### What this PR does / why we need it?
Part of #5304.
We have align with vLLM's latest change for `RotaryEmbeddingBase`. Don't
need this patch anymore.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.14.1
- vLLM main:
dc917cceb8
Signed-off-by: gcanlin <canlinguosdu@gmail.com>
## Summary
- Add automatic quantization format detection, eliminating the need to
manually specify `--quantization` when serving quantized models.
- The detection inspects only lightweight JSON files
(`quant_model_description.json` and `config.json`) at engine
initialization time, with no `.safetensors` reads.
- User-explicit `--quantization` flags are always respected;
auto-detection only applies when the flag is omitted.
## Details
**Detection priority:**
1. `quant_model_description.json` exists → `quantization="ascend"`
(ModelSlim)
2. `config.json` contains `"quant_method": "compressed-tensors"` →
`quantization="compressed-tensors"` (LLM-Compressor)
3. Neither → default float behavior
**Technical approach:**
Hooked into `NPUPlatform.check_and_update_config()` to run detection
after `VllmConfig.__post_init__`. Since `quant_config` is already `None`
at that point, we explicitly recreate it via
`VllmConfig._get_quantization_config()` to trigger the full quantization
initialization pipeline.
## Files Changed
| File | Description |
|------|-------------|
| `vllm_ascend/quantization/utils.py` | Added
`detect_quantization_method()` and `maybe_auto_detect_quantization()` |
| `vllm_ascend/platform.py` | Integrated auto-detection in
`check_and_update_config()` |
| `vllm_ascend/quantization/modelslim_config.py` | Improved error
handling for weight loading |
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd
---------
Signed-off-by: SlightwindSec <slightwindsec@gmail.com>
### What this PR does / why we need it?
New FIA operator requires queryT equal to the last element of
actualSequenceLengthQ.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Passed existing test (test_mtp_eagle_correctness.py).
- vLLM version: v0.15.0
- vLLM main:
9562912cea
---------
Signed-off-by: Wangbingjie <wangbj1207@126.com>
Signed-off-by: Wangbingjie <w30061490@china.huawei.com>
Co-authored-by: Wangbingjie <w30061490@china.huawei.com>
[Refactor] Modify the binding logic, added memory migration and
interrupt core binding functions.
### What this PR does / why we need it?
Controls the use of memory on a closer NUMA node to achieve a lower
memory access latency, while binding interrupts to different CPU cores
to prevent them form interrupting the inference process.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
b8eaaa073b
Signed-off-by: rowzwel_dx <1392851715@qq.com>
Signed-off-by: Rozwel-dx <1392851715@qq.com>
- vLLM version: v0.15.0
- vLLM main:
9562912cea
Signed-off-by: Rozwel-dx <1392851715@qq.com>
### What this PR does / why we need it?
Currently, the performance of multi-modal encoding (i.e.,
`AscendMMEncoderAttention` forward) is considerably bounded by the heavy
host pre-process operations.
We can see from the profiling results below, before the real computation
of Attention, there are long free time in the device, which will lead to
extremely low NPU utilization.
<img width="2264" height="1398" alt="iShot_2026-01-23_16 26 39"
src="https://github.com/user-attachments/assets/37f21d06-e526-4f28-82fe-005746cf13bd"
/>
---
**To opitimize this, this PR has proposed four changes:**
1. Use `seq_lens` CPU cache to avoid frequent d2h copy. Before this PR,
`AscendMMEncoderAttention` will copy the `cu_seqlens` from NPU to CPU in
every forward, since the op `_npu_flash_attention_unpad()` requires CPU
`cu_seqlens` (otherwise it will crash). Thus, we use
`seq_lens_cpu_cache` to cache this tensor, since it's shared between all
layers, but may change in different forward step. When the current
`layer_index` is `0`, we update the cache, otherwise we directly use the
cache to avoid frequent `diff` and `copy` operations, which are costful.
2. Pre-compute the scale value to avoid calculating it in every forward.
3. Move the judgment of `enable_pad` from forward to the `__init__`
method.
4. Revert https://github.com/vllm-project/vllm-ascend/pull/6204.
**Performance after these optimizations:**
- **TTFT** has been reduced by **7.43%** ⬇️.
- **Throughput** has been increased by **1.23%** ⬆️.
---
> [!NOTE]
> This PR requires https://github.com/vllm-project/vllm/pull/33674 be
merged.
