### What this PR does / why we need it?
- global_segment_size and local_buffer_size use constants for unified
management.
- Newly added support for input formats ending with GB, MB, KB, and B,
while being compatible with existing input methods.
### Does this PR introduce _any_ user-facing change?
- Users can use new input methods
- The documentation has also been modified
### How was this patch tested?
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/83f478bb19489b41e9d208b47b4bb5a95ac171ac
---------
Signed-off-by: 李子琦 <liziqi_ing@163.com>
### What this PR does / why we need it?
Refactor kv cache tensor initialization logic.
1. Unify the kvcache tensor initialization logic of deepseek and normal
models
2. spilt `initialize_kv_cache_tensors` into `_allocate_kv_cache_tensors`
and `_reshape_kv_cache_tensors`, following gpu modelrunner in vllm
### Does this PR introduce _any_ user-facing change?
N/A
### How was this patch tested?
CI passed with existing test.
1. prefill disaggregation scenario
4. deepseek + aclgraph/eager mode
5. qwen3 next
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/83f478bb19489b41e9d208b47b4bb5a95ac171ac
---------
Signed-off-by: MengqingCao <cmq0113@163.com>
### What this PR does / why we need it?
move quant before allgather in Allgather EP, rely on
https://github.com/vllm-project/vllm-ascend/pull/3334
Deepseek R1 W8A8 performance on A2 with
`HCCL_ALGO="level0:NA;level1:pipeline"`:
| Seq length | Mean TTFT (ms) main | Mean TTFT (ms) this PR |
|----------|----------|----------|
| 4k | 375.21 | 364.99 |
| 16k | 1465.23 | 1421.75 |
### 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/83f478bb19489b41e9d208b47b4bb5a95ac171ac
---------
Signed-off-by: realliujiaxu <realliujiaxu@163.com>
### What this PR does / why we need it?
Set adxl engine as the default Mooncake backend, because Ascend
Transport is no longer maintained.
Update README to include instructions for installing the adxl backend
Mooncake.
### Does this PR introduce _any_ user-facing change?
Users need to compile and install the mooncake backend for adxl
according to the revised README instructions.
### 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: nwpu-zxr <zhouxuerong2@huawei.com>
### What this PR does / why we need it?
Fix#3891.
The empty of `moe_comm_method` in the above issue is due to the wrong
check for MoE models. To be specific, the method `is_moe_model` only
checks whether a text-only model is a MoE model, without considering
multi-modal models, e.g., `VL` and `Omni`.
Check the config dict recursively to find if it has a key contains
"expert", without checking the model architecture.
It is worth noting that, we can't verify a model by if it contains
`FusedMoE` module because `is_moe_model` is called somewhere before the model loading, e.g., it's called when updating the ACLGraph config in
platform initialization.
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/83f478bb19489b41e9d208b47b4bb5a95ac171ac
---------
Signed-off-by: shen-shanshan <467638484@qq.com>
### What this PR does / why we need it?
Protect the scene where the first problem occurs. The execution should
be interrupted when the video memory application fails, rather than
waiting until an illegal address is accessed.
### Does this PR introduce _any_ user-facing change?
NA
### How was this patch tested?
NA
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/83f478bb19489b41e9d208b47b4bb5a95ac171ac
Signed-off-by: leo-pony <nengjunma@outlook.com>
### What this PR does / why we need it?
In the `update_*attn_params` functions, the
`torch.npu.stream(update_stream)` context manager was previously located
inside the for-loop that updates parameters for each layer. This
resulted in redundant stream initiations for every layer, adding
unnecessary overhead.
This commit refactors the code by moving the stream context manager to
wrap the entire for-loop. This ensures that the update stream is
initiated only once per function call, rather than for each layer. This
change reduces 90us in each decode model.
update stream in every layer:
<img width="1720" height="383" alt="image"
src="https://github.com/user-attachments/assets/70e4cb69-5bc1-4180-a67d-c99132134be6"
/>
remove update stream in every layer:
<img width="1269" height="175" alt="image"
src="https://github.com/user-attachments/assets/0e290edb-b0ce-48fe-b032-1b924ade6ae5"
/>
### 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/83f478bb19489b41e9d208b47b4bb5a95ac171ac
Signed-off-by: wangxiaoxin-sherie <wangxiaoxin7@huawei.com>
Co-authored-by: wangxiaoxin-sherie <wangxiaoxin7@huawei.com>
### What this PR does / why we need it?
As a validation for #3664, add end-to-end tests to monitor the InternVL
model and ensure its continuous proper operation. This PR is only for
single-card. So the models that have more parameters than 8B like 78B
are needed to test using multi-cards.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
`pytest -sv tests/e2e/singlecard/multi-modal/test_internvl.py`
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/83f478bb19489b41e9d208b47b4bb5a95ac171ac
---------
Signed-off-by: gcanlin <canlinguosdu@gmail.com>
### What this PR does / why we need it?
This PR temporarily bypasses the scenario where some models in vLLM
trigger a `ValueError` during the process of storing values in
`static_forward_context` when no `prefix` is specified for the linear
layers, which is a bug in some models in vLLM. The official fix will be
addressed by submitting a PR to the vLLM community that specifies a
prefix for the linear layers in each model.
