch tested?
relevant ut will be added to make sure the functionality of those
changes.
- vLLM version: v0.10.0
- vLLM main:
ad57f23f6a
---------
Signed-off-by: ganyi <pleaplusone.gy@gmail.com>
### What this PR does / why we need it?
cherry-pick #1675 to main
This PR adds validation checking to torchair_graph_config for better
reliability.
Co-authored-by: whx-sjtu <2952154980@qq.com>
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
- vLLM version: v0.10.0
- vLLM main:
2836dd73f1
Signed-off-by: 22dimensions <waitingwind@foxmail.com>
### What this PR does / why we need it?
Cherry pick #1291 from v0.9.1-dev, This pr implement the synchronization
of whether `dbo` is enabled across all dp ranks. specifically, it
performed allreduce op across multiple DP ranks, only when all the dp
rank is `enable_dbo`, it is enabled
Co-authored-by: shikang-hangzhou <459956190@qq.com>
Co-authored-by: wangli <wangli858794774@gmail.com>
- vLLM version: v0.10.0
- vLLM main:
2836dd73f1
---------
Signed-off-by: wangli <wangli858794774@gmail.com>
### What this PR does / why we need it?
cherry-pick #1501 from 0.9.1-dev to main
Currently, Ray is not compatible with ACL Graph, so we need to fall back
to eager mode when using the Ray backend.
co-authored: Yizhou Liu <liu_yizhou@outlook.com>
- vLLM version: v0.10.0
- vLLM main:
2836dd73f1
Signed-off-by: 22dimensions <waitingwind@foxmail.com>
### What this PR does / why we need it?
Cherry pick #1705 from v0.9.1-dev
Compared qwen2_5_vl.py, qwen2_5_vl_without_padding.py missing some
funtions. The purpose of this PR is to supplement these.
add:
- rot_pos_emb(self, grid_thw: torch.Tensor)
- get_window_index(self, grid_thw)
- _process_image_input(self, image_input)
- _process_video_input(self, video_input)
Co-authored-by: zheliuyu
[15750543867@163.com](mailto:15750543867@163.com)
Co-authored-by: wangli
[wangli858794774@gmail.com](mailto:wangli858794774@gmail.com)
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.10.0
- vLLM main:
207b750e19
Signed-off-by: wangli <wangli858794774@gmail.com>
### What this PR does / why we need it?
cherry pick #1749 from v0.9.1-dev
since the interface in vllm-ascend has changed so quickly, the
quantization function in mindie_turbo is no longer needed, so it needs
to be discarded.
Co-authored-by: zouyida [zouyida@huawei.com](mailto:zouyida@huawei.com)
Co-authored-by: wangli
[wangli858794774@gmail.com](mailto:wangli858794774@gmail.com)
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.10.0
- vLLM main:
207b750e19
Signed-off-by: wangli <wangli858794774@gmail.com>
### What this PR does / why we need it?
Fixed 310p failure when using the sampler feature.
The root cause is: torch_npu.npu_top_k_top_p uses the operator
aclnnApplyTopKTopP, but aclnnApplyTopKTopP currently does not support
310P.
First PR that has the issue is #1308.
### Does this PR introduce _any_ user-facing change?
No
- vLLM version: v0.10.0
- vLLM main:
207b750e19
Signed-off-by: leo-pony <nengjunma@outlook.com>
This PR fixes the bug `local variable 'decode_hs_or_q_c' referenced
before assignment` when running chunked-prefill with torchair. We should
calculate `decode_hs_or_q_c` whether or not torchair graphics mode is
enabled.
backport of #1378
fix https://github.com/vllm-project/vllm-ascend/issues/1369
- vLLM version: v0.10.0
- vLLM main:
0e36abf993
---------
Signed-off-by: whx-sjtu <2952154980@qq.com>
Signed-off-by: MengqingCao <cmq0113@163.com>
Co-authored-by: whx-sjtu <2952154980@qq.com>
### What this PR does / why we need it?
Fix vLLM startup failure when inferring DeepSeek R1 model in DP
scenario.
When running vLLM inference for the DeepSeek R1 model in DP32+TP1
configuration, the vLLM service fails to start with the following error.
<img width="1786" height="918" alt="21b2011042d4f77f36f5243fa64d9c18"
src="https://github.com/user-attachments/assets/df1963fe-587e-43ca-822e-a9094d0034fb"
/>
The root cause is a missing else branch after [this line of
code](d629f0b2b5/vllm_ascend/ops/fused_moe.py (L1411)).
