@skills · Owner
Agent skills by zechenzhangagi
121 skills indexed from github.com/zechenzhangagi. Reference any of them in AdaL, Claude Code, Cursor or any coding agent — nothing to install.
- 01-model-architecture · Collection · 11,472 stars
- 02-tokenization · Collection · 11,472 stars
- 03-fine-tuning · Collection · 11,472 stars
- 04-mechanistic-interpretability · Collection · 11,472 stars
- 05-data-processing · Collection · 11,472 stars
- 06-post-training · Collection · 11,472 stars
- 07-safety-alignment · Collection · 11,472 stars
- 08-distributed-training · Collection · 11,472 stars
- 09-infrastructure · Collection · 11,472 stars
- 10-optimization · Collection · 11,472 stars
- 11-evaluation · Collection · 11,472 stars
- 12-inference-serving · Collection · 11,472 stars
- 13-mlops · Collection · 11,472 stars
- 14-agents · Collection · 11,472 stars
- 15-rag · Collection · 11,472 stars
- 16-prompt-engineering · Collection · 11,472 stars
- 17-observability · Collection · 11,472 stars
- 18-multimodal · Collection · 11,472 stars
- 19-emerging-techniques · Collection · 11,472 stars
- 20-ml-paper-writing · Collection · 11,472 stars
- 21-research-ideation · Collection · 11,472 stars
- 22-agent-native-research-artifact · Collection · 11,472 stars
- academic-plotting · Skill · 11,472 stars
Generates publication-quality figures for ML papers from research context. Given a paper section or description, extracts system components and relationshi
- ai-research-skills · Collection · 11,472 stars
- ara-compiler · Skill · 11,472 stars
Compiles any research input — PDF papers, GitHub repositories, experiment logs, code directories, or raw notes — into a complete Agent-Native Research Arti
- ara-research-manager · Skill · 11,472 stars
Records research provenance as a post-task epilogue, scanning conversation history at the end of a coding or research session to extract decisions, experim
- ara-rigor-reviewer · Skill · 11,472 stars
Performs ARA Seal Level 2 semantic epistemic review on Agent-Native Research Artifacts, scoring six dimensions (evidence relevance, falsifiability, scope c
- audiocraft-audio-generation · Skill · 11,472 stars
PyTorch library for audio generation including text-to-music (MusicGen) and text-to-sound (AudioGen). Use when you need to generate music from text descrip
- autogpt-agents · Skill · 11,472 stars
Autonomous AI agent platform for building and deploying continuous agents. Use when creating visual workflow agents, deploying persistent autonomous agents
- autoresearch · Skill · 11,472 stars
Orchestrates end-to-end autonomous AI research projects using a two-loop architecture. The inner loop runs rapid experiment iterations with clear optimizat
- awq-quantization · Skill · 11,472 stars
Activation-aware weight quantization for 4-bit LLM compression with 3x speedup and minimal accuracy loss. Use when deploying large models (7B-70B) on limit
- axolotl · Skill · 11,472 stars
Expert guidance for fine-tuning LLMs with Axolotl - YAML configs, 100+ models, LoRA/QLoRA, DPO/KTO/ORPO/GRPO, multimodal support
- blip-2-vision-language · Skill · 11,472 stars
Vision-language pre-training framework bridging frozen image encoders and LLMs. Use when you need image captioning, visual question answering, image-text r
- brainstorming-research-ideas · Skill · 11,472 stars
Guides researchers through structured ideation frameworks to discover high-impact research directions. Use when exploring new problem spaces, pivoting betw
- chroma · Skill · 11,472 stars
Open-source embedding database for AI applications. Store embeddings and metadata, perform vector and full-text search, filter by metadata. Simple 4-functi
- clip · Skill · 11,472 stars
OpenAI's model connecting vision and language. Enables zero-shot image classification, image-text matching, and cross-modal retrieval. Trained on 400M imag
- constitutional-ai · Skill · 11,472 stars
Anthropic's method for training harmless AI through self-improvement. Two-phase approach - supervised learning with self-critique/revision, then RLAIF (RL
- creative-thinking-for-research · Skill · 11,472 stars
Applies cognitive science frameworks for creative thinking to CS and AI research ideation. Use when seeking genuinely novel research directions by leveragi
