Thinking Machines Lab

Thinking Machines Lab / History

History

Jun 2026-Sep 2026

September

Tue

LoRA rank <8 may fail training on MoE models

12:24 AM

The issue has been fully resolved.

Mon

Training on Inkling could fail with variable-shape loss inputs — SDK 0.24.1 required

7:38 AM

SDK version 0.25.0+ is required to use Tinker API.

Tinker SDK may be getting stuck on MacOS

7:37 AM

SDK version 0.25.0+ is required to use Tinker API.

August

Tue

Temporary outage due to billiing backend

10:35 PM

The issue has been fully resolved.

All sampling was down 6:36PM PT - 7:01PM PT on 08/08

4:04 PM

The issue was resolved.

Wed

Getting model info can fail for old models

9:51 PM

This is fully resolved now.

Tinker sampling slow or stuck for most models

12:40 AM

The issue is fully resolved now.

July

Tue

Tinker sampling failures

4:32 AM

The problematic code was reverted, the issue is fully mitigated.

Sat

Degraded Nemotron and Qwen3.5/6 Sampling

2:55 PM

We are rolling out a fix that will restore any lingering reduction in prefill throughput

Fri

Degraded Training Capacity

11:34 PM

We have deployed a fix and capacity has recovered

Tue

Training Degraded

7:46 AM

This issue has been resolved

Thu

Tinker fine tuning sampling slow

6:04 AM

The issue is fully resolved now.

June

Tue

Several Models Down

5:41 PM

All models should be back up now!