AI Model Training

ForgeTrain ForgeTrain

The world's first production-level large model pre-training framework written entirely by AI Agent with zero human code intervention

FS-DFM Apple

The few-step discrete flow matching text generation model launched by Apple can achieve thousand-step quality with 8 steps of sampling.

ACE-Ego ACE-Ego

One brain, multiple types of embodied operation VLA model research project

Gemini Diffusion Google DeepMind

Experimental text diffusion model launched by Google DeepMind, non-autoregressive generation, iterative text refinement

Agent Lightning Microsoft

Microsoft's agent training framework uses reinforcement learning and other methods to allow Agents to continuously optimize from experience.

Gemma 4 12B Google

Google's open source coder-less unified architecture multi-modal large model, the local running performance of 12B parameters exceeds the previous generation 27B