MLX-LM-LoRA 3.0 Released
HankYeomans · x · 2026-07-15
MLX-LM-LoRA version 3.0 has been released, featuring major updates:
- Dynamic Fine-Tuning (DFT)
- Final Token Preference Optimization (FTPO / Antidoom)
- More memory-efficient Chunked NLL training
- Expanded test coverage
- A leaner architecture
The post also provides the installation command pip install -U mlx-lm-lora and an example for LoRA training using mlxlmlora.train, highlighting that this update introduces substantial new training strategies and engineering optimizations rather than just a version bump.
More from coding & agent
- Google launches Gemini 3.5 Flash Cyber for CodeMender, with limited access for governments — GoogleAI · 2026-07-22
- Cursor doubles usage limits across all plans for Grok, Composer and new models — XFreeze · 2026-07-22
- Video-based proof of work is emerging as a feedback layer for coding agents — Vjeux · 2026-07-22
- Salesforce open-sources MCP+ to filter tool output before it reaches agents — msrivastav13 · 2026-07-22
- Anaconda acquires Kilo Code to expand into agentic software development — anacondainc · 2026-07-22
- Poolside’s laguna s 2.1 claims 118B MoE, 1M context, and 78.5% on SWE-bench Multilingual — ben_burtenshaw · 2026-07-22