### What this PR does / why we need it?
This PR adds missing arguments in `AscendRotaryEmbedding`,
`AscendYarnRotaryEmbedding` to conform with vLLM. Besides, corresponding
ut is introduced.
- vLLM version: v0.17.0
- vLLM main:
4034c3d32e
---------
Signed-off-by: Angazenn <supperccell@163.com>
### What this PR does / why we need it?
Replace the '_npu_flash_attention_unpad' operator with the
'npu_fusion_attention' operator to ensure that the Qwen VL model can run
in the A5 environment and remove the 'mrope' operator call restriction
for A5.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.16.0
- vLLM main:
4034c3d32e
Signed-off-by: 汪越 <wangyue361@h-partners.com>
### What this PR does / why we need it?
This PR aims to support aclgraph for model runner v2, please see RFC
#5208. The PR contains these modifications:
- adapt to newest commit of vllm main branch.
- supply a unified interface of extra forward context for both model
runner v1 and model runner v2.
- implement graph mode for main model.
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
- vLLM version: v0.16.0
- vLLM main:
4034c3d32e
---------
Signed-off-by: Ronald1995 <ronaldautomobile@163.com>
What this PR does / why we need it?
When using a draft model (e.g., in MTP speculative decoding) with shared
expert data parallelism (enabled via flashcomm), a shape mismatch error
occurs in the rotary embedding calculation for models like GLM-4.7. This
is because the positions tensor has an incorrect shape for this specific
configuration.
This PR fixes the issue by adding a check in
AscendRotaryEmbedding.forward_oot. If the model is a draft model and
shared expert DP is enabled, it processes the positions tensor using
torch.ops.vllm.maybe_all_gather_and_maybe_unpad to ensure its shape is
correct before applying the rotary embedding. This resolves the shape
mismatch error.
- vLLM version: v0.16.0
- vLLM main:
15d76f74e2
---------
Signed-off-by: Zhu Jiyang <zhujiyang2@huawei.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>
### 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?
This PR removes the custom `rotary_embedding` operator and its
associated C++ kernel implementation, PyTorch bindings, and tests.
The codebase now falls back to using the native
`torch_npu._npu_rotary_embedding` implementation. This change simplifies
the codebase by removing custom, platform-specific kernel code and
relying on the standard NPU library implementation, which is presumably
more optimized and easier to maintain.
### Does this PR introduce _any_ user-facing change?
No. This is an internal refactoring and does not introduce any
user-facing changes.
### How was this patch tested?
The tests for the custom `rotary_embedding` operator have been removed
along with the operator itself. The correctness of the fallback to the
native `torch_npu` implementation is verified by existing CI tests for
attention layers and models that use rotary embeddings.
- vLLM version: v0.15.0
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.15.0
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
### What this PR does / why we need it?
This pull request integrates comprehensive support for Mixture of
Experts (MoE) models on the Ascend 310P device within the vllm-ascend
framework. It achieves this by introducing specialized modules for
expert selection, fused MoE layers, and optimized all-gather
communication. The changes also refine existing NPU operations, making
them more consistent and efficient for 310P, ultimately enhancing the
performance and compatibility of MoE models on this hardware.
Highlights
310P MoE Support: Introduces dedicated implementations for Mixture of
Experts (MoE) models on Ascend 310P devices, including new modules for
expert selection, fused MoE layers, and communication.
All-Gather Communication: Enforces the use of ALLGATHER communication
for MoE operations on 310P, optimizing data transfer and leveraging
NPU-specific token dispatching.
Simplified NPU Operations: Removes conditional type casting for
npu_swiglu and enables custom rotary embedding kernels unconditionally,
suggesting improved native support for 310P.
New MoE Classes Registered: Registers AscendFusedMoE310 and
AscendSharedFusedMoE310 to integrate 310P-specific MoE layers into the
system's custom operation registry.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
offline test and server test, with qwen3-30b-a3b,tp/ep 4 on 310p
- vLLM version: v0.15.0
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.15.0
---------
Signed-off-by: pu-zhe <zpuaa@outlook.com>
### What this PR does / why we need it?
Improve Triton stability on Ascend for large grids
set `TRITON_ALL_BLOCKS_PARALLEL=1` when grids > 65535
- vLLM version: v0.14.1
- vLLM main:
dc917cceb8
Signed-off-by: hfadzxy <starmoon_zhang@163.com>
### What this PR does / why we need it?
change self.head_size to self.rotary_dim.
only the rotary part is processed here, the dimension should be rotary_dim.