---
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
Launch the server:
```bash
vllm serve /root/.cache/modelscope/hub/models/Qwen/Qwen3-VL-8B-Instruct \
--dtype bfloat16 \
--limit-mm-per-prompt '{"image": 1}' \
--max-model-len 16384 \
--max-num-batched-tokens 16384 \
--no-async-scheduling
```
Run benchmark:
```bash
vllm bench serve \
--model /root/.cache/modelscope/hub/models/Qwen/Qwen3-VL-8B-Instruct \
--backend openai-chat \
--endpoint /v1/chat/completions \
--dataset-name hf \
--hf-split train \
--dataset-path lmarena-ai/vision-arena-bench-v0.1 \
--num-prompts 500 \
--request-rate 10 \
--burstiness 5 \
--no-stream
```
Before this PR:
```
============ Serving Benchmark Result ============
Successful requests: 500
Failed requests: 0
Request rate configured (RPS): 10.00
Benchmark duration (s): 82.23
Total input tokens: 33418
Total generated tokens: 61543
Request throughput (req/s): 6.08
Output token throughput (tok/s): 748.45
Peak output token throughput (tok/s): 3203.00
Peak concurrent requests: 402.00
Total token throughput (tok/s): 1154.86
---------------Time to First Token----------------
Mean TTFT (ms): 10275.37
Median TTFT (ms): 6297.88
P99 TTFT (ms): 22918.26
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms): 263.02
Median TPOT (ms): 277.61
P99 TPOT (ms): 483.56
---------------Inter-token Latency----------------
Mean ITL (ms): 257.31
Median ITL (ms): 94.83
P99 ITL (ms): 1773.90
==================================================
```
After this PR:
```
============ Serving Benchmark Result ============
Successful requests: 500
Failed requests: 0
Request rate configured (RPS): 10.00
Benchmark duration (s): 81.20
Total input tokens: 33418
Total generated tokens: 61509
Request throughput (req/s): 6.16
Output token throughput (tok/s): 757.54
Peak output token throughput (tok/s): 2562.00
Peak concurrent requests: 395.00
Total token throughput (tok/s): 1169.11
---------------Time to First Token----------------
Mean TTFT (ms): 9511.91
Median TTFT (ms): 5479.78
P99 TTFT (ms): 21427.21
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms): 261.12
Median TPOT (ms): 276.03
P99 TPOT (ms): 446.99
---------------Inter-token Latency----------------
Mean ITL (ms): 254.04
Median ITL (ms): 97.71
P99 ITL (ms): 1516.67
==================================================
```
- vLLM version: v0.15.0
- vLLM main:
dc917cceb8
Signed-off-by: shen-shanshan <467638484@qq.com>
### What this PR does / why we need it?
In order to adapt to the GLM model, logits were passed in the sample,
which can cause accuracy issues in version 0.15.0.
- vLLM version: v0.15.0
- vLLM main:
83b47f67b1
Signed-off-by: GDzhu01 <809721801@qq.com>
### What this PR does / why we need it?
fix(mtp): resolve MTP core bugs and enhance eager mode test cases
1. Resolved critical issues in eager mode MTP core execution logic;
2. Fixed functional bugs in the _update_states_after_model_execute
function;
3. Updated and released test_mtp_qwen3_next.py to validate eager mode
acceptance rate.
### Does this PR introduce _any_ user-facing change?
None
### How was this patch tested?
- vLLM version: v0.15.0
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.15.0
Signed-off-by: Bowen-Leee <caoshankuangren@gmail.com>
### What this PR does / why we need it?
This PR supports the Kimi-K2.5 models on the NPU of bf16 and w4a8
weights.
The corresponding PR in the vllm community has been merged:
https://github.com/vllm-project/vllm/pull/34501
### Does this PR introduce _any_ user-facing change?
- No.
### How was this patch tested?
We test the Kimi-K2.5 weights. The weights path:
https://modelscope.cn/models/Eco-Tech/Kimi-K2.5-W4A8
Successfully ran on 910B NPU using vllm-ascend by the w4a8 weights.
- vLLM version: v0.15.0
- vLLM main:
9562912cea
---------
Signed-off-by: LoganJane <LoganJane73@hotmail.com>
### What this PR does / why we need it?
This PR fixes a bug in `vllm_ascend/worker/model_runner_v1.py` where the
`@torch.inference_mode` decorator was used without parentheses. Using
the decorator without instantiation is deprecated and may not correctly
disable gradient calculations, leading to performance degradation and
increased memory usage during inference. This change adds the required
parentheses to ensure `torch.inference_mode` is applied correctly.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
The change is a minor syntax correction. Existing CI tests should cover
this.
- vLLM version: v0.15.0
- vLLM main:
9562912cea
Signed-off-by: gcanlin <canlinguosdu@gmail.com>
### What this PR does / why we need it?
1. Following https://github.com/vllm-project/vllm/pull/32322, use the
`memory_profiling` context manager from vllm for profiling.
