Software | AI Infrastructure Database

Note sul campo

L'hardware fa notizia, ma il lock-in vive nel software. Una piattaforma qui è abbastanza vecchia da votare e decide ancora quali chip la gente acquista; le voci più recenti hanno a malapena due anni e sono già portanti per l'era degli agenti.

Inizia da questi tre

CU CUDA Due decenni di librerie — il fossato più profondo della tecnologia. VL vLLM Il motore open che serve una quota enorme dei token mondiali. MC Model Context Protocol (MCP) La porta universale dell'era degli agenti, adottata dall'intero settore in meno di un anno.

Compute platforms & kernels

CU CUDA NVIDIA's programming platform — the deepest moat in the industry. Attivo NC NCCL The collective-communications library every distributed training run calls. Attivo RO ROCm AMD's open-source answer to CUDA — the gap that narrowed at last. Attivo T( Triton (OpenAI) Python-level GPU kernels — the community's crowbar under the CUDA moat. Attivo X/ XLA / OpenXLA The compiler that makes TPUs programmable — now an open project. Attivo

Framework di addestramento

PY PyTorch The framework nearly every modern model is written in. Attivo DE DeepSpeed Microsoft's scaling toolkit — ZeRO sharding made huge models trainable. Attivo

JAX Composable function transforms — the framework Gemini is built with. Attivo MC Megatron-Core NVIDIA's reference recipe for trillion-parameter training. Attivo

TensorFlow The 2015 giant that industrialized deep learning — now in JAX's shadow. Mantenuto

Inference & serving

VL vLLM The open-source serving engine that made GPU inference efficient. Attivo OR ONNX Runtime The portable format for shipping models across hardware. Attivo OL Ollama One command to run an open model locally. Attivo SG SGLang The high-performance serving runtime born from RadixAttention research. Attivo TL TensorRT-LLM & Dynamo NVIDIA's tuned inference stack — peak tokens from its own silicon. Attivo LL llama.cpp The C++ project that put LLMs on laptops — and invented GGUF. Attivo

Orchestration & cluster management

Kubernetes The cloud's operating system, retrofitted for GPUs. Attivo RA Ray Distributed Python for the whole ML pipeline — Anyscale's engine. Attivo SL Slurm The HPC batch scheduler most big training runs still queue through. Attivo

Libraries, tooling & the agent layer

MC Model Context Protocol (MCP) The open standard connecting AI models to tools and data — USB-C for agents. Attivo HF Hugging Face Transformers The library that standardized how models are shared and loaded. Attivo L& LangChain & LangGraph The scaffolding layer of the agent era. Attivo W& Weights & Biases Experiment tracking for the training runs on this site — now CoreWeave-owned. Attivo