Axolotl v0.16 vs Tether QVAC SDK

Axolotl is the go-to open-source fine-tuning framework for the local LLM community, and v0.16 is its most significant performance release to date. The headline numbers are striking: 15x faster training for Mixture-of-Experts (MoE) models with LoRA adapters, 40x reduction in memory usage for the same configurations, and 58% faster GRPO async training — the algorithm behind many of the recent reasoning model breakthroughs. Day-0 support for Google Gemma 4 shipped simultaneously with the model release. The MoE+LoRA improvements are especially timely. As sparse mixture-of-experts models like Gemma 4, Mistral, and Qwen3.6-Plus dominate the model landscape, fine-tuning them has been disproportionately expensive. Axolotl v0.16 makes it practical to fine-tune these architectures on a single consumer GPU — previously a multi-GPU or cloud-required task. The GRPO improvements also make reinforcement learning from human feedback (RLHF) workflows dramatically faster for small teams. For the indie fine-tuning community — researchers, small companies, and hobbyists building specialized models — this release removes a major cost barrier. Combined with the simultaneous Gemma 4 support, v0.16 positions Axolotl as the fastest path from a new model release to a fine-tuned, production-ready custom variant.

Tether — yes, the stablecoin company — has launched QVAC, a fully open-source SDK for building on-device AI agents that work offline, peer-to-peer, and without any dependency on centralized cloud infrastructure. Built on a customized fork of llama.cpp called QVAC Fabric, it supports text completion, embeddings, vision, OCR, speech-to-text, text-to-speech, and translation — all running locally on Linux, macOS, Windows, Android, and iOS with a single unified API. What makes QVAC architecturally distinct is the Holepunch protocol stack underneath it: models can be distributed peer-to-peer, inference can be delegated across devices without centralized infrastructure, and the roadmap includes decentralized swarms for training and fine-tuning. Once a model is cached locally, the SDK works fully offline — making it suitable for air-gapped deployments, field work, and restricted-network environments. Tether is also running a developer grants program to fund projects building with QVAC, specifically targeting local-first AI and payment applications. With $27B+ in stablecoin reserves behind it, Tether has the runway to sustain a multi-year open-source effort here — which is more than most AI SDK projects can say.