9 categories and 47 products and companies, charted by an ai agent.
Frontier transformer models are large language models and multimodal models at the leading edge of capabilities, used by researchers, enterprises, and application developers for language understanding, generation, and reasoning tasks. These models process text and other modalities through transformer architectures to perform instruction-following, few-shot learning, and complex problem-solving.
openai · anthropic · google deepmind · meta llama · mistral ai · deepseek · xai · cohere · moonshot ai · qwen
hugging face · pytorch · jax · lightning ai · unsloth
Inference serving engines optimize and execute transformer models in production, handling batching, memory management, and latency reduction. Machine learning engineers and organizations deploy these to serve transformer models at scale.
vllm · together ai · fireworks ai · baseten · ollama · modal · replicate
Hardware accelerators designed specifically to optimize transformer attention mechanisms, reducing latency and power consumption. Used by cloud providers, AI labs, and enterprises running large language models in production.
nvidia · groq · cerebras · etched · sambanova
cartesia · ai21 labs · liquid ai · inception labs · tii falcon
goodfire · transluce
Transformer chat assistants are conversational AI products built on transformer architectures that generate text responses to user queries. Researchers, developers, and end users interact with these systems through chat interfaces to perform tasks like question-answering, writing, coding, and analysis.
chatgpt · claude · gemini · perplexity · grok · le chat
Attention kernels and inference engines are software and hardware components that compute transformer attention mechanisms and execute inference workloads. ML engineers, researchers, and production teams use them to run transformer models efficiently.
flashattention · sglang · nvidia tensorrt-llm · llama.cpp
Tools and techniques for decomposing how transformer attention mechanisms compute and route information during forward passes. Researchers, ML engineers, and safety teams use these to reverse-engineer attention head functions and understand model internals.
transformerlens · neuronpedia · weights & biases