Fix bug #6060
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Only a small section of code was modified to adjust the parameters, and
all standard tests were passed.
- vLLM version: v0.14.1
- vLLM main:
dc917cceb8
Signed-off-by: fengshi666 <fengshi666@adsl-99-12-210-25.dsl.hstntx.sbcglobal.net>
Co-authored-by: fengshi666 <fengshi666@adsl-99-12-210-25.dsl.hstntx.sbcglobal.net>
### What this PR does / why we need it?
Drop vLLM 0.13.0 support, upgrade to 0.14.0
- vLLM version: v0.13.0
- vLLM main:
d68209402d
---------
Signed-off-by: hfadzxy <starmoon_zhang@163.com>
### What this PR does / why we need it?
When running the Qwen2.5-Omni-7B model on Ascend NPU, the engine fails
during the profiling/warmup stage with the following error:
`AclNN_Runtime_Error(EZ9903): rtKernelLaunchWithHandleV2 failed: 507035.
The vector core execution is abnormal.`
error log:
https://github.com/vllm-project/vllm-ascend/actions/runs/21144534911/job/60806765393#step:17:6412
This error is specifically triggered by the `triton_mrope` kernel when
handling the unique `mrope_section` configurations of the Omni model.
Other multimodal models with standard sections (e.g., [16, 24, 24]) or
standard LLMs work correctly with Triton.
Modified vllm_ascend/ops/rotary_embedding.py to add a conditional check
before calling forward_triton.
1. For standard LLMs (mrope_interleaved = True ), it continues to use
Triton for acceleration.
2. For complex configurations (like Qwen2.5-Omni mrope_interleaved =
False ), it now falls back to the native super().forward_oot() path,
which uses the stable torch_npu or PyTorch implementation.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
d68209402d
Signed-off-by: hfadzxy <starmoon_zhang@163.com>
### What this PR does / why we need it?
1. Implement a **high-performance Triton custom kernel** for the rotary
position embedding (RoPE) operator on **Ascend NPU** platform
2. Fix critical bugs in the Triton RoPE kernel registration and
invocation process: including incorrect fake impl function name
matching, wrong torch ops namespace for kernel call, missing self
parameter in cos/sin slice fetching, and syntax errors in function type
annotations.
3. Achieve **extreme performance optimization** for the core RoPE
operator: the single inference latency is reduced from **57.1 μs** to
**9 μs**, with **6.34x performance improvement** and **84.24% latency
reduction**.
4. The RoPE operator is a **hot path** that is executed in every
transformer layer during LLM inference, the optimization will directly
reduce the overall inference latency and improve the throughput of LLM
serving on Ascend NPU.
5. Keep full backward compatibility: the Triton kernel is enabled only
when `HAS_TRITON=True`, and automatically fall back to the original
Ascend NPU native implementation if Triton is not available, no
functional regression.
### Does this PR introduce _any_ user-facing change?
**NO**
- No changes to any public APIs, interfaces or inference behaviors of
vLLM.
- No impact on the text generation quality and correctness of the large
model.
- The optimization is transparent to end users, only the inference speed
(latency/throughput) is improved without any functional change.
### How was this patch tested?
1. **Environment Validation**: Tested on Ascend NPU platform with
vLLM-Ascend framework, Triton library installed and enabled
(`HAS_TRITON=True`).
2. **Kernel Registration Test**: Verified the Triton RoPE kernel
(`rope_forward_triton`) is successfully registered to
`torch.ops._C_ascend` namespace without any
`ValueError/NameError/SyntaxError`.
3. **Functional Correctness Test**: Run large model (GLM4/MoE) inference
on the Ascend NPU platform, the generated text content is **completely
correct** (no garbled text, no logical errors), consistent with the
original implementation.
