Qwen3.5-9B on a PlayStation 4: Garlic LLM Needs No PC
Hobbyist Leonardo Alegria got a 9-billion-parameter model running on a PlayStation 4. What Garlic LLM can do and how fast it answers.
Hobbyist Leonardo Alegria has run the Qwen3.5-9B model, with 9 billion parameters, on a PlayStation 4. The project is called Garlic LLM and runs fully standalone: no Linux on the console and no external computer.
Performance reaches 1 to 39 tokens per second. To fit the model into memory, the author used quantization and shrank it to 4.5 GB. The console has only 8 GB of unified memory.
The result is that the PlayStation 4 works as an offline AI chatbot with voice control support.
The PlayStation 4 came out in 2013. Qwen is a family of language models from the Chinese company Alibaba. Quantization, in simple terms, stores the numbers inside a model at lower precision: the file takes less storage and memory, at the cost of some quality.
Source: iXBT