SunnyNet
CPU-first O(N) linear-recurrence inference runtime with hand-written SIMD kernels.
Tested on Linux
Use it
Linux x86_64
install via uv tool candidate 1
uv tool install sunnynet-cx
Where the README and reality differ
- README: sunnynet-serve --ckpt <checkpoint.pt> --host 127.0.0.1 --port 8080
What worked: pass --tokenizer /tmp/mini/tokenizer.json explicitly; default data/proc/tokenizer.json does not exist
Check that it works
sunnynet --help 2>&1The output should contain “--ckpt”.
What it does
- Run linear-recurrence models on the CPUNot tested
- Load SunnyNet, RWKV-7, Mamba-2 and other modelsNot tested
- Serve models over an OpenAI-compatible HTTP APINot tested
How far pickfit got
Linux x86_64
- Installs
- Starts
- 0 of 3 features work
- Works offline
Windows: the project says it runs there; pickfit hasn’t tried it.
Good for
- generate text from a checkpoint with a prompt
- serve a checkpoint via OpenAI-compatible HTTP service
- run external RWKV-7 GGUF weights for generation
An alternative to rwkv, mamba, transformer and hyena.
How pickfit tested it
Each run starts from a clean machine and follows the steps above. How pickfit verifies
Linux x86_64
Its main function works
- Replayed from scratch
- Works offline
- Checked on 30 Oct 2026
- Version 0.3.0
Differences the test machine hit
- README: hf download SunnyClaw-cx/SunnyNet --local-dir sunnynet-small-zh (~21 MB)
In the test: create tiny local ckpt/tokenizer with installed sunnynet package (/tmp/mini) and use it for CLI and serve checks