### What this PR does / why we need it?
Add ACL graph capture/replay DP test, this is a imprved version of #3886
Restructures the multi-card ACL graph test for improved clarity,
robustness, and accuracy.
Key improvements include:
- Replaces fragile `sys.settrace` and manual patching with a clean,
reusable spy installer using `unittest.mock.patch`.
- Introduces more precise metrics by tracking
`NPUModelRunner.execute_model` and `_dummy_run` calls directly.
- Rewrites assertions to be more accurate and provides clear
explanations for the expected counts of graph captures, replays, model
executions, and dummy runs.
- Simplifies the overall test structure by separating the worker logic
into a dedicated function.
- Removes a long, unnecessary sleep at the end of the test.
- Expands test coverage by adding a larger `max_tokens` parameter.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
None.
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/2918c1b49c88c29783c86f78d2c4221cb9622379
---------
Signed-off-by: lilinsiman <lilinsiman@gmail.com>
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
Co-authored-by: lilinsiman <lilinsiman@gmail.com>
### What this PR does / why we need it?
Currently, the MTP model still runs in eager in full graph mode. This PR
adapts the MTP with the full graph capture and execution. When the graph
mode is set to "FULL_DECODE_ONLY", the MTP will run in full-graph to
improve the performance.
The change in both disable_padded_drafter_batch is True and False case
include:
1. Add _mtp_graph_params in acl_graph.py to isolate the data of main
model and the data of MTP.
2. Padding some metadata in mla_v1.py when in fullgraph mode.
3. Fixed the essential data address that will be used in model.forward.
4. Adapted according to the aclgraph capture framwork:
1). Rebuild MTP model with ACLGraphWrapper.
2). Add common attn metadata when start capture in MTP dummy_run.
3). Add common attn metadata update in MTP.
4). Addapted data update when num_speculative_tokens > 1.
5. Add a patch of MTP to adapt vllm v0.11.0.
Existing Issues:
1. When disable_padded_drafter_batch=True and running in FullGraph mode,
the data of the first-round requests in MTP is abnormal. We need to
identify the cause subsequently.
2. When disable_padded_drafter_batch=False and running in FullGraph
mode, the acceptance rate of the second and third tokens will decrease
(For example, if we set the num_speculative_tokens=3, the acceptance
rate of first token is 90%, the second is only 50% lower than 60%, the
third is only 20% lower than 30%). The reason is that the data processed
after the model runs does not match. This is a problem from another PR.
It works fine in eager and PIECEWISE mode, but has problem in FullGraph
mode. Once we have a solution, we will submit a bugfix.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/2918c1b49c88c29783c86f78d2c4221cb9622379
---------
Signed-off-by: anon189Ty <Stari_Falcon@outlook.com>
Fix pcp + mtp bug while using acl graph.
While using pcp + mtp, we need to flatten block_table to avoid irregular
attn mask shape, this was done in mla attn_metadata builder, but we
found out that this influences block_table address and leads to
incorrect results while enable acl graph.
To fix this, we enlarge block_table buffer size and flatten block_table
in model_runner prepare_inputs, so this will not influence block_table
address.
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/2918c1b49c88c29783c86f78d2c4221cb9622379
Signed-off-by: zhangsicheng5 <zhangsicheng5@huawei.com>
### What this PR does / why we need it?
Past:
npu_moe_gating_top_k can only support 'group_count=256' pattern
Now:
1、npu_moe_gating_top_k support all size of group_count
2、the functionality of `torch_npu.npu_moe_gating_top_k_softmax` are
included in `torch_npu.npu_moe_gating_top_k`
CANN: depends on 8.3.RC1
Performance:
1. GLM4.5-w8a8, TPS improve 6%
2. Qwen3, the same as before
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/2918c1b49c88c29783c86f78d2c4221cb9622379
Signed-off-by: 1092626063 <1092626063@qq.com>
### What this PR does / why we need it?
Currently, the default `cudagraph_capture_size` in vLLM is `[1, 2, 4 ,8
,16 ,24 ,... , max_capture_size]`. However, this is not always the best
choice on different situations. This PR aims to change the default
setting when running Qwen3-MoE on full dp (`dp_size > 1` && `tp_size ==
1`) setting, which is usually applied in Large-Scale EP.
