Five labs, five minds: building a multi-model financial narrative game with small models
Original title:Five labs, five minds: building a multi-model finance drama on small models
AI summary
Thousand Token Wood v2 uses small models from four different labs (gpt-oss-20b, MiniCPM3-4B, Nemotron-Mini-4B, and a fine-tuned Qwen 0.5B) to drive agents in a financial simulation game. The core finding is that the friction in the heterogeneous serving layer lies in vLLM 0.22.1 requiring the CUDA toolkit, rather than in the models themselves. Through a tolerant JSON parsing layer, adding a model requires only one configuration. Information isolation ensures insider flags are not in prompts, and scanning tests verify no leakage. Memory uses emotional summary truncation to avoid being overwhelmed. The fine-tuned 0.5B model achieves 0% self-trading and 100% valid quotes, with zero leakage through the truth firewall. Small models are reliable format generators but unreliable reasoners, which can be compensated for through structure, prompts, and fine-tuning.
Key Takeaways
- Thousand Token Wood v2 uses small models from four different labs…
- 5B) to drive agents in a financial simulation game.
- The core finding is that the friction in the heterogeneous serving layer lies in vLLM 0.
- 1 requiring the CUDA toolkit,rather than in the models themselves.
- Through a tolerant JSON parsing layer,adding a model requires only one configuration.
Quick Facts
- Published
- Updated
- Source
- Hugging Face 社区博客(混合发现)
- Category
- Tutorials & Tips
- AI score
- 66
不是那种「我用 GPT 写了个游戏」的浅显分享,真在四个小模型上跑出了博弈感,里面 vLLM 踩坑和防火墙测试方法可以直接抄作业。
Related entities
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Five labs, five minds: building a multi-model financial narrative game with small models. June 7, 2026. https://new.nat6.net/en/news/cmq2rp55c00hlsl97h98gkd14/