Tsinghua Open-Sources Puro-2B: Low-Cost Pretraining on RTX 5090 Clusters

1 reports archived1 independent sourcesupdated 10/15/2026, 20:48:00

Synthesis & Latest Updates

1 Sources Cross-Validated

Tsinghua University released Puro-2B, a fully open-source low-cost pretraining pipeline. By utilizing an RTX 5090 cluster, the team trained a 2B-parameter model for approximately $4,400, drastically lowering the computational barrier for small teams to replicate large-scale LLM pretraining.

LATEST/Tsinghua University released Puro-2B, a fully open-source low-cost pretraining pipeline. By utilizing an RTX 5090 cluster, the team trained a 2B-parameter model for approximately $4,400, drastically lowering the computational barrier for small teams to replicate large-scale LLM pretraining.

HEAT TRENDHourly heat curve

31 fully observed hours

Now

6.1

Peak

11.921:00

24h change

-42%

The line compares only sources observed throughout; gap hours are interpolated to keep the trend continuous.

TIMELINECoverage timeline

Total 1 reports · Latest first

新智元 (微信公众号)T2·Oct 15, 20:4878 pts
Tsinghua Open-Sources Puro-2B: Low-Cost Pretraining on RTX 5090 Clusters
- Puro-2B provides a fully open-source pipeline (data, framework, weights) for reproducible low-cost pretraining.
- Achieved training of a 2B-parameter model on consumer-grade RTX 5090 clusters for only ~$4,400.
- Significantly reduces hardware and financial barriers compared to existing open-source alternatives like OLMo.
View Related Reports & Analysis Original Source