Search results

[Bug]:vllm0.13.0.rc1，打开full算子入图模式后精度异常

bugSomething isn't workingSomething isn't working

Status: Open.

#6456 In vllm-project/vllm-ascend;

· xieanxiean opened on Jan 31, 2026

[Bug]: The number of tokens generated by the deep thinking of the deepseek R1 model is less than that of deepseekV3.1 terminus

bugSomething isn't workingSomething isn't working

Status: Open.

#6450 In vllm-project/vllm-ascend;

· shitianlong1994 opened on Jan 31, 2026

[Bug]: GLLM-4.7 MTP infer error:RuntimeError: shape '[6, -1]' is invalid for input of size 512

bugSomething isn't workingSomething isn't working

Status: Open.

#6449 In vllm-project/vllm-ascend;

· Luu-Luu opened on Jan 31, 2026

[Bug]: 0.14.0rc1+GLM-4.7，KV Cache满载时TPS大幅下降

bugSomething isn't workingSomething isn't working

Status: Open.

#6446 In vllm-project/vllm-ascend;

· samsuzhang opened on Jan 31, 2026

[RFC]: eagle3 speculative decoding supports rotation quantization

RFCRequest For CommentsRequest For Comments

Status: Open.

#6445 In vllm-project/vllm-ascend;

· pjgao opened on Jan 30, 2026

·  

2026 Q1 RoadMap

[RFC]: DeepSeek MLA supports FA3

RFCRequest For CommentsRequest For Comments

Status: Open.

#6437 In vllm-project/vllm-ascend;

· pichangping opened on Jan 30, 2026

·  

2026 Q1 RoadMap

[RFC]: Deepseek 3.2 Indexer Operator Preprocessing Operator

RFCRequest For CommentsRequest For Comments

Status: Open.

#6436 In vllm-project/vllm-ascend;

· weijinqian0 opened on Jan 30, 2026

·  

2026 Q1 RoadMap

[RFC]: MOE fusion operator optimization

RFCRequest For CommentsRequest For Comments

Status: Open.

#6434 In vllm-project/vllm-ascend;

· weijinqian0 opened on Jan 30, 2026

·  

2026 Q1 RoadMap

[RFC]: Attention: Unified FIA operators.

RFCRequest For CommentsRequest For Comments

Status: Open.

#6433 In vllm-project/vllm-ascend;

· weijinqian0 opened on Jan 30, 2026

·  

2026 Q1 RoadMap

[RFC]: Add support for <code>CUDAGraphMode.FULL</code> mode

RFCRequest For CommentsRequest For Comments

Status: Open.

Feature

#6431 In vllm-project/vllm-ascend;

· yiz-liu opened on Jan 30, 2026

·  

2026 Q1 RoadMap

[Bug]: ( 0.13.0rc2+ torch-npu 2.8.0.post1) error: use of undeclared identifier 'aclOpExecutor' when deploy Qwen3-235B-A22B-Instruct-2507-w8a8-QuaRot

bugSomething isn't workingSomething isn't working

Status: Open.

#6429 In vllm-project/vllm-ascend;

· gao12312 opened on Jan 30, 2026

[RFC]: Adapting Qwen Dense Series Models to the 310P Chip

RFCRequest For CommentsRequest For Comments

Status: Open.

#6427 In vllm-project/vllm-ascend;

· wanghengkang opened on Jan 30, 2026

·  

2026 Q1 RoadMap