- crewai-multi-agent · Skill · 11,472 stars
Multi-agent orchestration framework for autonomous AI collaboration. Use when building teams of specialized agents working together on complex tasks, when
- deepspeed · Skill · 11,472 stars
Expert guidance for distributed training with DeepSpeed - ZeRO optimization stages, pipeline parallelism, FP16/BF16/FP8, 1-bit Adam, sparse attention
- distributed-llm-pretraining-torchtitan · Skill · 11,472 stars
Provides PyTorch-native distributed LLM pretraining using torchtitan with 4D parallelism (FSDP2, TP, PP, CP). Use when pretraining Llama 3.1, DeepSeek V3,
- dspy · Skill · 11,472 stars
Build complex AI systems with declarative programming, optimize prompts automatically, create modular RAG systems and agents with DSPy - Stanford NLP's fra
- evaluating-code-models · Skill · 11,472 stars
Evaluates code generation models across HumanEval, MBPP, MultiPL-E, and 15+ benchmarks with pass@k metrics. Use when benchmarking code models, comparing co
- evaluating-cosmos-policy · Skill · 11,472 stars
Evaluates NVIDIA Cosmos Policy on LIBERO and RoboCasa simulation environments. Use when setting up cosmos-policy for robot manipulation evaluation, running
- evaluating-llms-harness · Skill · 11,472 stars
Evaluates LLMs across 60+ academic benchmarks (MMLU, HumanEval, GSM8K, TruthfulQA, HellaSwag). Use when benchmarking model quality, comparing models, repor
- evolving-ai-agents · Skill · 11,472 stars
Provides guidance for automatically evolving and optimizing AI agents across any domain using LLM-driven evolution algorithms. Use when building self-impro
- experiment-tracking-swanlab · Skill · 11,472 stars
Provides guidance for experiment tracking with SwanLab. Use when you need open-source run tracking, local or self-hosted dashboards, and lightweight media
- faiss · Skill · 11,472 stars
Facebook's library for efficient similarity search and clustering of dense vectors. Supports billions of vectors, GPU acceleration, and various index types
- fine-tuning-openvla-oft · Skill · 11,472 stars
Fine-tunes and evaluates OpenVLA-OFT and OpenVLA-OFT+ policies for robot action generation with continuous action heads, LoRA adaptation, and FiLM conditio
- fine-tuning-serving-openpi · Skill · 11,472 stars
Fine-tune and serve Physical Intelligence OpenPI models (pi0, pi0-fast, pi0.5) using JAX or PyTorch backends for robot policy inference across ALOHA, DROID
- fine-tuning-with-trl · Skill · 11,472 stars
Fine-tune LLMs using reinforcement learning with TRL - SFT for instruction tuning, DPO for preference alignment, PPO/GRPO for reward optimization, and rewa
- gguf-quantization · Skill · 11,472 stars
GGUF format and llama.cpp quantization for efficient CPU/GPU inference. Use when deploying models on consumer hardware, Apple Silicon, or when needing flex
- gptq · Skill · 11,472 stars
Post-training 4-bit quantization for LLMs with minimal accuracy loss. Use for deploying large models (70B, 405B) on consumer GPUs, when you need 4× memory
- grpo-rl-training · Skill · 11,472 stars
Expert guidance for GRPO/RL fine-tuning with TRL for reasoning and task-specific model training
- guidance · Skill · 11,472 stars
Control LLM output with regex and grammars, guarantee valid JSON/XML/code generation, enforce structured formats, and build multi-step workflows with Guida
- hqq-quantization · Skill · 11,472 stars
Half-Quadratic Quantization for LLMs without calibration data. Use when quantizing models to 4/3/2-bit precision without needing calibration datasets, for
- huggingface-accelerate · Skill · 11,472 stars
Simplest distributed training API. 4 lines to add distributed support to any PyTorch script. Unified API for DeepSpeed/FSDP/Megatron/DDP. Automatic device
- huggingface-tokenizers · Skill · 11,472 stars
Fast tokenizers optimized for research and production. Rust-based implementation tokenizes 1GB in <20 seconds. Supports BPE, WordPiece, and Unigram algorit
- implementing-llms-litgpt · Skill · 11,472 stars
Implements and trains LLMs using Lightning AI's LitGPT with 20+ pretrained architectures (Llama, Gemma, Phi, Qwen, Mistral). Use when need clean model impl
- instructor · Skill · 11,472 stars
