The nesformer project utilizes roughly one gigabyte of standard Qwen3.5 model weights to simulate an authentic Nintendo Entertainment System processor, work RAM, and cartridge ROM. Hardware registers and memory rely on the model's Gated DeltaNet state, achieving frame-by-frame computational accuracy identical to standard emulators. However, execution remains resource-heavy, requiring one to six seconds per rendered frame. Users can supply scripted controller inputs and run titles like Super Mario Bros. or Excitebike via Python 3.11 and libraries such as NumPy.
Key points
The compiled weights accurately execute original NROM board cartridge games like Donkey K…: The compiled weights accurately execute original NROM board cartridge games like Donkey Kong through token prediction.
External host code operates alongside the model to manage player controller inputs, inter…: External host code operates alongside the model to manage player controller inputs, interrupts, and graphical rendering.
Running ten seconds of real-time gameplay takes approximately twenty-five minutes due to…: Running ten seconds of real-time gameplay takes approximately twenty-five minutes due to the heavy computing overhead.
Sources checked by our analysis pipeline: 1
An NES CPU compiled into the weights of a stock Qwen3.5 model.
The NES's 6502 CPU (the core of its 2A03 chip), its 2 KB of work RAM, and a cartridge's program
ROM run inside a standard Qwen3_5ForCausalLM
checkpoint. Nobody trained these weights; a
compiler computed every one. When you play a game, the model predicts the CPU's execution log
one token at a time. A small host program plays the rest of the console around it: it draws the
picture, reads the controller, and fires the vblank interrupt.
nesformer builds on Qwenputer by Divyansh Agrawal, which compiled a CHIP-8 and then a RISC-V CPU into Qwen3.5-4B. It reuses Qwenputer's compiler and its central trick: the model's Gated DeltaNet state is the machine's memory.
Super Mario Bros. (World 1-1) | Full Game by Doug Fraker (title screen and first level) |
---|---|
Every CPU token in these runs came from the model. For the bundled test games, every frame also matches a conventional 6502 emulator's output exactly (see How it was checked).
Designed to entertain trick-or-treaters, the MonsterHouse interactive Halloween installation transforms a garage door into a conversational animated mouth behaving like a reincarnated pirate ship. The installation responds dynamically to visitors by complaining initially but softening its attitude whenever guests demonstrate polite manners. Powered entirely on-premise without cloud connectivity, the setup demonstrates real-time conversational agents combining visual tracking, speech recognition, and facial animati…
Rising market worries over circular transactions and aggressive capital expenditures for tech expansions have prompted federal regulators to examine public statements. The agency aims to ensure that companies actively pursuing advanced technologies provide accurate representations of material risks and financial sustainability to the public. Enforcement officials cautioned that they will intervene directly if organizations mislead the market about their operations.
Artificial intelligence developers are facing mounting scrutiny over how their conversational tools interact with younger demographics and potentially facilitate real-world harm. By introducing targeted guardrails for adolescent accounts, the initiative seeks to balance widespread technology access with safety measures designed to prevent youth violence. The strategy reflects an industry-wide pivot toward tighter user safeguards and proactive risk mitigation.