All issues

Search results

perf(load): CXL→GPU load path does not scale with in-flight requests — TTFT ≈ N × 176 ms when decode is short

enhancementNew feature or requestNew feature or request

Status: Open.

#79 In xcena-dev/maru;

· kihwan-XCENA opened on Aug 26, 2026

feat(maru_vllm): maru-native CUDA kernel for packed GPU↔CXL KV transfers

Status: Open.

#74 In xcena-dev/maru;

· seohui-XCENA opened on Aug 19, 2026

[Bug] deferred load + write-behind 조합에서 요청 취소 시 EngineCore 사망

bugSomething isn't workingSomething isn't working

Status: Open.

#73 In xcena-dev/maru;

· youngrok-XCENA opened on Aug 19, 2026

resource manager: devices that appear after startup never join the pool (boot race with large CXL devices)

Status: Open.

#68 In xcena-dev/maru;

· jooho-XCENA opened on Jul 23, 2026

Defense-in-depth: bounds-check server-supplied offset/length before device ops in handler plugins

Status: Open.

#59 In xcena-dev/maru;

· hyunyul-XCENA opened on Jul 6, 2026

prepare pypi organization

Status: Open.

#52 In xcena-dev/maru;

· darjeeling opened on May 28, 2026

test to against multiple python versions

Status: Open.

#51 In xcena-dev/maru;

· darjeeling opened on May 28, 2026

recomand to adopt uv

Status: Open.

#50 In xcena-dev/maru;

· darjeeling opened on May 28, 2026

feat: event-driven CXL device hotplug monitoring

Status: Open.

#49 In xcena-dev/maru;

· seohui-XCENA opened on Apr 17, 2026

maru_sglang: chunk_size should be dynamically derived from per-token KV size

enhancementNew feature or requestNew feature or request

Status: Open.

#34 In xcena-dev/maru;

· kihwan-XCENA opened on Mar 27, 2026

[Feature] Implement eviction and capacity reclaim on CXL pool exhaustion

high-priorityHigh Priority itemHigh Priority item

Status: Open.

Feature

#4 In xcena-dev/maru;

· seohui-XCENA opened on Mar 6, 2026