xc-llm-ascend

Files

Delphine-Nic a3e9673137 [long seq feat]GQA support long-prefill-token-threshold and fixbug (#4209 )

### What this PR does / why we need it?
GQA chunk prefill with pcp and dcp support long-prefill-token-threshold

The markdown format results is as below:
| dataset | version | metric | mode | vllm-api-general-chat |
|----- | ----- | ----- | ----- | -----|
| gsm8kdataset | - | accuracy | gen | 96.13 |

- vLLM version: v0.11.0
- vLLM main:
2918c1b49c

---------

Signed-off-by: Delphine-Nic <tanwenqin@huawei.com>
Signed-off-by: Delphine-Nic <t00608739@china.huawei.com>
Co-authored-by: Delphine-Nic <tanwenqin@huawei.com>
Co-authored-by: Delphine-Nic <t00608739@china.huawei.com>

2025-11-19 18:10:27 +08:00

__init__.py

[Core] Make V1 work and enable V1 engine test (#389 )

2025-03-28 19:34:23 +08:00

attention_mask.py

Upgrade CANN to 8.3.rc1 (#3945 )

2025-11-03 20:21:07 +08:00

attention_v1.py

[long seq feat]GQA support long-prefill-token-threshold and fixbug (#4209 )

2025-11-19 18:10:27 +08:00

mla_v1.py

remove get_metadata_cls (#4087 )