Datadog and Baseten fine-tune Qwen 3.5 9B for a 50x faster LLM summarizer

baseten · x · 2026-11-19

Baseten partnered with Datadog to make the Patterns LLM topic-clustering summarizer 50x faster by fine-tuning Qwen 3.5 9B on production traces: latency dropped from 44.7s to 0.89s and judge pass rate rose from 45% to 93%.

They open-sourced pup-finetune (Apache-2.0), which curates datasets from Datadog Agent Observability trajectories (messages, tool calls, results), fine-tunes an open-weight model, and evaluates it against the base model — with a datadog-finetune skill so your coding agent can run the whole workflow.

Original post →

More from coding & agent

coding & agent channel →