sglang_issue_search1

decode-memory-bottleneck/sglang_issue_search1.txt 3761 chars
"Skip to content\nNavigation Menu\nPlatform\nSolutions\nResources\nOpen Source\nEnterprise\nPricing\nSign in\nSign up\nsgl-project\n/\nsglang\nPublic\nNotifications\nFork 7.2k\n Star 30.3k\nCode\nIssues\n706\nPull requests\n3.6k\nDiscussions\nActions\nSecurity and quality\nInsights\nIssues\nAssigned to me\nCreated by me\nMentioned\nRecent activity\nViews\nMilestones\nLabels\nFeedback\nPreview\nCollapse sidebar\nDevelopment Roadmap (2026 Q2)\n#22949 · merrymercy opened \n3\n[Feature] Overlap Spec Support\n#11762 · hnyls2002 opened \n25\nDeepSeek V4 Roadmap\n#23602 · fzyzcjy opened \n70\nIssues\nSearch Issues\nis:issue\n \ndisagg\n \nmemory\n \ndecode\nNew issue\nSearch results\nOpen\n20\n (20)\nClosed\n43\n (43)\nAuthor\nLabels\nProjects\nMilestones\nAssignees\nTypes\n[Bug] HiCache draft KV pool backup crashes with cudaMemcpyBatchAsync invalid argument in PD disaggregation\nStatus: Open.\n#31252 \nIn sgl-project/sglang;\n· F0rtiter opened \n[RFC] RuntimeContext as the configuration API\nStatus: Open.\n#30696 \nIn sgl-project/sglang;\n· ch-wan opened \n[Feature] Per-request opt-out of speculative decoding (RFC: disable_speculative_decoding sampling param)\nStatus: Open.\n#30263 \nIn sgl-project/sglang;\n· anencore94 opened \n[Bug] Deepseek V4 NVFP4 Disagg Decode node failure\nStatus: Closed (completed).\n#30242 \nIn sgl-project/sglang;\n· by shyeh25 was closed \n1\n5\ncomments\n[Bug] Compatibility issues with PP and Hicache causing errors\nStatus: Open.\n#30158 \nIn sgl-project/sglang;\n· enternal111 opened \n15\ncomments\n[Bug] The performance of deepseekv4 failed to pass the test on the main branch\nStatus: Open.\n#30082 \nIn sgl-project/sglang;\n· BJWang-ant opened \n1\ncomment\nMambaRadixCache shape mismatch in PD chunked prefill when caching unfinished req\nStatus: Not planned (skipped).\n#30019 \nIn sgl-project/sglang;\n· by 5SSjw was closed \n1\ncomment\nPD disaggregation decode deadlocks at high per-rank batch size with CUDA graphs (all ranks spinning in flashinfer moeA2ACombineKernel)\nStatus: Open.\n#30010 \nIn sgl-project/sglang;\n· yhyang201 opened \n1\ncomment\n[Failure Tracker] PR Test (AMD)\nStatus: Open.\n#27937 \nIn sgl-project/sglang;\n· amd-bot opened \n[RFC] ParisKV2 sparse attention integration for SGLang\nStatus: Open.\n#27652 \nIn sgl-project/sglang;\n· amy-77 opened \n[Bug] Unable to run MiMo-V2.5-Pro with pipeline parallelism: ValueError in srt/models/mimo_v2.py\nStatus: Open.\n#26893 \nIn sgl-project/sglang;\n· Kevsnz opened \n[Bug] DeepSeek V4 Pro on 8 * H20-3e requests hang after warmup without any response\nStatus: Open.\n#26783 \nIn sgl-project/sglang;\n· SAXEM1997 opened \n6\ncomments\n[Bug] Base-mode non-deterministic output under cancel/retry + long-context concurrent traffic\ninactive\nStatus: Closed (completed).\n#24845 \nIn sgl-project/sglang;\n· by Yunzez was closed \n3\ncomments\n[Bug] DeepSeek-V4-Pro,PD Disaggregation Input Length Hard-Capped at 18,432 Tokens Due to Incorrect SWA Limit Enforcement on Prefill Node\nStatus: Open.\n#24523 \nIn sgl-project/sglang;\n· txh1873749380 opened \n14\ncomments\nSegfault in NIXL UCX worker during disaggregated KV transfer\nStatus: Closed (completed).\n#23499 \nIn sgl-project/sglang;\n· by ishandhanani was closed \n1\ncomment\n[Roadmap] SGLang-Diffusion (26 Q2)\n0 / 3\nStatus: Open.\n#23035 \nIn sgl-project/sglang;\n· mickqian opened \n6\ncomments\n[Bug] ShmPointerMMData shared memory prematurely unlinked when using multimodal + dp_size > 1 + PD disaggregation\nStatus: Open.\n#23022 \nIn sgl-project/sglang;\n· AndyLi429 opened \n1\n3\ncomments\nPrevious123Next\nFooter\n© 2026 GitHub, Inc.\nFooter navigation\nTerms\nPrivacy\nSecurity\nStatus\nCommunity\nDocs\nContact\nManage cookies\nDo not share my personal information"