2. Fix wrong non-torch memory value recorded during profiling, which is
not its peak during inference.
---
**More details about point 2:**
After profling, the non-torch memory value we recorded is lower than
that in real inference. This is mainly because of the different memory
management behaviour between `torch.cuda.empty_cache()` and
`torch.npu.empty_cache()`.
With regard to `torch.cuda.empty_cache()`, it only recycle the unused
memory in pytorch memory pool (i.e., memory managed by pytorch caching
allocator), **with no affect to non-torch memory**. However, as for
`torch.npu.empty_cache()`, it has a totally different memory management
mechanism, i.e., it may call `aclrtSynchronize` and **enable Ascend
runtime to free up non-torch memory**.
Thus, the non-torch memory value we recorded after
`torch.npu.empty_cache()` is much lower than its peak during profling.
Resolution:
We record the peak non-torch memory value
(`non_torch_memory_before_empty_cache`) after profiling, but before
`torch.npu.empty_cache()`. Then, we add the diff
(`non_torch_memory_cleared_by_empty_cache =
non_torch_memory_before_empty_cache - self.non_torch_memory`) to
non-torch memory when calculating available KV cache memory, which will
lead to less KV cache memory (i.e., it's safer to avoid OOM issues).
---
> [!NOTE]
> This PR needs to wait for main2main aligning to latest vllm commit
before merging.
### Does this PR introduce _any_ user-facing change?
no.
### How was this patch tested?
Before this PR, the non-torch memory we used to calculate available KV
cache memory is **0.90 G**, whereas its peak during real inference is
**1.08 G**, diff: **182.00 M**.
After this PR, we add this diff to non-torch memory after profiling and
thus make the profiling results more accurate.
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd
---------
Signed-off-by: shen-shanshan <467638484@qq.com>
### What this PR does / why we need it?
- Keeps enable_cpu_binding default on, but skips binding on non‑ARM CPUs
inside bind_cpus, with a clear log.
- Uses a table-driven binding policy: A3 uses NUMA‑balanced binding;
other device types use NUMA‑affinity binding.
- Updates docs to reflect the exact behavior and adds/updates unit tests
for the new logic.
### Does this PR introduce _any_ user-facing change?
- Yes. CPU binding is now enabled by default via additional_config, and
documented in the user guide.
- CPU binding behavior differs by device type (A3 vs. others).
### How was this patch tested?
Added/updated unit tests:
test_cpu_binding.py
1. test_binding_mode_table covers A2 vs A3 binding mode mapping.
2. test_build_cpu_pools_fallback_to_numa_balanced covers fallback when
affinity info is missing.
3. TestBindingSwitch.test_is_arm_cpu covers ARM/x86/unknown arch
detection.
4. test_bind_cpus_skip_non_arm covers non‑ARM skip path in bind_cpus.
test_worker_v1.py
1. Updated mocks for enable_cpu_binding default True to align with new
config default.
- vLLM version: v0.14.1
- vLLM main: d7de043
---------
Signed-off-by: chenchuw886 <chenchuw@huawei.com>
Co-authored-by: chenchuw886 <chenchuw@huawei.com>
### What this PR does / why we need it?
The ds3.2 model adaptation supports the PCP feature.
The solution is as follows: When saving the KV cache, first perform an
allgather operation on the KVs, and then each node saves its own copy.
When the attention or indexer performs calculations, they all gather the
KV cache and then perform the calculations.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
02/12 23:05:10 - AISBench - INFO - Running 1-th replica of evaluation
02/12 23:05:10 - AISBench - INFO - Task [vllm-api-general-chat/gsm8k]:
{'accuracy': 96.35416666666667, 'type': 'GEN'}
02/12 23:05:10 - AISBench - INFO - time elapsed: 2.87s
02/12 23:05:12 - AISBench - INFO - Evaluation tasks completed.
02/12 23:05:12 - AISBench - INFO - Summarizing evaluation results...
dataset version metric mode vllm-api-general-chat
gsm8kdataset - accuracy gen 96.35
- vLLM version: v0.15.0
- vLLM main:
9562912cea
---------
Signed-off-by: weiguihua2 <weiguihua2@huawei.com>
Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>
### What this PR does / why we need it?
This PR extends the Ascend 310P attention backend to support the
`PrefillCacheHit` state. Previously, only `PrefillNoCache`,
`DecodeOnly`, and `ChunkedPrefill` were supported.
This PR handles this state by routing it to the existing
`forward_chunked_prefill_310` implementation, which is suitable for this
scenario.