Extract structured data from LLM responses with Pydantic validation, retry failed extractions automatically, parse complex JSON with type safety, and strea
- knowledge-distillation · Skill · 11,472 stars
Compress large language models using knowledge distillation from teacher to student models. Use when deploying smaller models with retained performance, tr
- lambda-labs-gpu-cloud · Skill · 11,472 stars
Reserved and on-demand GPU cloud instances for ML training and inference. Use when you need dedicated GPU instances with simple SSH access, persistent file
- langchain · Skill · 11,472 stars
Framework for building LLM-powered applications with agents, chains, and RAG. Supports multiple providers (OpenAI, Anthropic, Google), 500+ integrations, R
- langsmith-observability · Skill · 11,472 stars
LLM observability platform for tracing, evaluation, and monitoring. Use when debugging LLM applications, evaluating model outputs against datasets, monitor
- llama-cpp · Skill · 11,472 stars
Runs LLM inference on CPU, Apple Silicon, and consumer GPUs without NVIDIA hardware. Use for edge deployment, M1/M2/M3 Macs, AMD/Intel GPUs, or when CUDA i
- llama-factory · Skill · 11,472 stars
Expert guidance for fine-tuning LLMs with LLaMA-Factory - WebUI no-code, 100+ models, 2/3/4/5/6/8-bit QLoRA, multimodal support
- llamaguard · Skill · 11,472 stars
Meta's 7-8B specialized moderation model for LLM input/output filtering. 6 safety categories - violence/hate, sexual content, weapons, substances, self-har
- llamaindex · Skill · 11,472 stars
Data framework for building LLM applications with RAG. Specializes in document ingestion (300+ connectors), indexing, and querying. Features vector indices
- llava · Skill · 11,472 stars
Large Language and Vision Assistant. Enables visual instruction tuning and image-based conversations. Combines CLIP vision encoder with Vicuna/LLaMA langua
- long-context · Skill · 11,472 stars
Extend context windows of transformer models using RoPE, YaRN, ALiBi, and position interpolation techniques. Use when processing long documents (32k-128k+
- mamba-architecture · Skill · 11,472 stars
State-space model with O(n) complexity vs Transformers' O(n²). 5× faster inference, million-token sequences, no KV cache. Selective SSM with hardware-aware
- miles-rl-training · Skill · 11,472 stars
Provides guidance for enterprise-grade RL training using miles, a production-ready fork of slime. Use when training large MoE models with FP8/INT4, needing
- ml-paper-writing · Skill · 11,472 stars
Write publication-ready ML/AI papers for NeurIPS, ICML, ICLR, ACL, AAAI, COLM. Use when drafting papers from research repos, structuring arguments, verifyi
- ml-training-recipes · Skill · 11,472 stars
Battle-tested PyTorch training recipes for all domains — LLMs, vision, diffusion, medical imaging, protein/drug discovery, spatial omics, genomics. Covers
- mlflow · Skill · 11,472 stars
Track ML experiments, manage model registry with versioning, deploy models to production, and reproduce experiments with MLflow - framework-agnostic ML lif
- modal-serverless-gpu · Skill · 11,472 stars
Serverless GPU cloud platform for running ML workloads. Use when you need on-demand GPU access without infrastructure management, deploying ML models as AP
- model-merging · Skill · 11,472 stars
Merge multiple fine-tuned models using mergekit to combine capabilities without retraining. Use when creating specialized models by blending domain-specifi
- model-pruning · Skill · 11,472 stars
Reduce LLM size and accelerate inference using pruning techniques like Wanda and SparseGPT. Use when compressing models without retraining, achieving 50% s
- moe-training · Skill · 11,472 stars
Train Mixture of Experts (MoE) models using DeepSpeed or HuggingFace. Use when training large-scale models with limited compute (5× cost reduction vs dense
- nanogpt · Skill · 11,472 stars
Educational GPT implementation in ~300 lines. Reproduces GPT-2 (124M) on OpenWebText. Clean, hackable code for learning transformers. By Andrej Karpathy. P
- nemo-curator · Skill · 11,472 stars
GPU-accelerated data curation for LLM training. Supports text/image/video/audio. Features fuzzy deduplication (16× faster), quality filtering (30+ heuristi
