model-surgery

Open the hood on neural networks. Component-level model surgery, analysis, and composition.

A local AI model training, distillation, model surgery, GGUF conversion, and Ollama export workbench. 本地 AI 模型训练、蒸馏、模型手术与导出工具箱。

Deterministic tensor patch plans, patch application, and tensor path traversal for neural network model surgery.

Direct weight surgery for LLMs — edit weights, not gradients. Extract the arrow, leave the body intact.

Public EPI research scaffold for energy-aware AI evaluation, measurement discipline, and claim boundaries.

🧠🔪 Digital lobotomy for local LLMs: cut layers, swap words, inject moods — and compare the patient with a healthy donor. Web UI, runs 100% locally.

End-to-end audio-native SLM fine-tuning & model surgery framework (Qwen3-Omni backbone) for Indic telephony speech-to-speech.

Turn any HF LLM into an attention+recurrent hybrid in minutes on one consumer GPU: per-layer surgery, teacher distillation, executable-exam gating.

Nx SVD/SVF factorization primitives for neural network model surgery and TRINITY artifact export.

SafeTensors parsing, validation, bounded slicing, checksums, deterministic writing, and row-chunk helpers for Elixir and Nx.

Parallel AI reverse-engineering system: orchestrates REA, Heretic and OBLITERATUS to ablate refusals, map refusal geometry, and gate on capability preservation.

Visual model merging, layer surgery, and weight interpolation studio for local LLMs