4. **Performance Benchmark Test**: Measure the single execution latency
of the RoPE operator before/after optimization, confirm the latency is
stably reduced from 57.1 μs to 9 μs, the performance gain is valid and
stable.
5. **Fallback Mechanism Test**: Manually disable Triton
(`HAS_TRITON=False`), verify the code correctly falls back to the
original Ascend NPU native RoPE implementation, no service crash and
normal inference.
6. **Compatibility Test**: Test with different tensor shapes/sizes of
query/key, all cases work correctly with the Triton kernel, no shape
mismatch error.
- operator supply by Hexiang Wang
- vLLM version: v0.13.0
- vLLM main:
11b6af5280
---------
Signed-off-by: ZCG12345 <2097562023@qq.com>
### What this PR does / why we need it?
- Use upstream util function (`_pre_process()` and `_post_process()`) to
reduce redundant codes. (Find more details at
https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/layers/rotary_embedding/common.py#L184-L213)
- Merge Q/K split to simplify the logic of calling
`torch_npu.npu_rotary_mul()` for better performance (TPOT has been
reduced by **6.22%**).
### Does this PR introduce _any_ user-facing change?
no.
### How was this patch tested?
#### ✅ Functional test
Launch the server:
```bash
export VLLM_USE_MODELSCOPE=True
vllm serve /root/.cache/modelscope/hub/models/Qwen/Qwen2.5-VL-7B-Instruct \
--dtype bfloat16 \
--limit-mm-per-prompt '{"image": 1}' \
--max-model-len 16384 \
--max-num-batched-tokens 16384
```
Query the server:
```bash
curl -X POST http://localhost:8000/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "/root/.cache/modelscope/hub/models/Qwen/Qwen2.5-VL-7B-Instruct",
"messages": [
{"role": "system", "content": "You are a helpful assistant."},
{"role": "user", "content": [
{"type": "image_url", "image_url": {"url": "https://modelscope.oss-cn-beijing.aliyuncs.com/resource/qwen.png"}},
{"type": "text", "text": "What is the text in the illustrate? How does it look?"}
]}
],
"max_tokens": 100
}'
```
Output:
```
{"id":"chatcmpl-b2911ab6989ef098","object":"chat.completion","created":1768202780,"model":"/root/.cache/modelscope/hub/models/Qwen/Qwen2.5-VL-7B-Instruct","choices":[{"index":0,"message":{"role":"assistant","content":"The text in the illustration is \"TONGYI Qwen.\" The word \"TONGYI\" is written in blue, and \"Qwen\" is written in gray. The text appears to be part of a logo or branding design, with \"TONGYI\" being more prominent and \"Qwen\" being slightly smaller and positioned below it. The font style is modern and clean, with \"TONGYI\" having a slightly bolder appearance compared to \"Qwen.\"","refusal":null,"annotations":null,"audio":null,"function_call":null,"tool_calls":[],"reasoning":null,"reasoning_content":null},"logprobs":null,"finish_reason":"length","stop_reason":null,"token_ids":null}],"service_tier":null,"system_fingerprint":null,"usage":{"prompt_tokens":78,"total_tokens":178,"completion_tokens":100,"prompt_tokens_details":null},"prompt_logprobs":null,"prompt_token_ids":null,"kv_transfer_params":null}
```
#### ✅ Benchmark
Run:
```bash
export VLLM_USE_MODELSCOPE=False
export HF_ENDPOINT="https://hf-mirror.com"
vllm bench serve \
--model /root/.cache/modelscope/hub/models/Qwen/Qwen2.5-VL-7B-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 10 \
--no-stream
```
Before this PR:
```
============ Serving Benchmark Result ============
Successful requests: 10
Failed requests: 0
Benchmark duration (s): 5.96
Total input tokens: 7191
Total generated tokens: 996
Request throughput (req/s): 1.68
Output token throughput (tok/s): 167.05
Peak output token throughput (tok/s): 261.00
Peak concurrent requests: 10.00
Total token throughput (tok/s): 1373.16
---------------Time to First Token----------------
Mean TTFT (ms): 964.43
Median TTFT (ms): 858.48
P99 TTFT (ms): 1691.45
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms): 63.08
Median TPOT (ms): 40.86
P99 TPOT (ms): 241.30
---------------Inter-token Latency----------------
Mean ITL (ms): 40.16
Median ITL (ms): 33.61
P99 ITL (ms): 250.30
==================================================
```
After this PR:
