Home
Mianbi Intelligence’s ForgeTrain Uses AI to Forge Next-Gen Training Framework: 8 Hours to Catch Up, 2 Days to Surpass

As large model iteration cycles shrink to weeks, overcoming computing power constraints and engineering barriers has become the defining challenge in AI. On June 27, 2026, Mingshi Intelligence, together with the OpenBMB open-source community and AGI BAR, hosted an in-depth event called "AI4AI Fermentation Night." At the event, Li Yuxuan, technical lead of Mingshi Intelligence's AI Infra, detailed the company's proprietary production-grade pre-training framework, ForgeTrain, revealing the logic and practical breakthroughs behind the paradigm shift of "AI manufacturing AI."
Li Yuxuan noted that as the marginal returns from high-quality internet data and computing power diminish, the traditional approach of "piling on data and compute" has reached its limits. The industrial revolution saw "machines building machines"; the intelligence revolution is now moving toward "AI building AI." ForgeTrain validates this path with a single point: using AI to forge a specialized training framework tailored to specific models and hardware architectures from scratch, rather than relying on manually maintained general-purpose software stacks.
In performance tests, ForgeTrain demonstrated remarkable iteration efficiency. Through an automated process, it matched the performance of the industry-leading framework Megatron-LM within 8 hours and consistently surpassed it within 1.5 to 2 days, achieving a compute utilization (MFU) increase of about 8% to 10%. This capability has been successfully applied to models like MiniCPM4-0.5B/8B and is compatible with various hardware platforms, including H100 and Huawei Ascend NPU.
The success of ForgeTrain hinges on what Li Yuxuan calls the "Four-Stage Harness Optimization Process." Starting from the Anchor stage, which locks in binary consistency, it proceeds through Bit-for-Bit basic function generation, then Surpass performance sprint after removing constraints, and finally enters the Per-Op stage, which involves deep customization of each operator. The entire process is driven entirely by AI, with no manual intervention, successfully transforming NVIDIA's years of engineering moat into a technical problem that AI can automatically decouple.
This approach is summarized as "Forge Engineering" — a new engineering paradigm for the AI era. Li Yuxuan believes that in the future, everyone will be able to customize their own model assistant, and the software landscape will undergo a massive transformation.
Related article
U.S. Stocks Hit Historic Milestone as AI and Aerospace Giants Prepare for Trillion-Dollar Debut
Elon Musk, Sam Altman, and Dario Amodei, three titans of the technology sector, are advancing toward initial public offerings for their respective ventures. With SpaceX, OpenAI, and Anthropic—three industry behemoths nearing trillion-dollar valuation
Swedish AI Startup Lovable Eyes $13.2 Billion Valuation After Major Funding Round
As AI-driven coding tools gain traction, Swedish startup Lovable has secured a major funding round. The company aims to raise $3 billion, potentially boosting its valuation to $13.2 billion—double the $6.6 billion recorded last December. Menlo Ventur
Google Tests Remy AI Agent for Gemini as Focus Shifts to User Control
According to Business Insider, Google is testing Remy, a new AI personal agent for Gemini. This tool aims to execute tasks on behalf of users, streamlining both professional workflows and daily routines.Currently, Remy is undergoing testing in an int
Related Special Topic Recommendations
Comments (0)
0/500

As large model iteration cycles shrink to weeks, overcoming computing power constraints and engineering barriers has become the defining challenge in AI. On June 27, 2026, Mingshi Intelligence, together with the OpenBMB open-source community and AGI BAR, hosted an in-depth event called "AI4AI Fermentation Night." At the event, Li Yuxuan, technical lead of Mingshi Intelligence's AI Infra, detailed the company's proprietary production-grade pre-training framework, ForgeTrain, revealing the logic and practical breakthroughs behind the paradigm shift of "AI manufacturing AI."
Li Yuxuan noted that as the marginal returns from high-quality internet data and computing power diminish, the traditional approach of "piling on data and compute" has reached its limits. The industrial revolution saw "machines building machines"; the intelligence revolution is now moving toward "AI building AI." ForgeTrain validates this path with a single point: using AI to forge a specialized training framework tailored to specific models and hardware architectures from scratch, rather than relying on manually maintained general-purpose software stacks.
In performance tests, ForgeTrain demonstrated remarkable iteration efficiency. Through an automated process, it matched the performance of the industry-leading framework Megatron-LM within 8 hours and consistently surpassed it within 1.5 to 2 days, achieving a compute utilization (MFU) increase of about 8% to 10%. This capability has been successfully applied to models like MiniCPM4-0.5B/8B and is compatible with various hardware platforms, including H100 and Huawei Ascend NPU.
The success of ForgeTrain hinges on what Li Yuxuan calls the "Four-Stage Harness Optimization Process." Starting from the Anchor stage, which locks in binary consistency, it proceeds through Bit-for-Bit basic function generation, then Surpass performance sprint after removing constraints, and finally enters the Per-Op stage, which involves deep customization of each operator. The entire process is driven entirely by AI, with no manual intervention, successfully transforming NVIDIA's years of engineering moat into a technical problem that AI can automatically decouple.
This approach is summarized as "Forge Engineering" — a new engineering paradigm for the AI era. Li Yuxuan believes that in the future, everyone will be able to customize their own model assistant, and the software landscape will undergo a massive transformation.
U.S. Stocks Hit Historic Milestone as AI and Aerospace Giants Prepare for Trillion-Dollar Debut
Elon Musk, Sam Altman, and Dario Amodei, three titans of the technology sector, are advancing toward initial public offerings for their respective ventures. With SpaceX, OpenAI, and Anthropic—three industry behemoths nearing trillion-dollar valuation
Swedish AI Startup Lovable Eyes $13.2 Billion Valuation After Major Funding Round
As AI-driven coding tools gain traction, Swedish startup Lovable has secured a major funding round. The company aims to raise $3 billion, potentially boosting its valuation to $13.2 billion—double the $6.6 billion recorded last December. Menlo Ventur