- vLLM version: v0.11.0
- vLLM main:
https://github.com/vllm-project/vllm/commit/83f478bb19489b41e9d208b47b4bb5a95ac171ac
Signed-off-by: rjg-lyh <1318825571@qq.com>
### What this PR does / why we need it?
This patch mainly fix the the problem of not being able to determine the
exit status of the pod's entrypoint script and some other tiny
optimizations:
1. Shorten wait for server timeout
2. fix typo
3. fix the issue of ais_bench failing to correctly access the proxy URL
in a PD separation scenario.
### 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/83f478bb19489b41e9d208b47b4bb5a95ac171ac
---------
Signed-off-by: wangli <wangli858794774@gmail.com>
This is the follow-up PR to PR #3189, which continues to refactor sfa
into mla and finally remove deepseek_v3_2.py. This is the last PR of
deepseek modeling refactoring. After this, all deepseek-related model
codes are removed from vllm_ascend.
FurtherMore, after this PR deepseek v3.2 can run chunk-prefill with
correct accuracy.
- vLLM version: v0.11.0rc3
- vLLM main:
https://github.com/vllm-project/vllm/commit/83f478bb19489b41e9d208b47b4bb5a95ac171ac
---------
Signed-off-by: whx-sjtu <2952154980@qq.com>
### What this PR does / why we need it?
1. Refactor the file `mtp_proposer.py`, splits torchair related codes
into `mtp_torchair_proposer.py`
2. According to https://github.com/vllm-project/vllm/pull/24539,
implements padded speculative decoding as described in
https://github.com/vllm-project/vllm/issues/21984.
### Does this PR introduce _any_ user-facing change?
User can use `disable_padded_drafter_batch` to disable/enable padded
speculation, default is `False`.
offline example:
```
speculative_config={"method": "deepseek_mtp", "num_speculative_tokens": 1, "disable_padded_drafter_batch": False}
```
### How was this patch tested?
- [x] egaer with pad/unpad:
- [x] aclgraph with pad/unpad
- [x] torchair with pad/unpad
performance test of deepseek-r1 with tp16、dp1
aclgraph with pad ITL: 168ms
aclgraph with unpad ITL: 169ms
original: 178ms
- vLLM version: v0.11.0rc3
- vLLM main:
https://github.com/vllm-project/vllm/commit/83f478bb19489b41e9d208b47b4bb5a95ac171ac
---------
Signed-off-by: xuyexiong <xuyexiong@huawei.com>
We notice that sometimes user build vllm-ascend with incorrect torch
version. In this case, the build is passed, but when running the code,
the error `AttributeError: '_OpNamespace' '_C_ascend' object has no
attribute 'weak_ref_tensor'` is raised. Let's force the torch version to
2.7.1 and check the torch version when build from source to fix the
issue.
closes: #3342
- vLLM version: v0.11.0rc3
- vLLM main:
https://github.com/vllm-project/vllm/commit/c9461e05a4ed3557cfbf4b15ded1e26761cc39ca
Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>
Added instructions for resolving 'invalid tar header' error on Kylin OS with an ARM64 architecture on Atlas300I hardware during docker
pull, including steps for offline loading of docker images.
---
### What this PR does / why we need it?
The primary motivation for this PR is to address a critical `docker
pull` failure that occurs on specific, yet important, enterprise
environments. Specifically, when operating on **Kylin OS (麒麟操作系统) with
an ARM64 architecture on Atlas300I hardware**, users frequently
encounter an `archive/tar: invalid tar header` error, which completely
blocks the setup process. This issue has been consistently reproduced,
with multiple retries failing with the same error, confirming that it is
a persistent environmental problem rather than a transient network
issue.
<img width="2060" height="525" alt="image"
src="https://github.com/user-attachments/assets/6c1c5728-de27-476f-8df4-723564fc290b"
/>
This guide provides a robust, step-by-step workaround using an
offline-loading method (`docker save` on a host machine and `docker
load` on the target machine). This solution is crucial for enabling
users on this platform to use vLLM.
This contribution does not directly fix an existing issue number, but it
proactively solves a significant environmental and usability problem for
a growing user base.
### Does this PR introduce _any_ user-facing change?
No.It does not alter any code, APIs, interfaces, or existing behavior of
the vLLM project.
### How was this patch tested?
The instructions and troubleshooting steps in this guide were validated
through a real-world, end-to-end test case on the my hardware and OS.
The testing process was as follows:
1. **Problem Reproduction**: An attempt was made to directly `docker
pull` the `vllm-ascend:v0.10.0rc1-310p` image on a target machine
running Kylin OS (ARM64). The `invalid tar header` failure was
successfully and consistently reproduced, confirming the existence of
the problem.
2. **Solution Implementation**: The workaround detailed in the guide was
executed:
* On a separate host machine (Ubuntu x86_64), the image was successfully
pulled using the `--platform linux/arm64` flag.
* The image was then saved to a `.tar` archive using `docker save`.
* The `.tar` archive was transferred to the target Kylin OS machine.
* The image was successfully loaded from the archive using `docker load
-i ...`.
3. **End-to-End Validation**: After loading the image, the vLLM
container was launched on the target machine following the instructions
in the guide. Both online inference (via `curl` to the API server) and
offline inference (via the Python script) were executed successfully,
confirming that the entire workflow described in the document is
accurate and effective.
Since this is a documentation-only change based on a validated workflow,
no new unit or integration tests were added to the codebase.
- vLLM version: v0.11.0rc3
- vLLM main:
https://github.com/vllm-project/vllm/commit/83f478bb19489b41e9d208b47b4bb5a95ac171ac
---------
Signed-off-by: Liwx <liweixuan1014@gmail.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>