This PR fixes the issue.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
CI passed with new added/existing test.
- vLLM version: v0.10.0
- vLLM main:
5bbaf492a6
---------
Signed-off-by: zhanghaiwen <zhanghaiwen@cmss.chinamobile.com>
Co-authored-by: zhanghaiwen <zhanghaiwen@cmss.chinamobile.com>
bugfix cherry-pick from v0.9.1-dev
https://github.com/vllm-project/vllm-ascend/pull/2007
### What this PR does / why we need it?
Minimum reproducing code：
```python
# test.py
from vllm import LLM, SamplingParams
prompts = [
"Hello, my name is",
"The future of AI is",
]
sampling_params = SamplingParams(temperature=0.8, top_p=0.95)
llm = LLM(model="Qwen2.5-VL-7B-Instruct", max_model_len=26240)
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}")
```
```bash
export USE_OPTIMIZED_MODEL=0
python test.py
```
exception as follow:
```
[rank0]: File "/home/xxx/vllm_ascend/models/qwen2_5_vl_without_padding.py", line 84, in forward
[rank0]: q = torch_npu.npu_rotary_mul(q, cos, sin)
[rank0]: File "/home/anaconda3/envs/xxx/lib/python3.10/site-packages/torch/_ops.py", line 1116, in __call__
[rank0]: return self._op(*args, **(kwargs or {}))
[rank0]: RuntimeError: Expected all tensors to be on the same device, but found at least two devices, npu:0 and cpu! (when checking argument for argument r1 in method wrapper__npu_rotary_mul)
```
In `AscendQwen2_5_VisionAttention_Without_Padding`,
`torch_npu.npu_rotary_mul(q, cos, sin)`， `cos`/`sin` on cpu, but `q` on
npu, so there will be an error.
`qwen2_5_vl_without_padding.py` need this bugfix, because
`AscendQwen2_5_VisionTransformer_Without_Padding.rot_pos_emb` in
wen2_5_vl_without_padding.py is from vllm and `inv_freq` will create on
cpu.
40d86ee412/vllm/model_executor/models/qwen2_5_vl.py (L482)
```python
inv_freq = 1.0 / (theta**(torch.arange(0, dim, 2, dtype=torch.float, device='cpu') / dim))
```
`qwen2_5_vl.py` do not need, because
`AscendQwen2_5_VisionRotaryEmbedding` in qwen2_5_vl.py rewrite
`AscendQwen2_5_VisionRotaryEmbedding` and `inv_freq` will create on
device.
```python
inv_freq = 1.0 / (theta**(torch.arange(0, dim, 2, dtype=torch.float) / dim))
```
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
CI passed with new added/existing test.
- vLLM version: v0.10.0
- vLLM main:
18cc33dd60
Signed-off-by: pjgao <gaopengju3@huawei.com>
Co-authored-by: pjgao <gaopengju3@huawei.com>
### What this PR does / why we need it?
Fix#2033
Sync https://github.com/vllm-project/vllm/pull/14702 to solve
`grammar_bitmask` IndexError caused by outdated `apply_grammar_bitmask`
method
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Tested by upstream vllm
- vLLM version: v0.10.0
- vLLM main:
6e599eebe8
Signed-off-by: ApsarasX <apsarax@outlook.com>
### What this PR does / why we need it?
Adding `W4A8_DYNAMIC` quantization support for linear.
Dense models like Qwen3 can infer with `W4A8_DYNAMIC` quantization.
### Does this PR introduce _any_ user-facing change?
None
### How was this patch tested?
Adding ut case in `tests/ut/quantization/test_w4a8_dynamic.py`
Adding e2e case in
`tests/e2e/multicard/test_offline_inference_distributed.py::test_models_distributed_Qwen3_W4A8DYNAMIC`
to test qwen3 w4a8_dynamic quantized model
Note the w4a8_dynamic quantized model is quantized by `msit/msmodelslim`
of commit `d0abb0a47e1f1a473b866ad41b737fbc28fb1409`
1. Generate `W4A8_DYNAMIC` quantization weights using `msmodelslim`
```shell
git clone https://gitee.com/ascend/msit.git
cd msit/msmodelslim
git checkout d0abb0a47e1f1a473b866ad41b737fbc28fb1409
bash install.sh
```
2. Serve model using `vllm`
```shell
VLLM_USE_V1=1 python -m vllm.entrypoints.openai.api_server \
--model vllm-ascend/Qwen3-8B-W4A8 \
--port 8000 \
--quantization ascend \
--tensor_parallel_size 2 \
--enforce-eager
```
- vLLM version: v0.10.0
- vLLM main:
4cd7fe6cea
---------
Signed-off-by: ZhouXiang <zhouxiang100@huawei.com>
### What this PR does / why we need it?
This PR fixes a tensor shape mismatch in `add_lora_logits`.
