AI & LLMs agent skills is designed to orchestrate models, prompts, and AI workflows, ideal for builders who need smarter automation and better model outcomes.
Log ML experiments, visualize training in real-time, run hyperparameter sweeps, and manage model artifacts with Weights & Biases.
by NousResearch
Run local GGUF models with llama.cpp on any hardware — CPU, Apple Silicon, CUDA, ROCm, or Intel GPU — and discover models on Hugging Face Hub.
by NousResearch
Run a controlled Godmode skill workflow for authorized AI safety testing across Claude, Hermes, and other AI models.
by NousResearch
High-throughput LLM inference server with OpenAI-compatible API, PagedAttention, quantization (AWQ/GPTQ/FP8), and tensor parallelism for production deployments.
by NousResearch
Clinical Decision Support System patterns for drug interaction checking, dose validation, NEWS2/qSOFA scoring, and EMR alert integration — patient safety critical.
by affaan-m
Search, download, and upload models & datasets, manage repos, query datasets with SQL, deploy inference endpoints, and manage Spaces on Hugging Face Hub.
Remove refusal behaviors from open-weight LLMs without retraining, using mechanistic interpretability techniques like diff-in-means, SVD, and LEACE concept erasure.
by NousResearch
Benchmark LLMs across 60+ academic tasks (MMLU, GSM8K, HumanEval, HellaSwag) using the industry-standard EleutherAI evaluation harness.
by NousResearch
Expert guidance for fine-tuning LLMs with Axolotl — YAML configs, LoRA/QLoRA, DPO/KTO/ORPO/GRPO, multimodal support, and 100+ model architectures.
by NousResearch
Post-train and align LLMs using TRL — SFT, DPO, PPO, GRPO, and reward modeling for full RLHF pipelines.
by NousResearch
Get expert guidance on Unsloth for 2-5x faster LoRA/QLoRA fine-tuning with less VRAM — covering setup, APIs, debugging, and best practices.
by NousResearch
Build and auto-optimize declarative LM programs, RAG pipelines, and AI agents using DSPy — no manual prompt engineering required.
by NousResearch
Guarantee valid JSON, Pydantic, and regex outputs from local LLMs using FSM-based constrained token sampling — zero overhead, no retries needed.
by NousResearch