Search results

[Bug]: tool-call-parse 和 reasoning-parse 缺少deepseek v4这个参数

bugSomething isn't workingSomething isn't working

deepseek

llm-model

Status: Open.

#8875 In vllm-project/vllm-ascend;

· Alei-leia opened on May 5, 2026

[Bug]: gemma4 not possible even updating to transformers 5

bugSomething isn't workingSomething isn't working

Status: Open.

#8869 In vllm-project/vllm-ascend;

· celsowm opened on May 4, 2026

[Bug]: 910c两台机器，1p1d跑deepseekv4 flash w8a8模型，其中decode使用TP1DP16，做长稳压力测试，几个小时候decode处理请求的api server数量就在减少，应该proxy的load_balance_proxy_server_example.py负载分发机制有问题

bugSomething isn't workingSomething isn't working

deepseek

llm-model

Status: Open.

#8867 In vllm-project/vllm-ascend;

· xuyucheng220 opened on May 3, 2026

[Bug]: --enable-lora causes premature response truncation and wrong stop token (<|endoftext|> instead of <|im_end|>) with Qwen3

bugSomething isn't workingSomething isn't working

Status: Open.

#8863 In vllm-project/vllm-ascend;

· 18174045485 opened on May 1, 2026

[Bug]: kvpool layerwise loading crashes when use_layerwise=true

bugSomething isn't workingSomething isn't working

Status: Open.

#8858 In vllm-project/vllm-ascend;

· YuzhengWang5 opened on May 1, 2026

[Bug]: When deploying GLM5 and Deepseek V3.2 separately via PD architecture, there is a probabilistic issue of empty output or garbled characters.

bugSomething isn't workingSomething isn't working

deepseek

glm5

llm-model

Status: Open.

#8853 In vllm-project/vllm-ascend;

· zhenwenqi2024 opened on Apr 30, 2026

[Bug]: P/D proxy may leak resources on recomputed retry and mask metaserver errors after #8836

bugSomething isn't workingSomething isn't working

Status: Open.

#8852 In vllm-project/vllm-ascend;

· liziyu179 opened on Apr 30, 2026

[Feature]: ragflow调用310P部署qwen3-30b模型，报错不支持惩罚算子的问题，是未适配还是bug

feature request

Status: Open.

#8849 In vllm-project/vllm-ascend;

· hackerxu opened on Apr 30, 2026

[New Model]: OpenGuardrails-Text-2510模型执行的是GPTQ 算法的 W4A16 量化，不适用于昇腾 NPU 上的 vLLM-Ascend 框架，请求适配

feature request

Status: Open.

#8848 In vllm-project/vllm-ascend;

· hackerxu opened on Apr 30, 2026

[Feature]: support AsymGQA

feature request

Status: Open.

#8847 In vllm-project/vllm-ascend;

· kinnara1989 opened on Apr 30, 2026

[Bug]:For GLM 5/5.1 under a four-node PD separation setup with TP16 DP2 parallelism, the GPQA accuracy fell short of the standard, achieving only 67.17.

bugSomething isn't workingSomething isn't working

glm5

llm-model

Status: Open.

#8844 In vllm-project/vllm-ascend;

· ZYang6263 opened on Apr 30, 2026

[Bug]: LM-5.1 inference fails with reference to model documentation.

bugSomething isn't workingSomething isn't working

glm5

llm-model

Status: Open.

#8843 In vllm-project/vllm-ascend;

· zhangxinyuehfad opened on Apr 30, 2026