varun29ankuS/shodh-memory

vgliner-bi-edge-onnx-v1 Feature

This release adds 2 notable features for engineering teams evaluating rollout.

Published 13d AI Agents & Assistants

View tool

✓ No known CVEs patched

Read the diff → Tool health → What is this tool? →

✓ No known CVEs patched in this version

Topics

agent-memory agentic-ai ai-agents ai-memory claude cognitive-memory

+13 more

edge-ai graphrag knowledge-graph llm-free local-first mcp model-context-protocol robot-memory robotics ros2 rust vector-db zenoh

Summary

AI summary

ONNX export of gliner-bi-edge-v2.0 with perfect span-F1 parity on GDELT evaluation.

Changes in this release

Type	Severity	Summary	CVE
Feature
Feature	Low	Adds ONNX export of gliner-bi-edge-v2.0 model (149 MB fp32). Adds ONNX export of gliner-bi-edge-v2.0 model (149 MB fp32). Source: llm_adapter@2026-07-14 Confidence: high	—
Feature	Low	Provides label embeddings file `label_embeddings.bin` with 141×384 LE f32 vectors. Provides label embeddings file `label_embeddings.bin` with 141×384 LE f32 vectors. Source: granite4.1:30b@2026-07-14-audit Confidence: low	—
Feature	Low	Includes tokenizer configuration files (`tokenizer.json` and configs). Includes tokenizer configuration files (`tokenizer.json` and configs). Source: granite4.1:30b@2026-07-14-audit Confidence: low	—
Performance	Low	Reports parity vs PyTorch: span-F1 1.000 on GDELT eval (1269/1269 identical spans). Reports parity vs PyTorch: span-F1 1.000 on GDELT eval (1269/1269 identical spans). Source: granite4.1:30b@2026-07-14-audit Confidence: low	—

Full changelog

ONNX export of knowledgator/gliner-bi-edge-v2.0 (Apache-2.0) — the shodh-memory production entity typer.

Exported via scripts/export_gliner_bi_edge.py (gliner 0.2.27, onnxruntime 1.27.0, torch 2.10, opset 19). Parity vs torch: span-F1 1.000 on the 100-passage GDELT eval (1269/1269 identical spans) — see parity.json.

| asset | role |
|---|---|
| model.onnx (149 MB, fp32) | text tower + span scorer; inputs incl. labels_embeds |
| label_embeddings.bin | 141×384 LE f32, row-major, in entity-type-schema.json fine-label order |
| tokenizer.json + configs | tokenizer + GLiNER config |

Consumed by src/embeddings/gliner.rs (env: model dir). int8 deliberately not published (dynamic-quant parity 0.62 — static/QDQ calibration is the tracked follow-up).

View diff on GitHub

Weekly OSS security release digest.

The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.

No spam, unsubscribe anytime.

Share this release

Share on X Share on Bluesky

Track varun29ankuS/shodh-memory

Get notified when new releases ship.

About varun29ankuS/shodh-memory

Cognitive memory for AI agents with Hebbian learning, 3-tier architecture, and knowledge graphs. Single ~15MB binary, runs offline on edge devices.

All releases →