The changes also include refactoring the main `forward_impl` dispatch
method for better clarity and updating unit tests to cover the new state
and ensure correctness.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Accuracy test when chunked prefill is disabled.
- vLLM version: v0.15.0
- vLLM main:
9562912cea
---------
Signed-off-by: pu-zhe <zpuaa@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?
<!--
- 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 #
-->
Fix wrong computed_tokens when meet exception. This pull request
addresses a bug in the KV transfer mechanism where an exception during
token lookup operations could lead to an incorrect count of
computed_tokens. By modifying the exception handling in both the lookup
and lookup_scheduler functions to return 0 instead of the start index,
the system now correctly indicates that no tokens were successfully
processed when a remote connection failure occurs. This enhancement
improves the robustness and accuracy of token management within the
vllm_ascend distributed KV pool.
### 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.
-->
NO.
### 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: xleoken <xleoken@163.com>
### What this PR does / why we need it?
#6043 deleted the forward_before phase of the dynamic eplb. Currently,
the end-to-end precision is monitored in the UT, and the log is not
printed in the key place. As a result, the eplb does not take effect and
is not intercepted.
1. The forward_before function is added back.
2. Delete unnecessary logs and add key logs.
3. Warm-up of algorithm 3 is added.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
#### The conversation is normal.
Okay, the user is asking, \"What is deep learning?\" I need to explain
this in a clear and concise way. Let me start by recalling what I know
about deep learning. It's a subset of machine learning, right? So first,
I should mention that it's part of machine learning, which itself is a
branch of AI. Then, the key aspect of deep learning is the use of neural
networks with multiple layers. These are called deep neural
networks.\n\nWait, I should define neural networks first. Maybe start
with the basics. A neural network is inspired by the human brain, with
layers of nodes (neurons) that process data. But deep learning
specifically refers to networks with many layers—hence \"deep.\" So the
term \"deep\" comes from the number of layers. \n\nI should explain how
deep learning works. It involves training these networks on large
datasets, allowing them to automatically learn features from the data.
Unlike traditional machine learning, where you might have to manually
extract features, deep learning models can do this automatically. That's
a key point. For example, in image recognition, a deep learning model
can learn to detect edges, shapes, and then more complex patterns
without human intervention.\n\nApplications are important too. The user
might want to know where deep learning is used. Common examples include
image and speech recognition, natural language processing, autonomous
vehicles, and recommendation systems. Maybe mention specific
technologies like self-driving cars using computer vision or virtual
assistants like Siri or Alexa
- vLLM version: v0.15.0
- vLLM main:
13397841ab
Signed-off-by: shenchuxiaofugui <1311027364@qq.com>
### What this PR does / why we need it?
The incorrect regular expression syntax `.*[UE4M3|ue4m3].*` actually
ignores all words containing any of the following characters: `u, e, 4,
m, 3, |`
```yaml
extend-ignore-identifiers-re = [".*Unc.*", ".*_thw",
".*UE8M0.*", ".*[UE4M3|ue4m3].*", ".*eles.*", ".*fo.*", ".*ba.*",
".*ot.*", ".*[Tt]h[rR].*"]
```
===fix===>
```yaml
extend-ignore-identifiers-re = [".*Unc.*", ".*_thw",
".*UE8M0.*", ".*(UE4M3|ue4m3]).*", ".*eles.*", ".*fo.*", ".*ba.*",
".*ot.*", ".*[Tt]h[rR].*"]
```
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.15.0
- vLLM main:
9562912cea
Signed-off-by: MrZ20 <2609716663@qq.com>
### What this PR does / why we need it?
1. Fix a vec error caused by unaligned UB accesss in the
DispatchFFNCombine;
2. Fix expert_token_nums tensor defined on host instead of NPU in
moe_comm_method.py
3. Fix multi-core copy issue of expert_token_nums in dispatchffnCombine
op (single aiv copy is sufficient)
### Does this PR introduce _any_ user-facing change?
No, this PR does not introduce any user-facing changes. The fix only
addresses internal memory access logic and does not modify any public
APIs, interfaces, or user-visible behaviors.
### How was this patch tested?
`export VLLM_ASCEND_ENABLE_FUSED_MC2=1`
vLLM version: v0.15.0
- vLLM version: v0.15.0
- vLLM main:
9562912cea
Signed-off-by: xulei_ict <xulei292@huawei.com>
Co-authored-by: xulei_ict <xulei292@huawei.com>
NZ Format Support for Linear Layers: Implemented support for the NZ
(N-dimensional Z-order) format for linear layer weights on Ascend 310P,
enhancing performance for both quantized and unquantized layers.