```
============ Serving Benchmark Result ============
Successful requests: 10
Failed requests: 0
Benchmark duration (s): 5.71
Total input tokens: 7191
Total generated tokens: 996
Request throughput (req/s): 1.75
Output token throughput (tok/s): 174.45
Peak output token throughput (tok/s): 279.00
Peak concurrent requests: 10.00
Total token throughput (tok/s): 1433.95
---------------Time to First Token----------------
Mean TTFT (ms): 992.14
Median TTFT (ms): 938.30
P99 TTFT (ms): 1728.71
-----Time per Output Token (excl. 1st token)------
Mean TPOT (ms): 59.16
Median TPOT (ms): 37.65
P99 TPOT (ms): 234.89
---------------Inter-token Latency----------------
Mean ITL (ms): 36.55
Median ITL (ms): 30.73
P99 ITL (ms): 170.72
==================================================
```
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
---------
Signed-off-by: shen-shanshan <467638484@qq.com>
### What this PR does / why we need it?
this pr support use triton mrope like cuda_forward, which performance is
equal to ascendc ops
this triton ops should use cann 8.5.0
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
test in qwen3-vl-235b acc textvqa
native 81.82
npu triton 81.58
cuda triton 81.52
- vLLM version: v0.13.0
- vLLM main:
2f4e6548ef
Signed-off-by: shiyuan680 <917935075@qq.com>
### What this PR does / why we need it?
In scenarios where models like
[Moonlight](https://modelscope.cn/models/moonshotai/Moonlight-16B-A3B-Instruct)
(using MLA but without `rope_scaling` in config.json) invoke
`AscendRotaryEmbedding`. `_cos_cache` and `_sin_cache` are not recorded
correctly.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
- vLLM version: v0.13.0
- vLLM main:
45c1ca1ca1
Signed-off-by: Debonex <719893090@qq.com>
### What this PR does / why we need it?
Following https://github.com/vllm-project/vllm/pull/29873, register
`AscendApplyRotaryEmb` CustomOp and remove related patch.
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
#### ✅ Test Qwen2.5-VL
Run:
```bash
vllm serve /root/.cache/modelscope/hub/models/Qwen/Qwen2.5-VL-7B-Instruct \
--max_model_len 16384
```
Output:
```
{"id":"chatcmpl-b02c1ff3415d2462","object":"chat.completion","created":1766129265,"model":"/root/.cache/modelscope/hub/models/Qwen/Qwen2.5-VL-7B-In struct","choices":[{"index":0,"message":{"role":"assistant","content":"The text in the illustration is \"TONGYI Qwen.\" The word \"TONGYI\" is writ ten in blue, and \"Qwen\" is written in gray. The text appears to be part of a logo or branding design.","refusal":null,"annotations":null,"audio": null,"function_call":null,"tool_calls":[],"reasoning":null,"reasoning_content":null},"logprobs":null,"finish_reason":"stop","stop_reason":null,"tok en_ids":null}],"service_tier":null,"system_fingerprint":null,"usage":{"prompt_tokens":78,"total_tokens":129,"completion_tokens":51,"prompt_tokens_d
```
#### ✅ Test Qwen3-VL
Run:
```bash
vllm serve /root/.cache/modelscope/hub/models/Qwen/Qwen3-VL-8B-Instruct \
--max_model_len 16384
```
Output:
```
{"id":"chatcmpl-a3a7de5a900a9321","object":"chat.completion","created":1766129586,"model":"/root/.cache/modelscope/hub/models/Qwen/Qwen3-VL-8B-Instruct","choices":[{"index":0,"message":{"role":"assistant","content":"The text in the illustration is **“TONGYI Qwen”**.\n\n### How it looks:\n- **“TONGYI”** is written in **uppercase letters** in a **bold, modern sans-serif font**, colored **blue**.\n- **“Qwen”** is written in **lowercase letters** in a **slightly thinner, elegant sans-serif font**, colored **dark gray**.\n- The two lines of text are stacked vertically, with “TONG","refusal":null,"annotations":null,"audio":null,"function_call":null,"tool_calls":[],"reasoning":null,"reasoning_content":null},"logprobs":null,"finish_reason":"length","stop_reason":null,"token_ids":null}],"service_tier":null,"system_fingerprint":null,"usage":{"prompt_tokens":112,"total_tokens":212,"completion_tokens":100,"prompt_tokens_details":null},"prompt_logprobs":null,"prompt_token_ids":null,"kv_transfer_params":null}