- nemo-evaluator-sdk · Skill · 11,472 stars
Evaluates LLMs across 100+ benchmarks from 18+ harnesses (MMLU, HumanEval, GSM8K, safety, VLM) with multi-backend execution. Use when needing scalable eval
- nemo-guardrails · Skill · 11,472 stars
NVIDIA's runtime safety framework for LLM applications. Features jailbreak detection, input/output validation, fact-checking, hallucination detection, PII
- nnsight-remote-interpretability · Skill · 11,472 stars
Provides guidance for interpreting and manipulating neural network internals using nnsight with optional NDIF remote execution. Use when needing to run int
- openrlhf-training · Skill · 11,472 stars
High-performance RLHF framework with Ray+vLLM acceleration. Use for PPO, GRPO, RLOO, DPO training of large models (7B-70B+). Built on Ray, vLLM, ZeRO-3. 2×
- optimizing-attention-flash · Skill · 11,472 stars
Optimizes transformer attention with Flash Attention for 2-4x speedup and 10-20x memory reduction. Use when training/running transformers with long sequenc
- outlines · Skill · 11,472 stars
Guarantee valid JSON/XML/code structure during generation, use Pydantic models for type-safe outputs, support local models (Transformers, vLLM), and maximi
- peft-fine-tuning · Skill · 11,472 stars
Parameter-efficient fine-tuning for LLMs using LoRA, QLoRA, and 25+ methods. Use when fine-tuning large models (7B-70B) with limited GPU memory, when you n
- phoenix-observability · Skill · 11,472 stars
Open-source AI observability platform for LLM tracing, evaluation, and monitoring. Use when debugging LLM applications with detailed traces, running evalua
- pinecone · Skill · 11,472 stars
Managed vector database for production AI applications. Fully managed, auto-scaling, with hybrid search (dense + sparse), metadata filtering, and namespace
- presenting-conference-talks · Skill · 11,472 stars
Generates conference presentation slides (Beamer LaTeX PDF and editable PPTX) from a compiled paper with speaker notes and talk script. Use when preparing
- prompt-guard · Skill · 11,472 stars
Meta's 86M prompt injection and jailbreak detector. Filters malicious prompts and third-party data for LLM apps. 99%+ TPR, <1% FPR. Fast (<2ms GPU). Multil
- pytorch-fsdp2 · Skill · 11,472 stars
Adds PyTorch FSDP2 (fully_shard) to training scripts with correct init, sharding, mixed precision/offload config, and distributed checkpointing. Use when m
- pytorch-lightning · Skill · 11,472 stars
High-level PyTorch framework with Trainer class, automatic distributed training (DDP/FSDP/DeepSpeed), callbacks system, and minimal boilerplate. Scales fro
- pyvene-interventions · Skill · 11,472 stars
Provides guidance for performing causal interventions on PyTorch models using pyvene's declarative intervention framework. Use when conducting causal traci
- qdrant-vector-search · Skill · 11,472 stars
High-performance vector similarity search engine for RAG and semantic search. Use when building production RAG systems requiring fast nearest neighbor sear
- quantizing-models-bitsandbytes · Skill · 11,472 stars
Quantizes LLMs to 8-bit or 4-bit for 50-75% memory reduction with minimal accuracy loss. Use when GPU memory is limited, need to fit larger models, or want
- ray-data · Skill · 11,472 stars
Scalable data processing for ML workloads. Streaming execution across CPU/GPU, supports Parquet/CSV/JSON/images. Integrates with Ray Train, PyTorch, Tensor
- ray-train · Skill · 11,472 stars
Distributed training orchestration across clusters. Scales PyTorch/TensorFlow/HuggingFace from laptop to 1000s of nodes. Built-in hyperparameter tuning wit
- rwkv-architecture · Skill · 11,472 stars
RNN+Transformer hybrid with O(n) inference. Linear time, infinite context, no KV cache. Train like GPT (parallel), infer like RNN (sequential). Linux Found
- segment-anything-model · Skill · 11,472 stars
Foundation model for image segmentation with zero-shot transfer. Use when you need to segment any object in images using points, boxes, or masks as prompts
- sentence-transformers · Skill · 11,472 stars
Framework for state-of-the-art sentence, text, and image embeddings. Provides 5000+ pre-trained models for semantic similarity, clustering, and retrieval.