Unquantized Linear Method for Ascend 310P: Introduced
AscendUnquantizedLinearMethod310 to specifically handle and apply NZ
format casting to unquantized linear layer weights during the loading
process.
MRotaryEmbedding Integration: Extended Rotary Embedding support by
adding AscendMRotaryEmbedding310 to provide an Ascend-specific
implementation for MRotaryEmbedding.
Quantization Method Updates: Updated the w8a8_static quantization method
to directly transpose weights and apply NZ format casting, ensuring
consistency with the new format.
- vLLM version: v0.15.0
- vLLM main:
9562912cea
---------
Signed-off-by: Tflowers-0129 <2906339855@qq.com>
### What this PR does / why we need it?
This PR integrates the `npu_add_rms_norm` fused kernel for RMSNorm
operations with residual connections on 310P devices. This change
optimizes the computation by replacing a two-step process (manual
residual addition followed by RMSNorm) with a single, more efficient
fused operation. This is needed to improve the performance of models
utilizing RMSNorm with residual connections on the 310P architecture.
Fixes #
### Does this PR introduce _any_ user-facing change?
No, this PR introduces an internal optimization and does not change any
user-facing APIs or behaviors.
### How was this patch tested?
This patch was tested with updated unit tests
(`test_RMSNorm_forward_310p`) that mock the `npu_add_rms_norm` operation
to verify the correctness of the fused kernel integration.
---------
Signed-off-by: Tflowers-0129 <2906339855@qq.com>
### What this PR does / why we need it?
Integrating inductor pass and npugraph ex pass, see RFC:
https://github.com/vllm-project/vllm-ascend/issues/6347
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
all tests passed.
- vLLM version: v0.14.1
- vLLM main:
dc917cceb8
---------
Signed-off-by: wxsIcey <1790571317@qq.com>
### What this PR does / why we need it?
This PR adapts bugfixes from `norm_quant_fusion_pass` to
`graphex_norm_quant_fusion_pass` for the `npugraph_ex` backend.
The main changes are:
- Replaced `torch.ops.npu.npu_add_rms_norm` with
`torch.ops._C_ascend.npu_add_rms_norm_bias`.
- For patterns without bias, `None` is passed as the bias argument.
- For patterns with bias, the separate `add` operation for bias is
removed and the bias is passed directly to `npu_add_rms_norm_bias`. This
improves fusion.
These changes ensure consistency and correctness for RMSNorm and
quantization fusion patterns when using `npugraph_ex`.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.15.0
- vLLM main:
9562912cea
Signed-off-by: huyuanquan1 <huyuanquan1@huawei.com>
Co-authored-by: huyuanquan1 <huyuanquan1@huawei.com>
### What this PR does / why we need it?
mooncake layerwise support pcp function
PCP (Prefill Context Parallelism) Support: Introduced explicit support
for Prefill Context Parallelism (PCP) and Decode Context Parallelism
(DCP) in the Mooncake layerwise KV cache transfer mechanism, allowing
for more granular control and awareness of parallel configurations
during data transfer.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
By ci
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd
---------
Signed-off-by: wangxiaoteng <wangxiaoteng@huawei.com>
Signed-off-by: liziyu <liziyu16@huawei.com>
Co-authored-by: liziyu <liziyu16@huawei.com>
### What this PR does / why we need it?
refer to https://github.com/vllm-project/vllm-ascend/issues/6391,
Currently adapted the complete process of event publishing in vllm:
* `kv_connector_model_runner_mixin` invoke kv-connector
`get_kv_connector_kv_cache_events` func to collect kvevents
* in `scheduler.py` , it's `update_from_output` func will invoke
`_update_from_kv_xfer_finished` which invoke
`connector.update_connector_output` to collect kv-events from all
kv-worker, and then scheduler will invoke `connector.take_events` api to
collect all kv-events and add it to the events which from
`kv_cache_manager`
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
You can add `--kv-events-config` parameter to the `vllm server` command
to enable this feature.
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd
---------
Signed-off-by: yejj710 <abyss1999@163.com>
Co-authored-by: fems14 <1804143737@qq.com>
### What this PR does / why we need it?
vllm model runner v2 use uva buffer to prepare input data, but npu
doesn't support uva yet, this pr implement a uvawrapper class to mimic
gpu's uva backend. what's more, this pr make some modifications to adapt
to the newer main branch.
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
- vLLM main:
13397841ab
---------
Signed-off-by: Ronald1995 <ronaldautomobile@163.com>
### What this PR does / why we need it?
Fix the issue where, in graph mode, the fused `AddRMSNormQuant` operator
does not take effect when there is no bias.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd
---------
Signed-off-by: ZYang6263 <zy626375@gmail.com>
### What this PR does / why we need it?
This pull request enables the `npugraph_ex` backend by default to
improve performance on Ascend NPUs, as proposed in the
[RFC](https://github.com/vllm-project/vllm-ascend/issues/6214).