```
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c
---------
Signed-off-by: shen-shanshan <467638484@qq.com>
### What this PR does / why we need it?
This PR add additional check on creating global `_cos` and `_sin`, avoid
creating them when using `mrope` or encoder-decoder model.
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c
Signed-off-by: Angazenn <supperccell@163.com>
### What this PR does / why we need it?
Rename `_910B` to `A2`;
Rename `_910_93` to `A3`;
Rename `_910_95` to `A5`;
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c
Signed-off-by: zzzzwwjj <1183291235@qq.com>
### What this PR does / why we need it?
This PR add `qkv_rmsnorm_rope` operator and introduces a graph fusion
pass for `qknorm_rope` operations. The implementation includes a new
configuration flag, a pattern matching pass using
`torch._inductor.pattern_matcher`, and a custom Triton kernel for the
fused operation.
Co-authored-by: Angazenn
[supperccell@163.com](mailto:supperccell@163.com)
### Does this PR introduce _any_ user-facing change?
Yes, add new additional_config
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c
---------
Signed-off-by: wxsIcey <1790571317@qq.com>
### What this PR does / why we need it?
vllm-ascend support Ascend950 with Qwen dense model
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c
---------
Signed-off-by: wangyao <iwangyao@outlook.com>
Co-authored-by: weijinqian0 <1184188277@qq.com>
### What this PR does / why we need it?
Adds W4A16 quantization method for the Kimi-K2-Thinking model and
updates relevant modules to support the new quantization method.
- Implements complete W4A16 quantization method including weight
packing/unpacking, per-group quantization parameter generation,
post-processing logic and MoE method application.
- Adds parameters `use_int4_w4a16`, `w1_offset` and `w2_offset`, adjusts
`with_quant` conditional logic to support W4A16 matrix multiplication.
- Adds `packed_modules_model_mapping` for Kimi-K2-Thinking model and
processing logic for `weight_packed` field.
- vLLM version: v0.12.0
- vLLM main:
ad32e3e19c
---------
Signed-off-by: zhoux77899 <zhouxiang100@huawei.com>
Signed-off-by: Ruri <33858552+zhoux77899@users.noreply.github.com>
Signed-off-by: Ruri <zhouxiang100@huawei.com>
### What this PR does / why we need it?
Qwen2.5-VL mrope precision problem would been solved once this pr is
merged
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
Test on G8600 with textVQA dataset
- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2
---------
Signed-off-by: 李少鹏 <lishaopeng21@huawei.com>
Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>
### What this PR does / why we need it?
The main goal of this PR to alleviate the high maintenance burden from
model duplication when we are going to do the model optimization. Some
of our optimized models diverges a little from the vllm's modeling, but
needs to rewrite several part of original one, brings negligible
maintenance bruden to the vllm-ascend.In order to solve that, we propose
to leverage `torch.compile` and `inductor pattern matcher`,
automatically fuse the pattern we want to merge. For more details can
refer to the RFC https://github.com/vllm-project/vllm-ascend/issues/4239
This pr integrates `AddRMSNorm` and the `Quant` operator, which can
improve the inference speed of models using `w8a8 `quantization.
