--- title: Qwenputer-npu canonical_url: "https://www.modelscope.ai/models/Spartan1925/Qwenputer-npu" md_url: "https://www.modelscope.ai/models/Spartan1925/Qwenputer-npu.md" repository: Spartan1925/Qwenputer-npu last_updated: 2026-10-30 license: apache-2.0 pipeline_tag: text-generation tasks: - text-generation model_type: - qwen3_5_text architectures: - Qwen3_5ForCausalLM base_model: - Qwen/Qwen3.5-4B base_model_relation: merge parameters: 9.8B tensor_type: - BF16 - F32 library_name: - safetensors - pytorch frameworks: - pytorch downloads: 6 stars: 0 tags: - qwen3.5 - risc-v - rv32im - compiled-weights - gated-deltanet - qwenputer --- # Qwenputer-npu > Qwenputer-npu - An open-source model by Spartan1925 on ModelScope. Code on GitHub · Qwenputer · Base model: Qwen/Qwen3.5-4B · Apache-2.0 Spartan1925/Qwenputer-npu is a 9.8B-parameter text-generation model on ModelScope. licensed under apache-2.0. derived from Qwen/Qwen3.5-4B. - **Repository**: Spartan1925/Qwenputer-npu - **License**: apache-2.0 - **Tasks**: text-generation - **Parameters**: 9.8B - **Base model**: Qwen/Qwen3.5-4B - **Tags**: qwen3.5, risc-v, rv32im, compiled-weights, gated-deltanet, qwenputer - **Downloads**: 6 - **Stars**: 0 - **Last updated**: 2026-10-30 Source: https://www.modelscope.ai/models/Spartan1925/Qwenputer-npu ---

Qwenputer: a computer, compiled into the weights of a language model.

[Code on GitHub](https://github.com/divagr18/qwenputer/tree/npu) · [Qwenputer](https://www.modelscope.ai/models/Spartan1925/Qwen3.5-4B-Qwenputer) · [Base model: Qwen/Qwen3.5-4B](https://huggingface.co/Qwen/Qwen3.5-4B) · Apache-2.0 **Qwenputer-npu is Qwen3.5-4B with a 32-bit RISC-V CPU compiled into its weights, and that CPU now has a neural-network instruction set. It is fast enough to run a second language model: Karpathy's stories260K runs as a C program on the CPU inside Qwen and writes its stories.** Like [Qwenputer](https://www.modelscope.ai/models/Spartan1925/Qwen3.5-4B-Qwenputer), the CPU is 11 transformer layers in front of Qwen's own, every weight set by a compiler, with Qwen3.5's Gated DeltaNet state as its RAM. This version adds two instructions and a weight memory: - `VDOT rd, rs1, rs2`: the dot product of 8 int8 activations (two registers) with the next 8 int4 weights of the weight memory, added to `rd`. `WB rs1` sets where the weights start. - `SMAQA rd, rs1, rs2`: four signed byte products added to `rd`, as in the RISC-V P extension. - A weight memory of 32,640 words (261,120 int4 weights) in DeltaNet heads, loaded from the program's `.wrom` section. Each is one instruction for the CPU and one trace token, computed exactly by the compiled layers. ## A language model inside the language model stories260K is a 260K-parameter Llama trained on TinyStories. Here it is a C program (`guests/stories260k/`): its weights are data in the CPU's memory, and every multiply is an instruction the CPU executes. None of it is compiled into the checkpoint, so any other model that fits the memory runs the same way. | Build of stories260K | Trace tokens per story token | |---|---| | Plain C, int8 weights | 2,430,000 | | Unrolled kernels, no new instructions | 1,110,000 | | **VDOT + SMAQA, int4 weights** | **225,000** | The int4 weights come from distilling the float model on its own samples. The CPU writes the same story as the float model for all 60 tokens checked: > Once upon a time, there was a little girl named Lily. She loved to play outside in the park. > One day, she saw a big, red ball. She wanted to play with it, but it was too high A new fused executor runs the CPU at about 5,500 trace tokens a second on an L4, so a story token takes about 41 seconds. ## Checks - The checkpoint's own chip layers run VDOT and SMAQA test programs and the first 30,000 steps of the story program, teacher-forced against a reference model of the CPU: zero mismatches. - The executor runs the story free, checked against the reference: exact. - An IR replay of 39 of the `mul_big` conversations (114,736 positions, 93,381 of them the computer's) against the reference: zero mismatches. - `calc` still works the same way: every answer the CPU gave in these runs matched the reference model. The published sets, 40 problems each, batched decoding: | Benchmark | Base Qwen3.5-4B | **Qwenputer-npu** | |---|---|---| | Synthetic integer arithmetic | 75.0 | **90.0** | | DeepMind `mul_big` | 55.0 | **97.5** | The four arithmetic misses are Qwen repeating `calc` calls until it runs out of budget, with the CPU answering each one correctly; the published Qwenputer scores 97.5 on that set. Qwen's text behaviour matches base: KL 5.8e-4, against a noise floor of 5.1e-4. ## Quickstart ```bash git clone -b npu https://github.com/divagr18/qwenputer && cd qwenputer/experimental/riscv && uv sync --extra hf modelscope download Spartan1925/Qwenputer-npu --local_dir qwenputer-npu uv run python -m chipformer.rv.chat qwenputer-npu --firmware tests/data/rv/calc-bridge.elf \ --ask "Use calc: what is 123456789 * 987654321?" uv run python -m chipformer.rv.npucheck qwenputer-npu guests/stories260k/stories260k.elf uv run python guests/stories260k/storyrun.py guests/stories260k/stories260k.elf ``` `npucheck` runs the new instructions on this checkpoint's layers; `storyrun` writes the story on the CPU's executor. ## Gated DeltaNet as RAM

Gated DeltaNet state used as RAM: a one-hot key overwrites one row, a one-hot query reads it back, and the CPU's registers, memory and program counter live in DeltaNet heads.

## Model details | | | |---|---| | Base | Qwen/Qwen3.5-4B, Qwen's own weights unchanged | | Architecture | stock `Qwen3_5ForCausalLM` (text only); config-only changes | | Computer | 11 compiled layers before Qwen's 32; RV32IM + Zba + bignum unit + VDOT, WB, SMAQA | | Weight memory | 32,640 words of 8 int4 weights, in DeltaNet heads | | Hidden size | 4096 (Qwen's 2560 plus the computer's dims) | | Vocabulary | Qwen's, plus 263,092 trace tokens | | Size | 19.6 GB in bf16, mostly zero padding | | Precision | exact in bf16 | ## Citation ```bibtex @software{agrawal2026qwenputer, author = {Agrawal, Divyansh}, title = {{Qwenputer}: a computer compiled into the weights of a language model}, year = {2026}, version = {0.3.0}, url = {https://github.com/divagr18/qwenputer}, license = {MIT} } ``` ## License Apache-2.0. This is a modified version of [Qwen/Qwen3.5-4B](https://huggingface.co/Qwen/Qwen3.5-4B): the config is widened and computer layers are added, but Qwen's own weights are unchanged. stories260K is by Andrej Karpathy (MIT).