airllm Releases
145 releases of lyogavin/airllm
- v4.0.0AirLLM v4.0.0 — train Qwen3.8-Flash-Next under 6GB
This release introduces a memory-efficient training architecture known as Streamed LoRA, enabling the training of large models like Qwen3.8-Flash-Next on consumer-grade hardware with limited VRAM. It provides updated installation steps and specific usage examples for local dataset training.
Sep 5, 2026
- v3.3.0AirLLM v3.3.0 — Qwen3.8-Flash-Next on 5.95GB
This release focuses on optimizing the inference of the Qwen3.8-Flash-Next model, enabling it to run on a single GPU with minimal VRAM usage through advanced streaming techniques.
Aug 28, 2026
- v3.2.0AirLLM v3.2.0 — Qwen3.8-27B on 3.33GB
Aug 18, 2026
- v2.1.36
This release focuses on stability improvements for the Data Source Manager and LDAP authentication, preventing data clearing errors and handling complex UTF-8 syntax in user DNs.
Aug 5, 2026
- v2.1.35
A comprehensive release enhancing workflow capabilities with configurable post-success actions for batch updates and field search in import/export settings. It also resolves critical backend issues regarding database queries, worker management, and external data source synchronization.
Aug 5, 2026
- v2.8.2Seelen UI v2.8.2
This release introduces a new grid workspace system with navigation shortcuts and improves system memory management. It also fixes several UI bugs related to app menus, profile images, and keyboard navigation.
Aug 3, 2026
- v1.0.1
A maintenance release fixing documentation and CLI compatibility issues. It restores support for deprecated flags in the generator to ensure existing tutorials and guides work without error, while maintaining no behavioral changes to the framework runtime.
Jul 31, 2026
- v1.0.0v1.0.0 — Loco is stable
This is the first stable Loco release featuring a move to Sea-ORM 2.0, first-class LLM/agent support, priority queues, and a broad dependency modernization with significant breaking changes across queue, storage, config, and middleware subsystems.
Jul 29, 2026
- v3.1.0AirLLM v3.1.0 — Kimi K3 (2.8T) on a single card
Introduces support for the massive Kimi K3 (2.8T) model on a single consumer GPU while fixing general loading issues related to checkpoint handling and attention implementations.
Jul 29, 2026
- v2.8.1Seelen UI v2.8.1
This update focuses on UI refinements for the Seelen desktop environment, including configurable stack bar visibility, improved toolbar plugins, and various fixes for multimonitor setups and window management.
Jul 27, 2026
- 0.35.0
This release introduces an improved command palette for the Viewer, an experimental built-in catalog for local recordings, and enhanced data reading capabilities including HDF5 support and improved video/MCAP handling.
Jul 23, 2026
- v2.8.0Seelen UI v2.8.0
This release introduces a comprehensive plugins system for the dock and taskbar, adding specialized plugins for clock, network, bluetooth, notifications, and system tray, alongside enhancements for multimonitor support.
Jul 20, 2026
- 0.34.1
A maintenance release for the Rerun visualization toolbox focusing on bug fixes for data streaming, collection refreshing, and MP4 codec handling.
Jul 7, 2026
- 0.34.00.34.0 - Viewer MCP & Learning Course
A major feature release introducing an LLM agent integration (Viewer MCP), a new learning course, sparse voxel grid support, and gamepad navigation for 3D views.
Jul 6, 2026
- v0.9.6graphify 0.9.6
This release introduces Ruby container nodes and singleton-call resolution while fixing cross-language phantom imports and extractor bugs for Kotlin, Apex, and Swift.
Jul 4, 2026
- v0.9.5
Fixes regressions from 0.9.4 regarding indirect calls and community labels, enhances language extraction for several languages, and improves MCP server capabilities.
Jul 2, 2026
- v0.9.4
Introduces indirect call dispatch as a first-class dependency for various languages, fixes incremental update data issues, and improves robustness and error handling.
Jul 1, 2026
- v3.0.1AirLLM v3.0.1
A patch release that addresses installation failures by adding missing dependencies and improves the robustness of the package by making model-family imports defensive.
Jun 30, 2026
- v3.0.0
A major overhaul that enables inference of the largest open models (up to 405B parameters) on small GPUs, introduces native FP8 support, and ensures compatibility with modern Hugging Face libraries.
Jun 30, 2026
- v0.9.3
Adds cross-file member-call resolution for C++ and Objective-C, introduces a work-memory overlay for learning, and improves TypeScript and JavaScript resolution.
Jun 30, 2026