### Does this PR introduce _any_ user-facing change?
Yes, add new additional_config
### How was this patch tested?
```python
def main():
prompts = [
"The president of the United States is Mr.",
]
# Create a sampling params object.
sampling_params = SamplingParams(max_tokens=100, temperature=0.6, top_k=40, top_p=0.95)
# Create an LLM.
llm = LLM(
model="/root/.cache/modelscope/hub/models/vllm-ascend/Qwen3-8B-W8A8",
# enforce_eager=True,
tensor_parallel_size=1,
trust_remote_code=True,
gpu_memory_utilization=0.7,
quantization="ascend",
)
# Generate texts from the prompts.
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}")
```
```text
Prompt: 'The president of the United States is Mr.', Generated text: ' Trump. The president of the United States is Mr. Biden. Which of the following statements is correct? \n\nA. Mr. Trump is Mr. Biden. \nB. Mr. Trump is not Mr. Biden. \nC. The president of the United States is not Mr. Trump. \nD. The president of the United States is not Mr. Biden.\n\nThe question presents a contradiction: it states that "The president of the United States is Mr. Trump" and "The president of'
```
- vLLM version: 86e178f7c4d8c3b0eaf3c8e3f810a83f63b90e24
- vLLM main:
86e178f7c4
---------
Signed-off-by: Icey <1790571317@qq.com>
Signed-off-by: wxsIcey <1790571317@qq.com>
### What this PR does / why we need it?
Currently, there are two paths to judge the chip type in code,
`get_ascend_soc_version` use `get_soc_version` api in torch_npu, and
`is_310p` `use _build_info.__soc_version__`, which generate when
install. We need to unify the two paths.
We need to unify these codes based on the following points:
1. We need to ensure consistency in chip type judgment between compiling
and running states;
2. In compiling state, we need chip type to complete op's compilation,
but in running state, we only need device
type(910B/910_93/310P/910_95/etc) to make code branch judgement;
3. In compiling state, torch_npu may not have been installed yet, so we
can't use torch_npu's api.
Based on the above points, we have made the following changes:
1. When user set env `SOC_VERSION`, use it; when not set, query
soc_version by `npu-smi`;
2. generate device_type based on soc_version when compiling, and write
`__device_type__` instead of `__soc_version__` in `_build_info.py`;
3. In running state, use `__device_type__` to judge code branch.
### Does this PR introduce _any_ user-facing change?
When not set env `SOC_VERSION`, it will not be `ASCEND910B1` by default,
we will query soc_version by `npu-smi`. And env `SOC_VERSION` must be in
the list `soc_to_device` in `setup.py`.
- vLLM version: v0.11.0
- vLLM main:
2918c1b49c
Signed-off-by: zzzzwwjj <1183291235@qq.com>
### What this PR does / why we need it?
avoid mrope fusion op when running qwen2.5-vl on a+x machine
### Does this PR introduce _any_ user-facing change?
no
### How was this patch tested?
Test text VQA accuracy on G8600 with aisbench
- vLLM version: v0.11.0
- vLLM main:
2918c1b49c
---------
Signed-off-by: 李少鹏 <lishaopeng21@huawei.com>
This reverts commit 646c1db5d7.
this new ops may lead accuracy problem
### What this PR does / why we need it?
### Does this PR introduce _any_ user-facing change?
### How was this patch tested?
- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0
### What this PR does / why we need it?
YaRN scaling is used to improve long seq accuracy for models like Qwen3.
In vLLM, YaRN scaling refers to `YaRNScalingRotaryEmbedding` class which
inherits from original `RotaryEmbedding`. Although
`YaRNScalingRotaryEmbedding` does not rewrite the `forward` function of
`RotaryEmbedding` , using YaRN on npu still run into the native
implementation of foward in `RotaryEmbedding`, rather than forward_oot
in vLLM-Ascend. Thus I register another custom op here to enable the oot
implementation for YaRN in vLLM-Ascend, similar to #3151 .
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
- vLLM version: v0.11.0rc3
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.0
---------
Signed-off-by: Angazenn <supperccell@163.com>
**Background:**
There are two principles about operator registration in PyTorch
- The same namespace can be only registered once by `TORCH_LIBRARY`
- The operator signatures can be only registered once by `def`
Considering that all custom operators defined in the current repo are
only used by Ascend, instead of defining a common operator schema by
vLLM, all accelerators then follow this operator schema and complete the
implementation based on their respective hardware, which is conducive to
functional abstraction.
Therefore, we can rename the operator registration namespace to an
Ascend-specific namespace(**_C_ascend**).
