miHoYo Announces General Visual Language Action Base Model Gamer
MiHoYo announces the universal visual language action foundation model Gamer, which does not require reading internal game data or relying on APIs. It collects game screen inputs at a rate of 5 frames per second (5 Hz) to generate native keyboard and mouse command simulations equivalent to 30 Hz, simulating real human operations. With NVIDIA H200 hardware and the Streaming-vLLM engine, end-to-end request latency is reduced to 65.12 milliseconds.
Same story1 other outlet also reported this
- Hugging Face Daily Papers··First
miHoYo Honkai AI R&D team introduces the general visual language action foundation model Gamer for long-term real-time game intelligent agents
- ITHome · AI··+35 h 26 min·Reading now
miHoYo Announces General Visual Language Action Base Model Gamer
Adds:NVIDIA H200 hardware reduces end-to-end request latency to 65.12 milliseconds.
In order of publication · later reports are added as they appear · Compared by AI from each outlet’s title and summary
Source: ITHome · AI · ithome.com