Repository Issue Activity (beta)

#17050 [Tracking] CI Test Failures and Fixes

#26340 CUDA Coredump Tracker

#32817 [Feature] Support runtime attach/detach of HiCache storage backend on UnifiedRadixCache

#35582 Qwen-VL: literal vision marker in text is parsed as extra image when request contains an image

#20865 [Feature] Improve Unit Test Coverage

#33522 [Roadmap]Fast Engine Recovery: Weight Cache Daemon

#36071 [Bug][AMD] DSA dense/k-only decode graph produces all-NaN logits with DP attention

#36083 [Feature] Extend cross-layer index-topk reuse (dsa_layer_skips_topk) to DeepSeek-V4

#36081 [Bug] DSV4F0731+DSPARK ERROR IN SGLANG 0.5.18

#36010 [Bug] NVFP4 KV cache + speculative decoding: extend_prefix_lens_cpu is None in the dequant workspace path

#36058 [Bug] /usr/bin/ld: cannot find -lcuda: No such file or directory collect2: error: ld returned 1 exit status ninja: build stopped: subcommand failed.

#34737 [Bug] Stale staging watermark subscribers survive prefill failure

#36048 [Bug] Qwen3.8-27B AWQ has no usable fast path on single Ada SM89 32GB: CUDA graph hangs and native MTP cannot allocate requests

#36033 [PD] Inconsistent prefill→decode failure notification across transfer backends; NIXL has none

#32321 [RFC] Replace the MLX runner-stub split with one Torch-owned SRT path and an exported whole-model MLX region

#35591 [Bug] ROCm images pin AITER below the FlyDSL MXFP4 MoE kernels, making tuned AITER_CONFIG_FMOE tables unusable on gfx950

#35949 [Bug] Qwen 3.8 27b NVFP4 incorrect coordinate detection on image

#29093 [Bug] Model Gateway docs list wrong CLI flag names for tokenizer cache options

#29061 [Bug] OpenAI tool schemas are mutated before apply_chat_template, causing train-serving prompt mismatch

#29034 [Bug] --hicache-size over-allocates Mamba host cache for hybrid Mamba models (e.g. Qwen3.5)

#29013 [Bug] MiMo-Pro-V2.5 with Hicache: Mooncake Store crashes with `AttributeError: 'tuple' object has no attribute 'data_ptr'` on AsymmetricMHA models

#26783 [Bug] DeepSeek V4 Pro on 8 * H20-3e requests hang after warmup without any response

#28414 [Bug] DeepSeek-V4 has slight accuracy issue

#21379 [Bug] evict_mamba crashes with AssertionError: evict leaf node invalid when leaf's KV is locked by in-flight request

#25682 [Bug] JIT compilation failure (ninja status 1) in ActivationKernel during CUDA graph capture for Qwen3.5-122B-A10B