All pull requests

Pull requests

• cua_s1: retain bounded vision cache across screenshot grids

  #127·Levius-Fubuki opened Oct 9, 2026Collaborator·4 tasks·

  ·

• docs: publish roadmap for models, features, and agent integration

  #126·hsliuustc0106 opened Oct 8, 2026Contributor·4 tasks done·

  ·

• [Model] openjev/openjev: request compilation and calibrated answers (CPU half of #95)

  #124·chaimaerachdi opened Oct 8, 2026·4 tasks done·

  ·
  1

• feat(native): add multimodal language and shared vision CUDA Graph replay

  #123·Levius-Fubuki opened Oct 8, 2026Collaborator·4 tasks done·

  ·
  1

• jemm: load verified calibration and return typed errors

  #122·Levius-Fubuki opened Oct 8, 2026Collaborator·

  ·
  1

• omnijev: add the request contract, preparation, heads and export

  #120·twu3202 opened Oct 8, 2026Contributor·4 tasks done·

  ·
  3

• jemm: add native decision worker and pinned weight export

  #118·Levius-Fubuki opened Oct 8, 2026Collaborator·4 tasks done·

  ·
  3

• qwen: share vision execution for 4B and 27B models

  #117·Levius-Fubuki opened Oct 8, 2026Collaborator·4 tasks done·

  ·
  1

• bench: add frozen HTTP workload and concurrency matrix

  #116·Levius-Fubuki opened Oct 8, 2026Collaborator·4 tasks done·

  ·
  2

• Valen: serve Valen-Preview-0923 through a Python reference worker

  #115·MrDongsls opened Oct 7, 2026Collaborator·4 tasks done·

  ·
  2

• decider: reuse request prefixes with conservative auto selection

  #113·Levius-Fubuki opened Oct 7, 2026Collaborator·4 tasks done·

  ·
  3

• feat(decider): independent CUDA Graph replay mode

  #112·Levius-Fubuki opened Oct 7, 2026Collaborator·4 tasks done·

  ·
  1

• feat(decider): bounded request-local CUDA batching

  #111·Levius-Fubuki opened Oct 7, 2026Collaborator·4 tasks done·

  ·
  1

• decider: add verified eager native CUDA worker

  high priority
  #110·Levius-Fubuki opened Oct 7, 2026Collaborator·4 tasks done·

  ·
  1

• runtime: optional bounded admission

  high priority
  #109·xiaoyu-xyz opened Oct 7, 2026Contributor·4 tasks done·

  ·
  1

• cua_s1: add native vision CUDA Graph replay

  high priority
  #108·Levius-Fubuki opened Oct 7, 2026Collaborator·4 tasks done·

  ·
  1

• docs: add Apple Silicon CPU install path and venv interpreter to gett…

  #107·MinhaoLi0318 opened Oct 7, 2026·4 tasks done·

  ·
  1

• qwen3_5: retain multimodal Graph regressions in workspace tests

  high priority
  #106·Levius-Fubuki opened Oct 6, 2026Collaborator·4 tasks done·

  ·
  2

• cua_s1: add native multimodal language CUDA Graph replay

  high priority
  #103·Levius-Fubuki opened Oct 6, 2026Collaborator·4 tasks done·

  ·
  5

• ci: compile every declared CUDA backend

  high priority
  #101·xiaoyu-xyz opened Oct 6, 2026Contributor·

  ·

• laya: pin the TileLang target instead of inferring it

  high priority
  #100·xiaoyu-xyz opened Oct 6, 2026Contributor·

  ·

• qwen3_5: run prompts that share prefixes once per prefix

  #99·twu3202 opened Oct 6, 2026Contributor·4 tasks done·

  ·

• qwen3_5: add a GEMM handle with one algorithm per weight shape

  #98·twu3202 opened Oct 6, 2026Contributor·4 tasks done·

  ·

• qwen3_5: add prefix continuation operations

  #97·twu3202 opened Oct 6, 2026Contributor·4 tasks done·

  ·
  1

• jev_vl: add native inference and prefix caching

  high priority
  #96·linear3735 opened Oct 6, 2026·4 tasks·

  ·
  6