Waveforce

An independent local development lab focusing on structural knowledge distillation, precision alignments, and high-efficiency model convergence configurations. In this organization, we continuously build and release specialized large language models, high-performance architectures, and optimized distilled models engineered for maximum efficiency and speed.

Project Aura

We develop standalone, high-speed reasoning setups trained via direct Kullback-Leibler (KL) logit alignment distribution. Our architectures aim to extract the operational density of larger, native multimodal teacher models directly into compact, consumer-grade student baselines.

Core Focus Areas

• Structural Distillation: True logit-level cross-entropy probability curve alignment.
• Native Deployments: Outputting unquantized, high-precision safetensors weights optimized natively for edge hardware wrappers and advanced local runtimes.

Notice: Gated Models Require A HuggingFace Account With Access To The Model Repo Of The Model You Want To Access And A HuggingFace Access Token (HF_TOKEN)