Orchestra-Research/AI-Research-SKILLs

这个仓库里有 98 个技能，GitHub 星标 ★ 12,734。

• autoresearchOrchestrates end-to-end autonomous AI research projects using a two-loop archite
• implementing-llms-litgptImplements and trains LLMs using Lightning AI's LitGPT with 20+ pretrained archi
• mamba-architectureState-space model with O(n) complexity vs Transformers' O(n²). 5× faster inferen
• nanogptEducational GPT implementation in ~300 lines. Reproduces GPT-2 (124M) on OpenWeb
• rwkv-architectureRNN+Transformer hybrid with O(n) inference. Linear time, infinite context, no KV
• distributed-llm-pretraining-torchtitanProvides PyTorch-native distributed LLM pretraining using torchtitan with 4D par
• huggingface-tokenizersFast tokenizers optimized for research and production. Rust-based implementation
• sentencepieceLanguage-independent tokenizer treating text as raw Unicode. Supports BPE and Un
• axolotlExpert guidance for fine-tuning LLMs with Axolotl - YAML configs, 100+ models, L
• llama-factoryExpert guidance for fine-tuning LLMs with LLaMA-Factory - WebUI no-code, 100+ mo
• peft-fine-tuningParameter-efficient fine-tuning for LLMs using LoRA, QLoRA, and 25+ methods. Use
• unslothExpert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80%
• nnsight-remote-interpretabilityProvides guidance for interpreting and manipulating neural network internals usi
• pyvene-interventionsProvides guidance for performing causal interventions on PyTorch models using py
• sparse-autoencoder-trainingProvides guidance for training and analyzing Sparse Autoencoders (SAEs) using SA
• transformer-lens-interpretabilityProvides guidance for mechanistic interpretability research using TransformerLen
• nemo-curatorGPU-accelerated data curation for LLM training. Supports text/image/video/audio.
• ray-dataScalable data processing for ML workloads. Streaming execution across CPU/GPU, s
• grpo-rl-trainingExpert guidance for GRPO/RL fine-tuning with TRL for reasoning and task-specific
• miles-rl-trainingProvides guidance for enterprise-grade RL training using miles, a production-rea
• openrlhf-trainingHigh-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, R
• simpo-trainingSimple Preference Optimization for LLM alignment. Reference-free alternative to
• slime-rl-trainingProvides guidance for LLM post-training with RL using slime, a Megatron+SGLang f
• torchforge-rl-trainingProvides guidance for PyTorch-native agentic RL using torchforge, Meta's library
• fine-tuning-with-trlFine-tune LLMs using reinforcement learning with TRL - SFT for instruction tunin
• verl-rl-trainingProvides guidance for training LLMs with reinforcement learning using verl (Volc
• constitutional-aiAnthropic's method for training harmless AI through self-improvement. Two-phase
• llamaguardMeta's 7-8B specialized moderation model for LLM input/output filtering. 6 safet
• nemo-guardrailsNVIDIA's runtime safety framework for LLM applications. Features jailbreak detec
• prompt-guardMeta's 86M prompt injection and jailbreak detector. Filters malicious prompts an
• huggingface-accelerateSimplest distributed training API. 4 lines to add distributed support to any PyT
• deepspeedExpert guidance for distributed training with DeepSpeed - ZeRO optimization stag
• training-llms-megatronTrains large language models (2B-462B parameters) using NVIDIA Megatron-Core wit
• pytorch-fsdp2Adds PyTorch FSDP2 (fully_shard) to training scripts with correct init, sharding
• pytorch-lightningHigh-level PyTorch framework with Trainer class, automatic distributed training
• ray-trainDistributed training orchestration across clusters. Scales PyTorch/TensorFlow/Hu
• lambda-labs-gpu-cloudReserved and on-demand GPU cloud instances for ML training and inference. Use wh
• modal-serverless-gpuServerless GPU cloud platform for running ML workloads. Use when you need on-dem
• skypilot-multi-cloud-orchestrationMulti-cloud orchestration for ML workloads with automatic cost optimization. Use
• awq-quantizationActivation-aware weight quantization for 4-bit LLM compression with 3x speedup a
• quantizing-models-bitsandbytesQuantizes LLMs to 8-bit or 4-bit for 50-75% memory reduction with minimal accura
• optimizing-attention-flashOptimizes transformer attention with Flash Attention for 2-4x speedup and 10-20x
• gguf-quantizationGGUF format and llama.cpp quantization for efficient CPU/GPU inference. Use when
• gptqPost-training 4-bit quantization for LLMs with minimal accuracy loss. Use for de
• hqq-quantizationHalf-Quadratic Quantization for LLMs without calibration data. Use when quantizi
• ml-training-recipesBattle-tested PyTorch training recipes for all domains — LLMs, vision, diffusion
• evaluating-code-modelsEvaluates code generation models across HumanEval, MBPP, MultiPL-E, and 15+ benc
• evaluating-llms-harnessEvaluates LLMs across 60+ academic benchmarks (MMLU, HumanEval, GSM8K, TruthfulQ
