AI model running printer
The story so far
Omar Sanseverino, a member of the Google DeepMind technology team, said that he ran a 270M Gemma 3 model with 14.1MB of memory, by compiling the Transformer inference engine and the quantized model into PostScript print jobs and using the printer's built-in interpreter to execute computations, with a processing time of about 60 seconds per token.
AI-written from the reports · Updated 26 minutes ago
Coverage timeline
Follow the coverage to see every side of the story.
- ITHome · AIGoogle engineers fit the 270M parameter Gemma 3 model with 270 million parameters into a printer to run
Google DeepMind engineer Omar Sanseverino ran a 270M parameter Gemma 3 270M model with 14.1MB of memory, taking about 60 seconds per token, by compiling the Transformer inference engine and the quantized model into a PostScript print job and executing the computation using the printer's built-in interpreter.
Heat over time
Not enough data yet to draw a trend.