Skip to content

Releases: nylon-memory/NylonME

v0.3.5

Choose a tag to compare

@github-actions github-actions released this 22 Sep 09:35

Full Changelog: v0.3.4...v0.3.5

Full Changelog: v0.3.4...v0.3.5

Full Changelog: v0.3.4...v0.3.5

v0.3.4

Choose a tag to compare

@github-actions github-actions released this 20 Sep 09:51

Python SDK 0.2.2

  • Adds sync/async feedback() client methods for the ReportFeedback API introduced in v0.3.3.
  • Ships regenerated gRPC stubs for the feedback contract.
  • Engine-side feedback-driven reflection remains disabled by default; set NYLON_FEEDBACK_REFLECT=1 to enable LLM consumption of recorded feedback.

Full Changelog: v0.3.3...v0.3.4

v0.3.3

Choose a tag to compare

@github-actions github-actions released this 20 Sep 09:15

Full Changelog: v0.3.2...v0.3.3

Full Changelog: v0.3.2...v0.3.3

v0.3.2

Choose a tag to compare

@github-actions github-actions released this 18 Sep 10:18

Full Changelog: v0.3.1...v0.3.2

Full Changelog: v0.3.1...v0.3.2

Full Changelog: v0.3.1...v0.3.2

修复

  • fix(engine): HNSW 索引重启丢失(生产级) — 此前向量索引只在编织时构建,服务每次重启后向量种子通道静默失效,检索退化为纯词面。现启动时自动从持久化节点回填索引。
  • fix(engine): resonate 种子补齐 — 扩散阶段可能因 budget 截断丢失部分直接命中种子,现强制并回候选集。

评测基础设施

  • 编织结果磁盘缓存(查询侧实验迭代 1 小时 → 分钟级)
  • 全证据命中口径 / 逐跳缺口解剖 / 多通道 RRF 融合旋钮(默认关闭)

基准

LoCoMo 十会话 recall@10: 85.4% → 85.8%(引擎修复后新基线)

V0.3.2- HNSW restart loss repair (production level)

Repair

-* * fix (engine): HNSW index lost during restart (production level) * * - Previously, vector indexes were only built during weaving, and the vector seed channel silently failed after each service restart, resulting in retrieval degradation to pure word faces. Automatically backfill indexes from persistent nodes upon startup.
-* * fix (engine): Resonate seed replenishment * * - During the diffusion phase, the missing portion may be directly hit by the seed due to budget truncation. Now, it is forced to merge with the candidate set.

Evaluate infrastructure

-Weaving result disk cache (query side experiment iteration 1 hour → minute level)
-Full evidence hit caliber/hop by hop gap dissection/multi-channel RRF fusion knob (default off)

Benchmark

LoCoMo Ten Sessions recall@10 : 85.4% → 85.8% (new baseline after engine repair)

v0.3.1

Choose a tag to compare

@github-actions github-actions released this 18 Sep 06:20

Full Changelog: v0.3.0...v0.3.1

Full Changelog: v0.3.0...v0.3.1

Full Changelog: v0.3.0...v0.3.1

v0.3.0

Choose a tag to compare

@github-actions github-actions released this 16 Sep 06:42

Full Changelog: v0.2.1...v0.3.0

Full Changelog: v0.2.1...v0.3.0

Full Changelog: v0.2.1...v0.3.0

v0.2.1

Choose a tag to compare

@github-actions github-actions released this 12 Aug 09:19

Full Changelog: v0.2.0...v0.2.1

Full Changelog: v0.2.0...v0.2.1

Full Changelog: v0.2.0...v0.2.1

v0.2.0

Choose a tag to compare

@github-actions github-actions released this 12 Aug 07:34

Full Changelog: v0.1.0...v0.2.0

Full Changelog: v0.1.0...v0.2.0

Full Changelog: v0.1.0...v0.2.0

v0.1.0

Choose a tag to compare

@github-actions github-actions released this 12 Aug 06:53