Perplexity’s recent open-source projects span AI models, local inference and security tools

A post lists seven recent Perplexity open-source contributions, including two AI models, the Lily inference engine, the PII-Tracer classifier, a research-agent benchmark and two security tools. It claims Lily is 1.23 times faster at prefill and 1.35 times faster at decoding than MLX-LM on an M5 Max. A Perplexity employee said on October 3 that another contribution was coming the following week.