LangSmith Fine-Tuning launches smithtune CLI for end-to-end model training

LangSmith now offers smithtune, a CLI that turns agent trajectories into custom fine-tuned models. It handles data, training, and evaluation.

Key points

smithtune supports supervised fine-tuning using agent trajectories
It works with Fireworks and Baseten for training and deployment
Trajectories are recorded with exact context for accurate training
Evaluation compares base and fine-tuned models with built-in replay testing

Why it matters: This simplifies training specialized models, reducing cost and latency while improving task performance.

LangSmith has introduced a new tool called smithtune, a command-line interface (CLI) for fine-tuning models. It helps teams turn agent trajectories stored in LangSmith into custom models. The process includes dataset creation, training, and evaluation, all in one workflow.

smithtune supports supervised fine-tuning, which trains models based on examples of good behavior. It uses trajectories that record what the agent saw at each step, preserving context. This makes training more accurate, especially for complex tasks.

The CLI can pull data from LangSmith, filter and label it, and prepare it for training. It also splits data into training, validation, and test sets. Users can review and adjust training settings before starting the process.

Training runs on Fireworks or Baseten, which handle the infrastructure. smithtune monitors performance and saves the best model based on validation results. After training, it compares the fine-tuned model with the original base model using built-in evaluation tools.

This new tool makes it easier for teams to create specialized models without building data pipelines or managing infrastructure. It aims to lower costs and improve speed for custom AI solutions.