- sentencepiece · Skill · 11,472 stars
Language-independent tokenizer treating text as raw Unicode. Supports BPE and Unigram algorithms. Fast (50k sentences/sec), lightweight (6MB memory), deter
- serving-llms-vllm · Skill · 11,472 stars
Serves LLMs with high throughput using vLLM's PagedAttention and continuous batching. Use when deploying production LLM APIs, optimizing inference latency/
- sglang · Skill · 11,472 stars
Fast structured generation and serving for LLMs with RadixAttention prefix caching. Use for JSON/regex outputs, constrained decoding, agentic workflows wit
- simpo-training · Skill · 11,472 stars
Simple Preference Optimization for LLM alignment. Reference-free alternative to DPO with better performance (+6.4 points on AlpacaEval 2.0). No reference m
- skypilot-multi-cloud-orchestration · Skill · 11,472 stars
Multi-cloud orchestration for ML workloads with automatic cost optimization. Use when you need to run training or batch jobs across multiple clouds, levera
- slime-rl-training · Skill · 11,472 stars
Provides guidance for LLM post-training with RL using slime, a Megatron+SGLang framework. Use when training GLM models, implementing custom data generation
- sparse-autoencoder-training · Skill · 11,472 stars
Provides guidance for training and analyzing Sparse Autoencoders (SAEs) using SAELens to decompose neural network activations into interpretable features.
- speculative-decoding · Skill · 11,472 stars
Accelerate LLM inference using speculative decoding, Medusa multiple heads, and lookahead decoding techniques. Use when optimizing inference speed (1.5-3.6
- stable-diffusion-image-generation · Skill · 11,472 stars
State-of-the-art text-to-image generation with Stable Diffusion models via HuggingFace Diffusers. Use when generating images from text prompts, performing
- systems-paper-writing · Skill · 11,472 stars
Comprehensive guide for writing systems papers targeting OSDI, SOSP, ASPLOS, NSDI, and EuroSys. Provides paragraph-level structural blueprints, writing pat
- tensorboard · Skill · 11,472 stars
Visualize training metrics, debug models with histograms, compare experiments, visualize model graphs, and profile performance with TensorBoard - Google's
- tensorrt-llm · Skill · 11,472 stars
Optimizes LLM inference with NVIDIA TensorRT for maximum throughput and lowest latency. Use for production deployment on NVIDIA GPUs (A100/H100), when you
- torchforge-rl-training · Skill · 11,472 stars
Provides guidance for PyTorch-native agentic RL using torchforge, Meta's library separating infra from algorithms. Use when you want clean RL abstractions,
- training-llms-megatron · Skill · 11,472 stars
Trains large language models (2B-462B parameters) using NVIDIA Megatron-Core with advanced parallelism strategies. Use when training models >1B parameters,
- transformer-lens-interpretability · Skill · 11,472 stars
Provides guidance for mechanistic interpretability research using TransformerLens to inspect and manipulate transformer internals via HookPoints and activa
- unsloth · Skill · 11,472 stars
Expert guidance for fast fine-tuning with Unsloth - 2-5x faster training, 50-80% less memory, LoRA/QLoRA optimization
- verl-rl-training · Skill · 11,472 stars
Provides guidance for training LLMs with reinforcement learning using verl (Volcano Engine RL). Use when implementing RLHF, GRPO, PPO, or other RL algorith
- weights-and-biases · Skill · 11,472 stars
Track ML experiments with automatic logging, visualize training in real-time, optimize hyperparameters with sweeps, and manage model registry with W&B - co
- whisper · Skill · 11,472 stars
OpenAI's general-purpose speech recognition model. Supports 99 languages, transcription, translation to English, and language identification. Six model siz