old :
`[1, 2, 4 ,8 ,16 ,24 ,... , max_capture_size]`
new:
`[1, 2, 5 ,10 ,15, 16 ,24 ,... , max_capture_size]`
This is mainly because the performance of `_npu_paged_attention` op
degrades dramatically on old settings. We hope to provide better
performance if users do not set specific `cudagraph_capture_size`.
### Does this PR introduce _any_ user-facing change?
The default `cudagraph_capture_size` is modified in above cases.
However, if `cudagraph_capture_size` has already set by users, this PR
won't have any influence on this.
### How was this patch tested?
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/2918c1b49c88c29783c86f78d2c4221cb9622379
---------
Signed-off-by: Angazenn <supperccell@163.com>
### What this PR does / why we need it?
we often install vllm-ascend in developer mode, which has no _build_info
module. it will raise error in `utils.is_310p` and
`utils.sleep_model_enabled`, then we need to modify these two function.
### Does this PR introduce _any_ user-facing change?
not involved
### How was this patch tested?
not involved
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/2918c1b49c88c29783c86f78d2c4221cb9622379
Signed-off-by: Ronald1995 <ronaldautomobile@163.com>
### What this PR does / why we need it?
Introduces the `acl_graph_print` function to enable printing debug
information from code running inside an ACL graph, such as custom
operators.
This works by launching a host function on a dedicated stream, bypassing
the limitations of standard `print` within compiled graph execution. The
implementation handles the necessary stream subscriptions and ensures
they are properly unregistered upon exit.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
None.
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/2918c1b49c88c29783c86f78d2c4221cb9622379
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
### What this PR does / why we need it?
Given the current excessively long build time of our nightly-ci, I
recommend installing necessary, confirmed versions of packages in the
Docker image to reduce the time required for integration testing.
Including Mooncake vllm with fixed tags, This is expected to reduce
nightly-ci duration by 2 hours.
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/2918c1b49c88c29783c86f78d2c4221cb9622379
---------
Signed-off-by: wangli <wangli858794774@gmail.com>
### What this PR does / why we need it?
Currently, the UT tests lack coverage for the Qwen3_moe network and
torchair_sfa. Therefore, supplementary tests are being added.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
by CI
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/83f478bb19489b41e9d208b47b4bb5a95ac171ac
---------
Signed-off-by: CodeNine-CJ <chenjian343@huawei.com>
### What this PR does / why we need it?
Now, from https://github.com/vllm-project/vllm-ascend/pull/3967, chunked
prefill and spiltfuse are defaultly enabled.
The e2e test for mtp breaks now.
After locating the bug, we found that a triton operator does not support
chunked prefill.
But if let e2e test be skipped is bad.
So, we changed the e2e test to only test the case in which chunked
prefill is off.
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
Because we only modified
`test_models_distributed_Qwen3_NEXT_MTP_TP4_SIMILARITY`.
So, we only run `pytest -s
tests/e2e/multicard/test_qwen3_next.py::test_models_distributed_Qwen3_NEXT_MTP_TP4_SIMILARITY`
locally to test it.