### Does this PR introduce _any_ user-facing change?
Yes. `npugraph_ex` is now enabled by default. Users can disable it by
setting `enable: false` in the `npugraph_ex_config` section of the
`additional_config`.
### How was this patch tested?
CI passed. The changes are covered by existing and new E2E tests
(`test_aclgraph_accuracy.py`) and unit tests (`test_ascend_config.py`)
that have been updated to reflect the new default behavior. The tests
verify correctness and consistency with `npugraph_ex` enabled and
disabled, as well as with the new static kernel option.
Signed-off-by: huyuanquan1 <huyuanquan1@huawei.com>
Co-authored-by: huyuanquan1 <huyuanquan1@huawei.com>
### What this PR does / why we need it?
GLM5 adaptation
1. use torch_npu.npu_lightning_indexer for GLM5
2. forbid eagle proposer when fullgraph mode is enabled because of bugs
3. add quatization config for GLM5
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
by ci
- vLLM main:
978a37c823
---------
Signed-off-by: yydyzr <liuyuncong1@huawei.com>
Signed-off-by: shenchuxiaofugui <1311027364@qq.com>
Co-authored-by: shenchuxiaofugui <1311027364@qq.com>
### What this PR does / why we need it?
This PR aims to update `target_probs` to `target_logits` in
`rejection_sample`, for catching up with
https://github.com/vllm-project/vllm/pull/32852. Otherwise, sampling
with temperature will incur accuracy problem where tokens can be
accepted or rejected unreasonably.
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
by ci
- vLLM version: v0.15.0
- vLLM main:
13397841ab
Signed-off-by: Zetong Li <slippersss@126.com>
### What this PR does / why we need it?
This PR extends original `rope_triton_forward` and
`split_qkv_rmsnorm_rope` to support `cos_sin_cache` && `positions` as
inputs. This fully aligns to vLLM RoPE api interface. Compared with
earlier implementation for RoPE, the benefits are:
1. avoiding pre-computation of `cos` `sin` before model execution, which
helps to remove redundant codes.
2. allowing eagle3 draft model to have different rope parameters with
main model (see #6612 ). This help to recover accept rate && accuracy in
that case.
In addition, this kernel change only introduces very small performance
degradation. Those `index_select` or `chunk` operations are now changed
into simple memory access in triton kernel (For example,
https://github.com/vllm-project/vllm-ascend/pull/5450/changes#diff-a4c2d3071530df193b98f9bf38553874bc4d47571336711f116c26d019cfbb6aR77-R81).
**Highlights**
- **RoPE Cache Unification**: Replaced separate _sin and _cos global
tensors with a unified cos_sin_cache and explicit positions tensor for
Rotary Positional Embeddings (RoPE), streamlining data handling.
- **Triton Kernel Integration**: Updated Triton kernels
(split_qkv_rmsnorm_rope_kernel, _triton_rope) to directly consume the
cos_sin_cache and positions for more efficient and integrated RoPE
calculations.
- **Custom Operation Registration**: Registered `rope_forward_oot` as a
new custom operation, allowing its use in fused compilation passes and
providing a dedicated entry point for the new RoPE implementation.
- **Refactored RoPE Forward Pass**: Modified the rope_forward_oot
function to accept the new cos_sin_cache and positions arguments,
enabling a more flexible and integrated RoPE application within the
system.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
5326c89803
Additional test on Qwen3-235b accuracy:
| Aime2024 | GSM8K | Livecodebench |
| -------- | -------- | -------- |
| 83.33 | 96.26 | 70.23 |
---------
Signed-off-by: Angazenn <supperccell@163.com>
## Summary
- Remove unused `set_rotation_config` and `apply_rotation` methods from
`AscendW4A4LaosDynamicLinearMethod`
- Remove unused `rotation_type` field and associated conditional
quantization parameters (`heads_rotation`, `kronecker_rotation_n`,
`kronecker_rotation_m`)
These rotation-related functions and parameters are never called in the
current W4A4 LAOS dynamic quantization workflow.
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd
Signed-off-by: SlightwindSec <slightwindsec@gmail.com>
This pull request introduces a new capability to monitor the health of
NPU cards directly from the Worker class. This enhancement allows for
proactive detection of NPU issues by executing the npu-smi command,
improving system reliability and operational visibility within the
vllm_ascend worker environment.
