All pull requests

Pull requests

• Fix exact-fit admission of leased cached prefixes

  #2734·bmehta001 opened Oct 14, 2026Contributor·

  ·
  1

  4

• Add opt-in NVTX ranges for Engine and DFlash profiling

  #2733·Tianlei Wu (tianleiwu) opened Oct 14, 2026Contributor·

  ·
  1

• Stabilize Qwen CUDA graph input and KV-cache buffers

  #2732·Yen-Shi Wang (yen-shi) opened Oct 14, 2026Contributor·

  ·
  4

• Add static-graph Engine backend results next to the paged Engine results

  #2730·Tianlei Wu (tianleiwu) opened Oct 13, 2026Contributor·2 of 3 tasks·

  ·

• Add Qwen3.8-Flash Engine qualification tools and package builders

  #2729·Tianlei Wu (tianleiwu) opened Oct 13, 2026Contributor·4/4·3 tasks done·

  ·

• Host the Qwen3.8-Flash CUDA-graph decode loop in the Engine's static path

  #2728·Tianlei Wu (tianleiwu) opened Oct 13, 2026Contributor·3/4·3 tasks done·

  ·

• Add the qwen38_flash_graph model type and its device decode engine

  #2727·Tianlei Wu (tianleiwu) opened Oct 13, 2026Contributor·2/4·3 tasks done·

  ·

• Add the Qwen3.8-Flash CUDA data plane library and its host loader

  #2726·Tianlei Wu (tianleiwu) opened Oct 13, 2026Contributor·1/4·3 tasks done·

  ·

• Support rank-4 multimodal embedding buffers

  #2724·Xiaoyu Z (xiaoyu-work) opened Oct 13, 2026Member·

  ·
  1

• Keep SDK state alive through telemetry shutdown and update to 3.10.281.1

  #2723·bmehta001 opened Oct 13, 2026Contributor·

  ·
  9

  

• Cover sparse pools and evicted ancestry in checkpoint scan benchmarks

  #2721·bmehta001 opened Oct 13, 2026Contributor·

  ·
  2

  

• Add diarization support to streaming ASR examples

  #2720·Mason Corey (themason2011) opened Oct 12, 2026·

  ·

• [Diarization] Adding GenAI Processor Modifications for Diarization Using Nemotron-3-Diarization

  #2719·Mason Corey (themason2011) opened Oct 12, 2026·

  ·
  6

• Add Qwen-3.8 Flash Next benchmarks and reproduction results

  #2715·kunal-vaishnavi opened Oct 12, 2026Contributor·3/3·

  ·

• Add build and runtime support for Qwen-3.8 Flash Next

  #2714·kunal-vaishnavi opened Oct 12, 2026Contributor·2/3·

  ·
  28

• Add Qwen-3.8 Flash Next to model builder

  #2713·kunal-vaishnavi opened Oct 12, 2026Contributor·1/3·

  ·
  35

• Add experimental Nemotron-H hybrid MTP support on CUDA

  #2710·Rui Ren (rui-ren) opened Oct 12, 2026Contributor·

  ·
  20

• Guard graph buffer packed-row and position dimensions

  #2709·Tianlei Wu (tianleiwu) opened Oct 12, 2026Contributor·3 tasks done·

  ·

• Isolate MTP head metadata and decoder configuration

  #2708·Tianlei Wu (tianleiwu) opened Oct 12, 2026Contributor·3 tasks done·

  ·

• Add LoRA adapter support to native and Python benchmarks

  #2707·Jiajia Qin (qjia7) opened Oct 12, 2026Contributor·

  ·
  3

• Add opt-in Engine-wide snapshot cache budgets

  #2706·bmehta001 opened Oct 11, 2026Contributor·

  ·
  3

• Allow opting out of AMDGPU graph capture via provider option

  #2705·AMDmoore opened Oct 10, 2026Contributor·

  ·
  1

• Add focused Engine parity QA and first-difference reports

  #2701·bmehta001 opened Oct 9, 2026Contributor·

  ·
  5

• integrate gated delta net compact state replay on webgpu

  #2700·aciddelgado opened Oct 9, 2026Contributor·

  ·

• Fix ignored context limits for Qwen3.6-35B-A3B in C/C++ examples

  #2692·Hrishith Thadicherla (hthadicherla) opened Oct 8, 2026Contributor·

  ·
  14