Baseten ramps post‑training
Baseten pushed truss v0.15.8 and posted a Post‑Training Research Engineer role on Mar 23 — a clear near‑term signal of increased fine‑tuning/post‑training work. The company is also named as a customer in a recent RunSybil funding story, suggesting they already buy third‑party infra. (github.com) (linkedin.com) (techcompanynews.com)
PyPI shows the truss CLI package at version 0.15.7 with a release timestamp of Mar 17, 2026, confirming recent packaging activity ahead of any later tags. (pypi.org) Baseten’s docs and repo describe Truss as a CLI that packages model serving logic, handles containerization and dependency management, exposes GPU/memory allocation in config.yaml, and includes commands for launching training jobs (truss-train). (docs.baseten.co) Job listings aggregated on Baseten’s careers/ATS and industry job boards describe the Post‑Training Research Scientist/Engineer role as focused on post‑training methodology, performant inference, translating research into production systems, and running experiments at multi‑node scale for 1T+ parameter models. (jobs.ashbyhq.com) Commercial job aggregators show the posting live across sites and list pay ranges; ZipRecruiter surfaces an estimated salary band of roughly $210,000–$285,000 for the Post‑Training Research Scientist role. (ziprecruiter.com) RunSybil’s March 18, 2026 funding announcement — a $40M round led by Khosla Ventures — explicitly named Baseten among early customers, which corroborates that Baseten purchases third‑party AI/security infra. (runsybil.com) Baseten’s corporate site and case studies highlight both inference‑optimized infrastructure and a “Run Training” surface with single‑tenant and self‑hosted deployment options, indicating product-level support for integrating post‑training and fine‑tuning workflows into customer deployments. (baseten.co)