Related ISSUE: https://github.com/vllm-project/vllm-ascend/issues/2742
- vLLM version: main
- vLLM main:
f592b3174b
Signed-off-by: FFFrog <ljw1101.vip@gmail.com>
### What this PR does / why we need it?
`torch_npu.npu_apply_rotary_pos_emb` only support head_size and
rotary_dim equal 128. Error occurs when running GLM
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
- vLLM version: main
- vLLM main:
404c85ca72
Signed-off-by: realliujiaxu <realliujiaxu@163.com>
### What this PR does / why we need it?
Optimize rope by caching sin and cos at the first layer in Qwen Models.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
CI passed with new added/existing test.
- vLLM version: v0.10.1.1
- vLLM main:
562663a044
---------
Signed-off-by: MengqingCao <cmq0113@163.com>
Signed-off-by: ZYang6263 <zy626375@gmail.com>
Signed-off-by: rjg-lyh <1318825571@qq.com>
Co-authored-by: Mengqing Cao <cmq0113@163.com>
Co-authored-by: ZYang6263 <51255902183@stu.ecnu.edu.cn>
Co-authored-by: ZYang6263 <zy626375@gmail.com>
### What this PR does / why we need it?
The current implementation will result in duplicate generation of
`sin_cos_cache` in rope when `kv_seqlen` > 4k, because the
initialization length of the `sin_cos_cache` is only 4k.
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
After this PR merged, sin_cos_cache will not increase in forward func,
so `test_native_rope_deepseek_forward_cache_handling` is not necessary.
- vLLM version: v0.10.1.1
- vLLM main:
60f0843ef8
Signed-off-by: zzzzwwjj <1183291235@qq.com>
### What this PR does / why we need it?
Fix long context seq accuracy problem for `GLM4.5`.
When `max_tokens=1000`, there is cyclic output problem like:
```bash
00 00 00 00 00 00 00 00 00 00 00 00 00 00
```
### Does this PR introduce _any_ user-facing change?
No.
### How was this patch tested?
```python
import os
os.environ["VLLM_USE_MODELSCOPE"] = "True"
os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
from vllm import LLM, SamplingParams
def main():
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=1000, temperature=0.0)
# Create an LLM.
llm = LLM(model="/root/.cache/modelscope/hub/models/ZhipuAI/GLM-4___5",
tensor_parallel_size=8,
enforce_eager=True,
trust_remote_code=True,
max_model_len=1024)
# Generate texts from the prompts.
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}")
if __name__ == "__main__":
main()
```
- vLLM version: v0.10.1.1
- vLLM main:
0235103cbb
---------
Signed-off-by: Shanshan Shen <87969357+shen-shanshan@users.noreply.github.com>
Signed-off-by: shen-shanshan <467638484@qq.com>
Clean up useless code which is only used for torchair in rotary_embedding
- vLLM version: v0.10.1.1
- vLLM main:
a344a5aa0a
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
### What this PR does / why we need it?
After moved torchair related rope ops into torchair_ops, split the
torchair from the origin rope ops to make the code clean.
### Does this PR introduce _any_ user-facing change?
No
### How was this patch tested?
vLLM version: main
vLLM main:
ab9f2cfd19
- vLLM version: v0.10.1.1
- vLLM main:
81eea3d348
Signed-off-by: hust17yixuan <303660421@qq.com>
### What this PR does / why we need it?
Register RotaryEmbedding instead of overwrite forward
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
CI passed with new added/existing test.
- vLLM version: v0.10.0
- vLLM main:
808d2e9aa0
---------
Signed-off-by: Icey <1790571317@qq.com>
Signed-off-by: wxsIcey <1790571317@qq.com>
### What this PR does / why we need it?
This PR supports torchair graph mode with non-mla backend on both 800IA2
and 300I Duo platforms. The main change is to add
`attention_v1_torchair.py` to support specific attention related
operations that are required by torchair.
### Does this PR introduce _any_ user-facing change?
Before this PR, vLLM-Ascend only allows deepseek to use torchair. Now we
can also use it with pangu. Besides, we add a support model list to
control which type of models that can use torchair.
### How was this patch tested?
We have test it with PanguProMoE on both 800IA2 and 300I Duo platforms,
and model generates answer normally.
