web-llm Releases

This release addresses a security vulnerability related to Cypher injection in the MCP server by bumping the dependency on graphiti-core to version 0.28.2 or higher.

Mar 11, 2026

This release prioritizes security by hardening search filters against Cypher injection, while also introducing the GLiNER2 hybrid LLM client and updating documentation.

Mar 11, 2026

This patch release fixes a critical hang issue in the video recording process, ensuring the recorder resolves properly after all steps complete.

Mar 2, 2026

Minimal release notes: "### Patch Changes - Add homepage and bugs URLs to package.json files. Fix README video embeds for GitHub LFS."

Mar 1, 2026

This version improves MP4 video encoding specifically for iPhone compatibility by adding BT.709 color metadata and the faststart flag, allowing videos to be saved to iOS photo galleries and play faster.

Mar 1, 2026

Minimal release notes: "### Patch Changes - Fix published package: resolve `workspace:*` dependency to actual version."

Mar 1, 2026

Minimal release notes: "### Minor Changes - Initial release. Record scripted browser demos as MP4, GIF, or WebM with cursor animation, keystrok"

Mar 1, 2026

Major release featuring Engine and API overhaul with ChatModule refactored into Engine/MLCEngine, consolidated constructor/reload behavior, multi-model loading, and improved worker lifecycle. Includes OpenAI API mirroring, expanded prebuilt model support across Llama, Mistral, Gemma, Qwen, and Phi families, WebGPU performance improvements, and XGrammar integration for JSON-schema constrained generation.

Feb 18, 2026

This release introduces several rendering and UI improvements including a free list strategy for sprite batches, new animation reset functionality, and fixes for layout size notifications and routing.

Feb 12, 2026

This patch release fixes a specific tap interaction issue where taps starting inside a component but ending outside were not being canceled correctly.

Feb 12, 2026

This is a major update to the PaddlePaddle framework focusing on training efficiency, development experience, and domestic hardware support, featuring significant upgrades to memory management and parallelism.

Jan 31, 2026

PaddlePaddle 3.2.2 focuses on enhancing performance and stability through optimizations in distributed training, operator mechanisms, and hardware adaptation. Major additions include new interfaces for model parameter management and support for HuggingFace checkpoint formats.

Dec 8, 2025

PaddlePaddle 3.2.0 significantly enhances large model training and inference capabilities with FlashMask V3 for attention efficiency and FP8 mixed precision training. It introduces a CUDA-like hardware adapter with high operator reuse rates and improves developer experience through extensive API compatibility with industry standards and Safetensors support.

Sep 8, 2025

PaddlePaddle 3.1.1 focuses on stabilizing large model training by fixing numerical precision issues, 0-size tensors, and distributed stability problems. It optimizes FP8 performance and introduces new APIs for device management and tensor type conversion, alongside support for Safetensors loading.

Aug 21, 2025

This release focuses on optimizing the framework for automatic parallelism and FP8 low-precision training, offering 10-20% performance improvements and enhanced hardware compatibility.

Jun 29, 2025

This release marks a significant milestone with a complete framework rewrite, introducing a fresh UI and numerous enhancements and bug fixes.

Feb 3, 2025

Major update focusing on UI enhancements, new language support, and a complete framework rewrite.

Sep 12, 2024

Maintenance release fixing bugs, adding custom themes, and optimizing images.

Jun 8, 2024

Minor update introducing tab cloaking features to the interface.

May 23, 2024

May 21, 2024