Previously, `lora_a_stacked` was passed as shape `[num_loras, in_dim,
rank]`, which does not match the expected einsum pattern `"bi, boi ->
bo"` used in `bgmv_shrink`.
This causes runtime errors like:
RuntimeError: einsum(): subscript i has size 3 for operand 1 which does
not broadcast with previously seen size 4
This fix transposes `lora_a_stacked` and `lora_b_stacked` to match the
expected shapes:
- `lora_a`: `[num_loras, rank, in_dim]`
- `lora_b`: `[num_loras, out_dim, rank]`
All unit tests pass after this fix.
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
```
import torch
import pytest
from unittest.mock import patch, PropertyMock, ANY
from vllm_ascend.lora.punica_wrapper.punica_npu import PunicaWrapperNPU
@pytest.fixture
def wrapper_cpu():
cfg = {"max_num_batched_tokens": 10, "max_batches": 2, "device": "cpu"}
w = PunicaWrapperNPU(**cfg)
w.is_prefill = True
w.no_lora = False
return w
def test_add_lora_logits(wrapper_cpu):
batch_size = 2
hidden_size = 4
lora_rank = 3
vocab_size = 5
y = torch.zeros(batch_size, vocab_size)
x = torch.randn(batch_size, hidden_size)
num_loras = 1
lora_a = torch.randn(num_loras, hidden_size, lora_rank)
lora_b = torch.randn(num_loras, lora_rank, vocab_size)
with patch.object(wrapper_cpu.__class__, "sampler_indices",
new_callable=PropertyMock) as mock_idx:
mock_idx.return_value = torch.zeros(batch_size, dtype=torch.long)
wrapper_cpu.add_lora_logits(y, x, lora_a, lora_b, scale=1.0)
assert y.shape == (batch_size, vocab_size)
assert not torch.allclose(y, torch.zeros_like(y))
Signed-off-by: hongfugui <hongfugui_yewu@cmss.chinamobile.com>
Refactor Sampler implementation from patch way to inherit from vLLM
Sampler interface.
Next step: Make the op `TopKTopPSampler` in vLLM support custom ops
register mechanism
- vLLM version: v0.10.0
- vLLM main:
61a6905ab0
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
This PR designs the shared expert multi-stream parallelism of
w8a8-dynamic-quantized MoE stage in more detail to achieve better
performance.
- vLLM version: v0.10.0
- vLLM main:
2cc571199b
Signed-off-by: whx-sjtu <2952154980@qq.com>
### What this PR does / why we need it?
Add two custom kernels(bgmv_shrink and bgmv expand) to solve the
performance of LoRA
### Does this PR introduce _any_ user-facing change?
no user-facing change
### How was this patch tested?
we add Unit Test file to test the custom ascendc kernel. See
vllm-ascend/tests/e2e/singlecard/ops/test_bgmv_expand.py and
vllm-ascend/tests/e2e/singlecard/ops/test_bgmv_expand.py
Based on the actual test of the QWen2.5 7B model using vllm-ascend
version v0.9.2.rc1, the TTFT, TPOT and throughput have increased by
about 70%.
- vLLM version: v0.9.2
- vLLM main:
40d86ee412
---------
Signed-off-by: taoxudonghaha <justsheldon@163.com>
Support the inference of the Deepseekr1-w8a8-mtp model with
statically-quantized shared_head in MTP layers.
- vLLM version: v0.9.2
- vLLM main:
6eca337ce0
Signed-off-by: curryliu <120010041@link.cuhk.edu.cn>
This PR removes the restriction that TP cannot be greater than 4 in
torchair scenario, because current newest version of CANN has fixed this
bug.
- vLLM version: v0.10.0
- vLLM main:
04ff4be310
Signed-off-by: whx-sjtu <2952154980@qq.com>
Clean up useless import from vllm to make code more clear.
- vLLM version: v0.10.0
- vLLM main:
18cc33dd60
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
### What this PR does / why we need it?
it'll execute allreduce and malmul seperately in vllm RowParallelLinear
forward funcion, this function use torch_npu.npu_mm_all_reduce_base to
execute allreduce and matmul in a fused kernel way. this will gain a 20%
performance
promotion in eager mode.