---------
Signed-off-by: angazenn <zengyanjia@huawei.com>
Signed-off-by: tianyitang <tangtianyi4@huawei.com>
Co-authored-by: angazenn <zengyanjia@huawei.com>
Co-authored-by: tianyitang <tangtianyi4@huawei.com>
### What this PR does / why we need it?
Add initial experimental support for Ascend 310P, this patch squash
below PR into one to help validation:
- https://github.com/vllm-project/vllm-ascend/pull/914
- https://github.com/vllm-project/vllm-ascend/pull/1318
- https://github.com/vllm-project/vllm-ascend/pull/1327
### Does this PR introduce _any_ user-facing change?
User can run vLLM on Altlas 300I DUO series
### How was this patch tested?
CI passed with:
- E2E image build for 310P
- CI test on A2 with e2e test and longterm test
- Unit test missing because need a real 310P image to have the test,
will add in a separate PR later.
- Manually e2e test:
- Qwen2.5-7b-instruct, Qwen2.5-0.5b, Qwen3-0.6B, Qwen3-4B, Qwen3-8B:
https://github.com/vllm-project/vllm-ascend/pull/914#issuecomment-2942989322
- Pangu MGoE 72B
The patch has been tested locally on Ascend 310P hardware to ensure that
the changes do not break existing functionality and that the new
features work as intended.
#### ENV information
CANN, NNAL version: 8.1.RC1
> [!IMPORTANT]
> PTA 2.5.1 version >= torch_npu-2.5.1.post1.dev20250528 to support NZ
format and calling NNAL operators on 310P
#### Code example
##### Build vllm-ascend from source code
```shell
# download source code as vllm-ascend
cd vllm-ascend
export SOC_VERSION=Ascend310P3
pip install -v -e .
cd ..
```
##### Run offline inference
```python
from vllm import LLM, SamplingParams
prompts = ["水的沸点是100摄氏度吗？请回答是或者否。", "若腋下体温为38摄氏度，请问这人是否发烧？请回答是或者否。",
"水的沸点是100摄氏度吗？请回答是或者否。", "若腋下体温为38摄氏度，请问这人是否发烧？请回答是或者否。"]
# Create a sampling params object.
sampling_params = SamplingParams(temperature=0.0, top_p=0.95, max_tokens=10)
# Create an LLM.
llm = LLM(
model="Qwen/Qwen2.5-7B-Instruct",
max_model_len=4096,
max_num_seqs=4,
dtype="float16", # IMPORTANT cause some ATB ops cannot support bf16 on 310P
disable_custom_all_reduce=True,
trust_remote_code=True,
tensor_parallel_size=2,
compilation_config={"custom_ops":['none', "+rms_norm", "+rotary_embedding"]},
)
# Generate texts from the prompts.
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}")
```
---------
Signed-off-by: Vincent Yuan <farawayboat@gmail.com>
Signed-off-by: Yikun Jiang <yikunkero@gmail.com>
Signed-off-by: angazenn <zengyanjia@huawei.com>
Co-authored-by: Vincent Yuan <farawayboat@gmail.com>
Co-authored-by: angazenn <zengyanjia@huawei.com>
Co-authored-by: wangxiyuan <wangxiyuan1007@gmail.com>
Co-authored-by: leo-pony <nengjunma@outlook.com>
Co-authored-by: shen-shanshan <467638484@qq.com>
This change ensures proper functionality for longer sequences by
correctly invoking the _set_cos_sin_cache method with self as the first
argument.
For example, with DeepSeek R1, if this change isn't made, the program
will crash when the input sequence exceeds 4096.
Signed-off-by: Jade Zheng <zheng.shoujian@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?
Deepseek v3 now adopt vanilla chunked prefill on MLA part which is
ineffcient for computing but necessary for chunked prefill. Since PR
https://github.com/vllm-project/vllm-ascend/pull/543 bring v0 scheduler
into vllm-ascend, we can now adopt torch_npu._npu_flash_attention inside
the mla backend for more performance boost. Also there are some
redundant computation inside the rope, which is also removed. This PR
should bring some performance gain for deepseek eager mode inference.
---------
Signed-off-by: ganyi <pleaplusone.gy@gmail.com>