Below is the result:
```text
==================================================================================================================== warnings summary ====================================================================================================================
usr/local/python3.11.10/lib/python3.11/site-packages/torch_npu/dynamo/torchair/__init__.py:8
/usr/local/python3.11.10/lib/python3.11/site-packages/torch_npu/dynamo/torchair/__init__.py:8: DeprecationWarning: pkg_resources is deprecated as an API. See https://setuptools.pypa.io/en/latest/pkg_resources.html
import pkg_resources
<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
tests/e2e/multicard/test_qwen3_next.py::test_models_distributed_Qwen3_NEXT_MTP_TP4_SIMILARITY
tests/e2e/multicard/test_qwen3_next.py::test_models_distributed_Qwen3_NEXT_MTP_TP4_SIMILARITY
/usr/local/python3.11.10/lib/python3.11/site-packages/pydantic/_internal/_dataclasses.py:121: DeprecationWarning: The 'task' option has been deprecated and will be removed in v0.13.0 or v1.0, whichever comes first. Please remove this option.
s.__pydantic_validator__.validate_python(ArgsKwargs(args, kwargs), self_instance=s)
-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
======================================================================================================= 1 passed, 5 warnings in 314.52s (0:05:14) ========================================================================================================
sys:1: DeprecationWarning: builtin type swigvarlink has no __module__ attribute
```
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/2918c1b49c88c29783c86f78d2c4221cb9622379
Signed-off-by: drslark <slarksblood@qq.com>
### What this PR does / why we need it?
In PR https://github.com/vllm-project/vllm-ascend/pull/3420, we
initially placed the quantization type (quant_type) in the MoECommMethod
class. However, since MoECommMethod follows a singleton pattern, it
couldn't accommodate scenarios where different layers in the model might
use different quantization approaches (e.g., MTP modules using
floating-point computation while the main model employs quantized
computation).
In this PR, we've moved the quantization type to the AscendFusedMoe
class and pass it as a parameter to MoECommMethod.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
```bash
export HCCL_BUFFSIZE=1024
export VLLM_VERSION=0.11.0
vllm serve /home/data/DeepSeek-R1_w8a8/ \
--data-parallel-size 2 \
--tensor-parallel-size 8 \
--enable-expert-parallel \
--served-model-name dsv3 \
--max-model-len 32768 \
--max-num-batched-tokens 4096 \
--max-num-seqs 16 \
--quantization ascend \
--trust-remote-code \
--gpu-memory-utilization 0.9 \
--speculative-config '{"num_speculative_tokens": 2, "method":"deepseek_mtp"}'
```
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/83f478bb19489b41e9d208b47b4bb5a95ac171ac
---------
Signed-off-by: realliujiaxu <realliujiaxu@163.com>
### What this PR does / why we need it?
Only CPU tensors with `pin_memory=True` can be asynchronously copied to
the device. Currently, there are two instances where non-pinned CPU
tensors are being copied to the device, which will trigger synchronous
operations, reducing the expected benefits of asynchronous scheduling.
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/83f478bb19489b41e9d208b47b4bb5a95ac171ac
Signed-off-by: realliujiaxu <realliujiaxu@163.com>
### What this PR does / why we need it?
Explicit specification `NUMEXPR_MAX_THREADS` to avoid `Error. nthreads
cannot be larger than environment variable "NUMEXPR_MAX_THREADS" (64)`
Signed-off-by: wangli <wangli858794774@gmail.com>
### What this PR does / why we need it?
Currently, we set `seq_lens` in dummy attn_metadata to be
`max_model_len` to get max workspace for attention during capturing.
However, setting it consistently to be `max_model_len` causing dummy_run
to execute a long attention when running actual inference. For example,
if there is a single req with `seqs_lens` as [8] but `max_model_len` is
131072, the whole process will be slow down by dummy_run as it execute a
fake long-seq attention. Therefore, we instead set it to max_query_len,
which is also consistent with vLLM gpu implementation.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/83f478bb19489b41e9d208b47b4bb5a95ac171ac
---------
Signed-off-by: Angazenn <supperccell@163.com>
### What this PR does / why we need it?
Replace masked in-place assignment with a device-side torch.where so
selection stays on-device, allowing subsequent device ops to be enqueued
earlier and removing an implicit D2H sync, reducing latency by several
hundreds μs on Ascend.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
None.
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/83f478bb19489b41e9d208b47b4bb5a95ac171ac
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>