### Does this PR introduce _any_ user-facing change?
this PR introduce a new env `VLLM_ASCEND_ENABLE_MATMUL_ALLREDUCE` to
control whether enable the feature or not.
### How was this patch tested?
the patch is tested by adding a new test file `test_patch_linear.py` to
guard the ut
- vLLM version: v0.10.0
- vLLM main:
7728dd77bb
Signed-off-by: Ronald1995 <ronaldautomobile@163.com>
### What this PR does / why we need it?
A refactoring of forward_context and model_runner_v1, add some context
which is necessary in model inference into forward_context, and refactor
dummy_run logic, make it more reasonable.
Some details for this PR:
Add `ascend_forward_context`;
Update mc2_v2 op, and support `active_mask` param;
Update scripts in examples dir;
refactor `dummy_run` logic;
Add soc_version for A2 and A3;
### Does this PR introduce _any_ user-facing change?
No change at user-facing.
### How was this patch tested?
- vLLM version: v0.10.0
- vLLM main:
57c22e57f9
Signed-off-by: zzzzwwjj <1183291235@qq.com>
### What this PR does / why we need it?
Fix num_hidden_layers when Qwen2-Audio 7B and #1760 ：
```
INFO 07-15 04:38:53 [platform.py:174] PIECEWISE compilation enabled on NPU. use_inductor not supported - using only ACL Graph mode
Traceback (most recent call last):
File "/workspace/test1.py", line 58, in <module>
main(audio_count)
File "/workspace/test1.py", line 38, in main
llm = LLM(model="Qwen/Qwen2-Audio-7B-Instruct",
File "/vllm-workspace/vllm/vllm/entrypoints/llm.py", line 271, in __init__
self.llm_engine = LLMEngine.from_engine_args(
File "/vllm-workspace/vllm/vllm/engine/llm_engine.py", line 494, in from_engine_args
vllm_config = engine_args.create_engine_config(usage_context)
File "/vllm-workspace/vllm/vllm/engine/arg_utils.py", line 1286, in create_engine_config
config = VllmConfig(
File "/usr/local/python3.10.17/lib/python3.10/site-packages/pydantic/_internal/_dataclasses.py", line 123, in __init__
s.__pydantic_validator__.validate_python(ArgsKwargs(args, kwargs), self_instance=s)
File "/vllm-workspace/vllm/vllm/config.py", line 4624, in __post_init__
current_platform.check_and_update_config(self)
File "/vllm-workspace/vllm-ascend/vllm_ascend/platform.py", line 180, in check_and_update_config
update_aclgraph_sizes(vllm_config)
File "/vllm-workspace/vllm-ascend/vllm_ascend/utils.py", line 307, in update_aclgraph_sizes
num_hidden_layers = vllm_config.model_config.hf_config.num_hidden_layers
File "/usr/local/python3.10.17/lib/python3.10/site-packages/transformers/configuration_utils.py", line 211, in __getattribute__
return super().__getattribute__(key)
AttributeError: 'Qwen2AudioConfig' object has no attribute 'num_hidden_layers'
```
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
Closes: https://github.com/vllm-project/vllm-ascend/issues/1780https://github.com/vllm-project/vllm-ascend/issues/1760https://github.com/vllm-project/vllm-ascend/issues/1276https://github.com/vllm-project/vllm-ascend/issues/359
- vLLM version: v0.10.0
- vLLM main:
7728dd77bb
Signed-off-by: hfadzxy <starmoon_zhang@163.com>
### What this PR does / why we need it?
- Upgrade to v0.10.0
- Drop v0.9.2 version compatibility
- Add patch for
`vllm_ascend/patch/worker/patch_common/patch_sampler_gather_logprobs.py`
as workaround of
f3a683b7c9
for v0.10.0 and also add e2e test `test_models_prompt_logprobs`
- Pin transformers<4.54.0 as workaround of
https://github.com/vllm-project/vllm-ascend/issues/2034
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
- Test locally:
`VLLM_USE_MODELSCOPE=true pytest -sv
tests/e2e/singlecard/test_offline_inference.py::test_models_prompt_logprobs`
- CI passed
- vLLM version: v0.9.2
- vLLM main:
7728dd77bb
---------
Signed-off-by: Yikun Jiang <yikunkero@gmail.com>
Signed-off-by: wuzhongjian <wuzhongjian_yewu@cmss.chinamobile.com>
### What this PR does / why we need it?
Fix duplicate 'torch.' prefix in qwen2-vl, qwen2.5-vl
- vLLM version: v0.9.2
- vLLM main:
dde295a934