- vLLM version: v0.15.0
- vLLM main:
13397841ab
---------
Signed-off-by: liziyu <liziyu16@huawei.com>
Signed-off-by: wangxiaoteng <wangxiaoteng@huawei.com>
Signed-off-by: luomin2005 <luomin2005@huawei.com>
Co-authored-by: liziyu <56102866+liziyu179@users.noreply.github.com>
Co-authored-by: wangxiaoteng <wangxiaoteng@huawei.com>
### What this PR does / why we need it?
This PR introduces support for W8A8 dynamic quantization for
Mixture-of-Experts (MoE) models on Ascend 310P devices. This is achieved
by:
- Implementing a new quantization scheme
`AscendW8A8DynamicFusedMoEMethod310`.
- Adding a unified MLP implementation (`unified_apply_mlp`) for 310P
that handles both quantized and unquantized paths.
- Refactoring the MoE and quantization configuration logic to correctly
route to the new 310P-specific implementations.
- Adding new e2e and unit tests to verify the functionality of MoE W8A8
quantization.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
- Added a new e2e test `test_qwen3_moe_tp2_w8a8` to test MoE W8A8
quantization in a multi-card setup.
- Added several new unit tests for the 310P-specific MoE components,
including `experts_selector`, `fused_moe`, `moe_comm_method`, `moe_mlp`,
and the new `w8a8_dynamic` quantization method.
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd
---------
Signed-off-by: pu-zhe <zpuaa@outlook.com>
### What this PR does / why we need it?
To prevent confusion between different QuantType classes, we remove**
QuantType in prepare_finalize.py
- vLLM version: v0.15.0
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.15.0
Signed-off-by: shenchuxiaofugui <1311027364@qq.com>
### What this PR does / why we need it?
1. Currently, eplb registers different attributes for different models,
but these attributes are not actually used. Now, these attributes are
directly deleted.
2. Add some log about eplb.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
#### Deepseek v3.1 chat
Of course! Here is a comprehensive explanation of deep learning, broken
down for clarity.\n\n### The Simple Analogy: A Child Learning to
Recognize a Cat\n\nImagine teaching a child what a cat is. You don't
give them a rulebook with instructions like \"has pointy ears, whiskers,
and a tail.\" Instead, you show them many pictures, saying \"this is a
cat\" or \"this is not a cat.\" The child's brain gradually learns to
identify the complex patterns—the combination of shapes, colors, and
textures—that define \"cat-ness.\"\n\n**Deep learning is essentially
this, but for computers.** It's a method for teaching computers to learn
from examples and recognize patterns directly from data (like images,
sound, or text) without being explicitly programmed with rigid
rules.\n\n---\n\n### The Technical Definition\n\n**Deep Learning is a
subfield of machine learning, which itself is a subfield of artificial
intelligence (AI).** It uses artificial **neural networks** with many
layers (\"deep\" networks) to model and understand complex patterns in
data.\n\nHere are the key concepts in that definition:\n\n1.
**Artificial Intelligence (AI):** The broad science of making machines
smart and capable of performing tasks that typically require human
intelligence.\n2. **Machine Learning (ML):** A subset of AI that gives
computers the ability to learn from data *without* being explicitly
programmed for every single rule.\n3. **Deep Learning (DL):** A
specific, powerful
- vLLM version: v0.15.0
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.15.0
Signed-off-by: shenchuxiaofugui <1311027364@qq.com>
### What this PR does / why we need it?
there are batch invariant ops implemented by triton and ascendc, this pr
aims to choose which kind of ops to be used to enable batch invariant.
#5487
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd
---------
Signed-off-by: Ronald1995 <ronaldautomobile@163.com>
### What this PR does / why we need it?
This PR upgrades the core vLLM dependency to a newer version from the
main branch (`13397841ab469cecf1ed425c3f52a9ffc38139b5`). This is
necessary to keep our project up-to-date with the latest features and
fixes from upstream vLLM.
1.
ac32e66cf9
pass file is moved.
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd
---------
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
Signed-off-by: wxsIcey <1790571317@qq.com>
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
Co-authored-by: wxsIcey <1790571317@qq.com>
### What this PR does / why we need it?
This PR fixes an `AttributeError: 'Parameter' object has no attribute
'data'` that occurs when MLAPO is enabled with vLLM v0.15.0.
The error is caused by a monkey-patch on
`MLAAttention.process_weights_after_loading` which is incompatible with
changes in vLLM v0.15.0. This is likely related to PyTorch's deprecation
of the `.data` attribute on `torch.nn.Parameter` objects.
This change makes the monkey-patch conditional, so it is not applied for
vLLM v0.15.0 and newer versions, resolving the crash.
