Issues

• [Feature]: Support DeepSeek-V3 with PyTorch-based graph construction on Ascend NPU

  feature
  xLLM-AI/xllm#2407 · Filter by author XuZhang99 opened on Sep 30, 2026

• [Bug]: PD 分离部署乱码

  bug
  xLLM-AI/xllm#2359 · Filter by author renren123 opened on Sep 22, 2026

• [Bug]: [NPU][ACL Graph][DP] Single request crashes Qwen3.5 on an empty DP shard

  bug
  xLLM-AI/xllm#2135 · Filter by author ShareableXue opened on Aug 5, 2026

• [Feature]: Support kimi k3

  feature
  xLLM-AI/xllm#2085 · Filter by author zhanglistar opened on Jul 30, 2026

• Feature: Runtime CP<->DP mode switching for xLLM workers

  enhancementfeature
  xLLM-AI/xllm#1967 · Filter by author cchh05 opened on Jul 17, 2026
  2

• [Question]: NPU上，针对短上下文推理加速的卡内PD并行技术。实验室已有原型，想确认对业务是否有帮助。

  question
  xLLM-AI/xllm#1950 · Filter by author mingzhehao637-droid opened on Jul 16, 2026
  3

• [Question]: models/llm/qwen3_5.h 必须编译 NPU 头文件，而不是 common 的

  question
  xLLM-AI/xllm#1916 · Filter by author cding-nv opened on Jul 10, 2026
  2

  

• [Question]: 请问BW1000 上部署Minimax M3/M2.7，token吞吐率实测效果咋样？是否支持专家并行？

  question
  xLLM-AI/xllm#1905 · Filter by author HardenGale opened on Jul 7, 2026
  1

  

• [Question]: A3单机跑GLM5.2-W8A8最大支持多少上下文？

  question
  xLLM-AI/xllm#1862 · Filter by author zhujianwei-ops opened on Jul 1, 2026
  1

• [Feature]: MACA / MetaX C500 Adaptation Issue Draft

  xLLM-AI/xllm#1851 · Filter by author Dayuxiaoshui opened on Jun 30, 2026
  3

• [Feature]: Support DSpark for DeepSeek V4

  feature
  Feature

  xLLM-AI/xllm#1837 · Filter by author shanshanpt opened on Jun 27, 2026
  

• [Question]: Missing MTTOplib

  question
  xLLM-AI/xllm#1557 · Filter by author nvgruel123 opened on May 26, 2026
  5

• [Bug]: ACL Graph 模式下 ATB CustomPagedAttention context_lens 冻结导致 decode 全零输出

  bug
  xLLM-AI/xllm#1466 · Filter by author sunbaosong opened on May 16, 2026

• [Feature]: Support Prefix Cache for Qwen-Next Model (Mamba Cache Mode)

  feature
  xLLM-AI/xllm#1016 · Filter by author JC-ut0 opened on Mar 9, 2026

• [Question]: GLM-5-W8A8如何配置tool-call-parser和reasoning_parser

  question
  xLLM-AI/xllm#1004 · Filter by author Stonesan233 opened on Mar 5, 2026
  1

• [Question]: 想问下将来有可能适配Atlas 300I 系列的NPU吗

  question
  xLLM-AI/xllm#995 · Filter by author Sp00n-X opened on Mar 4, 2026

• [Bug]: GLM-5-w8a8 fails to launch in 910C

  bug
  xLLM-AI/xllm#967 · Filter by author Pastens opened on Mar 2, 2026

• [Feature]: 能支持一下Qwen3-Omni系列的模型吗

  feature
  xLLM-AI/xllm#962 · Filter by author Lcy0801 opened on Feb 28, 2026
  1

• [Bug]: Qwen3-8B crash with v0.7.2 on ascend 910C

  bug
  xLLM-AI/xllm#959 · Filter by author liujinguang0125 opened on Feb 28, 2026

• [Bug]: beam search does not work on 910B

  bug
  xLLM-AI/xllm#917 · Filter by author Euphonious opened on Feb 12, 2026

• qwen3的模型，请求时候怎么配，能没有think的过程，直接输出结果

  xLLM-AI/xllm#887 · Filter by author w066650 opened on Feb 5, 2026
  1

• MTP k>1 crashes during warmup due to uninitialized embed_tokens weights

  xLLM-AI/xllm#835 · Filter by author QwertyJack opened on Jan 29, 2026

• [Bug]: 使用xllm-0.7.2-release-hb-rc2-arm镜像PD 分离部署运行 xllm_master_serving 时缺少 librust_tokenizers.so

  bug
  xLLM-AI/xllm#833 · Filter by author w1ida opened on Jan 29, 2026
  

• [Question]: xllm 技术报告性能复现问题：与 vllm_ascend 最新版本实测结果不一致

  question
  xLLM-AI/xllm#832 · Filter by author w1ida opened on Jan 29, 2026
  1

• KV cache memory leak when prefix cache is enabled

  xLLM-AI/xllm#828 · Filter by author QwertyJack opened on Jan 29, 2026
  4