• nemo-evaluator-sdkEvaluates LLMs across 100+ benchmarks from 18+ harnesses (MMLU, HumanEval, GSM8K
• llama-cppRuns LLM inference on CPU, Apple Silicon, and consumer GPUs without NVIDIA hardw
• sglangFast structured generation and serving for LLMs with RadixAttention prefix cachi
• tensorrt-llmOptimizes LLM inference with NVIDIA TensorRT for maximum throughput and lowest l
• serving-llms-vllmServes LLMs with high throughput using vLLM's PagedAttention and continuous batc
• mlflowTrack ML experiments, manage model registry with versioning, deploy models to pr
• experiment-tracking-swanlabProvides guidance for experiment tracking with SwanLab. Use when you need open-s
• tensorboardVisualize training metrics, debug models with histograms, compare experiments, v
• weights-and-biasesTrack ML experiments with automatic logging, visualize training in real-time, op
• evolving-ai-agentsProvides guidance for automatically evolving and optimizing AI agents across any
• autogpt-agentsAutonomous AI agent platform for building and deploying continuous agents. Use w
• crewai-multi-agentMulti-agent orchestration framework for autonomous AI collaboration. Use when bu
• langchainFramework for building LLM-powered applications with agents, chains, and RAG. Su
• llamaindexData framework for building LLM applications with RAG. Specializes in document i
• chromaOpen-source embedding database for AI applications. Store embeddings and metadat
• faissFacebook's library for efficient similarity search and clustering of dense vecto
• pineconeManaged vector database for production AI applications. Fully managed, auto-scal
• qdrant-vector-searchHigh-performance vector similarity search engine for RAG and semantic search. Us
• sentence-transformersFramework for state-of-the-art sentence, text, and image embeddings. Provides 50
• dspyBuild complex AI systems with declarative programming, optimize prompts automati
• guidanceControl LLM output with regex and grammars, guarantee valid JSON/XML/code genera
• instructorExtract structured data from LLM responses with Pydantic validation, retry faile
• outlinesGuarantee valid JSON/XML/code structure during generation, use Pydantic models f
• langsmith-observabilityLLM observability platform for tracing, evaluation, and monitoring. Use when deb
• phoenix-observabilityOpen-source AI observability platform for LLM tracing, evaluation, and monitorin
• audiocraft-audio-generationPyTorch library for audio generation including text-to-music (MusicGen) and text
• blip-2-vision-languageVision-language pre-training framework bridging frozen image encoders and LLMs.
• clipOpenAI's model connecting vision and language. Enables zero-shot image classific
• evaluating-cosmos-policyEvaluates NVIDIA Cosmos Policy on LIBERO and RoboCasa simulation environments. U
• llavaLarge Language and Vision Assistant. Enables visual instruction tuning and image
• fine-tuning-serving-openpiFine-tune and serve Physical Intelligence OpenPI models (pi0, pi0-fast, pi0.5) u
• fine-tuning-openvla-oftFine-tunes and evaluates OpenVLA-OFT and OpenVLA-OFT+ policies for robot action
• segment-anything-modelFoundation model for image segmentation with zero-shot transfer. Use when you ne
• stable-diffusion-image-generationState-of-the-art text-to-image generation with Stable Diffusion models via Huggi
• whisperOpenAI's general-purpose speech recognition model. Supports 99 languages, transc
• knowledge-distillationCompress large language models using knowledge distillation from teacher to stud
• long-contextExtend context windows of transformer models using RoPE, YaRN, ALiBi, and positi
• model-mergingMerge multiple fine-tuned models using mergekit to combine capabilities without
• model-pruningReduce LLM size and accelerate inference using pruning techniques like Wanda and
• moe-trainingTrain Mixture of Experts (MoE) models using DeepSpeed or HuggingFace. Use when t
• speculative-decodingAccelerate LLM inference using speculative decoding, Medusa multiple heads, and
• academic-plottingGenerates publication-quality figures for ML papers from research context. Given
• ml-paper-writingWrite publication-ready ML/AI papers for NeurIPS, ICML, ICLR, ACL, AAAI, COLM. U
• presenting-conference-talksGenerates conference presentation slides (Beamer LaTeX PDF and editable PPTX) fr
• systems-paper-writingComprehensive guide for writing systems papers targeting OSDI, SOSP, ASPLOS, NSD
• brainstorming-research-ideasGuides researchers through structured ideation frameworks to discover high-impac
• creative-thinking-for-researchApplies cognitive science frameworks for creative thinking to CS and AI research
• ara-compilerCompiles any research input — PDF papers, GitHub repositories, experiment logs,
• ara-research-managerRecords research provenance as a post-task epilogue, scanning conversation histo
• ara-rigor-reviewerPerforms ARA Seal Level 2 semantic epistemic review on Agent-Native Research Art