p-e-w/heretic
- Source
- GitHub
- First trending
- Category
- Model inference
- GitHub stars
- 31,616
- Main language
- Python
This page introduces an external open-source repository. It is not an HDATF product.
What it does
A command-line tool that removes refusal behavior, called safety alignment, from transformer language models using directional ablation. An Optuna-based optimizer picks parameters that reduce refusals while keeping the model close to the original.
How it helps ATF
Not related to our products. Listed because it appeared in the ranked lists.
License
AGPL-3.0 Network copyleft. Offering a modified version as a service also requires releasing the source. Review before any product use.
More in this category
- AlexsJones/llmfit
llmfit checks a machine's CPU, RAM, GPU and VRAM and recommends which open-source LLMs and quantizations it can run. It offers a terminal UI, a web dashboard, a REST API, and local benchmarks of real tokens per second. - RunanywhereAI/runanywhere-sdks
SDKs built on one shared C++ core that run LLMs, vision, speech, voice agents, embeddings, RAG and image generation locally on phones, browsers, desktops and servers. A capability registry routes each call to an engine on the device. - maderix/ANE
A research proof of concept that trains neural networks directly on Apple's Neural Engine through reverse-engineered private APIs, without CoreML training APIs, Metal or the GPU. The author states it is not a maintained framework. - microsoft/BitNet
bitnet.cpp is the official inference framework for 1-bit LLMs, with optimized kernels for CPUs and GPUs. The project has also released 1-bit embedding models and a guide for converting and running them. - karpathy/nanochat
A minimal codebase for training and chatting with your own LLM on a single GPU node, covering tokenization, pretraining, finetuning, evaluation and inference. One depth setting determines the other hyperparameters.
Only repositories in the ranked Trendshift lists are included, and the lists are used only to find candidates. We do not copy their ranks. Descriptions, licenses and star counts come from each GitHub repository. The notes are our own reading. We have not tested these projects, and a place on a trending list does not prove quality.