Roman | RyzenAdvanced รีทวีตแล้ว
We just uploaded our GLM-5's tech report onto arxiv. Hope it helpful! takeaway keywords:
- dsa arch with 750B parameter (40B active) + 30T data
- slime RL toolkit
- asynchronous agentic RL that is really helpful (+3 points improvements on several major benchmarks)
- adapt to seven Chinese chips
results:
- top #1 open model on AA (artificialanalysis.ai)
- top #1 open model on text arena and code arena
- top #1 open model on vending bench and design bench
todo:
-need improvements on infra
-need more GPUs
Z.ai@Zai_org
Presenting the GLM-5 Technical Report! arxiv.org/abs/2602.15763 After the launch of GLM-5, we’re pulling back the curtain on how it was built. Key innovations include: - DSA Adoption: Significantly reduces training and inference costs while preserving long-context fidelity - Asynchronous RL Infrastructure: Drastically improves post-training efficiency by decoupling generation from training - Agent RL Algorithms: Enables the model to learn from complex, long-horizon interactions more effectively Through these innovations, GLM-5 achieves SOTA performance among open-source models, with particularly strong results in real-world software engineering tasks.
English
