This release fixes issues for SREs watching stability and regressions.
✓ No known CVEs patched in this version
Summary
AI summaryFixed crash caused by Qwen3 KV cache accumulating tokens across sequential embedding requests.
Full changelog
🐛 Bug Fixes
- Embedding Engine (Qwen3 KV Cache): Fixed a crash (
Embedding error: shape mismatch in broadcast_add) that occurred when generating multiple embeddings sequentially. The Qwen3 KV cache was accumulating tokens across independent requests. Sinceclear_kv_cacheis private incandle-transformers 0.9.2, this is resolved by retaining the configuration andVarBuilderin memory, and reconstructing a freshQwen3Modelinstance for each forward pass. BecauseVarBuilderusesArcunder the hood, this reconstruction is virtually instant (zero-copy) and completely prevents cross-contamination of embeddings.
Full Changelog: https://github.com/pomazanbohdan/memory-mcp-1file/compare/v0.4.3...v0.4.4
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
About pomazanbohdan/memory-mcp-1file
A self-contained Memory server with single-binary architecture (embedded DB & models, no dependencies). Provides persistent semantic and graph-based memory for AI agents.
Related context
Beta — feedback welcome: [email protected]