Enables comparison of responses from multiple LLMs (OpenAI, Anthropic, Gemini) to the same prompt, returning a validated divergence score based on sentence embeddings.
Enables classification of GPU workloads as training, inference, or idle from telemetry data, with tools for one-shot classification, benchmarking, and audit log verification.
MCP server for analyzing transformer language model internals, exposing tools to trace attention heads and neurons responsible for predictions, get activation statistics, ablate components, and sketch circuits.