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd
Signed-off-by: Meihan-chen <jcccx.cmh@gmail.com>
### What this PR does / why we need it?
The speculative inference acceptance rate decreases after the vllm
version is upgraded to v0.15.0. This issue is resolved.
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
UT and tests case
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd
---------
Signed-off-by: lilinsiman <lilinsiman@gmail.com>
### What this PR does / why we need it?
For deepseek v3.2, DSA use FullAttentionSpec, allocate 2 * mla page size
bytes, and we use half of that for k cache in DSA
However, the actual proportion of k cache is not high, which results in
a large amount of kvcache being wasted. The proportion of discarded
kvcache is (576-128)/(576 x 2) = 0.388.
Run the same script to start DeepSeek V3.2 on a single A3 server. The
following shows the comparison of kvcache usage:
Before refactoring
```
[kv_cache_utils.py:1307] GPU KV cache size: 15,872 tokens
```
After refactoring
```
[kv_cache_utils.py:1307] GPU KV cache size: 25,984 tokens
```
This pull request refactors the KV cache allocation for Deepseek v3.2
models that use sparse attention. It replaces the use of
`FullAttentionSpec` with `MLAAttentionSpec` and introduces a more
principled way of calculating KV cache tensor split factors based on
model configuration.
This change removes hardcoded values and correctly sizes the cache
tensors, leading to optimized memory usage and improved code
maintainability.
### Does this PR introduce _any_ user-facing change?
No, this is an internal optimization and does not introduce any
user-facing changes.
### How was this patch tested?
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd
---------
Signed-off-by: Wang Kunpeng <1289706727@qq.com>
### What this PR does / why we need it?
This PR adds DCP support to the SFA backend.
Please note that due to operator constraints, the current implementation
has to all-gather the entire KV cache and modify the block table to
satisfy the operator input requirements. This results in significantly
increased communication overhead and peak memory usage. Therefore, this
is only a temporary workaround and will be refactored once the operator
provides proper support.
Additionally, because of the above limitations,
`cp_kv_cache_interleave_size` is currently required to be equal to
`block_size`. This restriction will also be removed after the refactor.
#### Test
accuracy test using DeepSeek-V3.2-Exp-W8A8 with dp2tp8dcp8
| dataset | version | metric | mode | vllm-api-general-stream |
|----- | ----- | ----- | ----- | -----|
| gsm8kdataset | - | accuracy | gen | 96.35 |
- vLLM version: v0.15.0
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.15.0
---------
Signed-off-by: QiuChunshuo <qiuchunshuo@huawei.com>
### What this PR does / why we need it?
Issue: If a model such as Ling-1T adopts partial rotary position
embedding (partial RoPE), but config.json uses the rotary_dim parameter
instead of partial_rotary_factor, it will trigger a RuntimeError: The
expanded size of the tensor (128) must match the existing size (64) at
non-singleton dimension 3.
<img width="1681" height="472" alt="image"
src="https://github.com/user-attachments/assets/ba03d7df-ecba-4d6f-9ec1-4dc55f59799e"
/>
This PR addresses an issue where models using partial rotary position
embedding (partial RoPE) with the `rotary_dim` parameter in
`config.json` (instead of `partial_rotary_factor`) would encounter a
`RuntimeError`.
This change adds support for the `rotary_dim` parameter in
`vllm_ascend/ops/rotary_embedding.py` to correctly calculate the
`rope_dim`, resolving the tensor size mismatch error.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
The patch was tested successfully with the Ling-1T model, which
previously triggered the error.
- vLLM version: v0.15.0
- vLLM main:
d7e17aaacd
Signed-off-by: GoCHug <93277779+GoCHug@users.noreply.github.com>
### What this PR does / why we need it?
Part of #5304.
After https://github.com/vllm-project/vllm/pull/32523 merge, we could
remove the patch of `MiniCPMAttention`.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
Test it locally.
- vLLM version: v0.13.0
- vLLM main:
2c24bc6996
---------
Signed-off-by: gcanlin <canlinguosdu@gmail.com>
### What this PR does / why we need it?
layerwise connector support recompute scheduler.
NOTE：
Triggering recompute will invoke the tokenizer again, which may lead to
precision fluctuations.
[RFC]: CDCP Scheduling for Disaggregated Prefilling with KV Cache
Layerwise Push Support
https://github.com/vllm-project/vllm-ascend/issues/4842
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
bde38c11df
---------
Signed-off-by: liziyu <liziyu16@huawei.com>
Signed-off-by: wangxiaoteng <wangxiaoteng@huawei.com>
Co-authored-by: wangxiaoteng <wangxiaoteng@huawei.com>
