Sparticle62ops/pssa
A custom AI architecture being developed in rust
About the project
PSSA is a research prototype language model written in Rust with no ML frameworks: a recurrent state-space core, an episodic memory bank, and plastic weights instead of transformer attention. The author reports it learns faster and generates text about 12x faster than a transformer on CPU at 1.5M parameters.
Useful for
- Build and run the prototype with cargo build --release to train on your own corpus
- Compare PSSA and transformer learning curves on the same corpus and tokenizer
- Validate CPU gradients against the CUDA path via cargo test --release
README summarized by DeepSeek V4.1 Flash. Details may be inaccurate.
Why it’s trending
- 0 stars so far today, about 13 expected by the end of the day. The usual pace is 0 per day, so that's 13× as much.
- Before this, the repository barely got any stars — about 0 per day.
- The spike has held for 2 days in a row — not a one-off blip.
- The repository is 42 days old and already has 20 stars.
- Hacker News: “PSSA: A non-transformer language model written from scratch in Rust” — 37 points, 2 h ago.
Stars per day
Bars are daily stars, the line is the usual pace. Red marks spike days.
Numbers
- Total stars
- 20
- Today
- 0 · ≈ 13 by evening
- Forks
- 1
- Issues and pull requests
- 2
- Watchers
- 0
- Language
- Rust
- License
- GPL-3.0
- Created
- August 19, 2026
- Last push
- September 30, 2026
Star trust
Growth looks organic: forks and discussion are in line with active projects, and stars arrive unevenly, the way people give them.
These are heuristics, not a verdict: we judge by the repository’s behavior, not by a list of stargazers.
Hacker News discussions
- PSSA: A non-transformer language model written from scratch in Rust37 points, September 30, 2026
Spotted in
- September 30, 2026Spotted on Hacker News
Share
README badge
Paste it into your README — the badge shows the current magnitude and links to this page.
[](https://gitnova.dev/en/r/Sparticle62ops/pssa)More in this category
-
7.7
firelex/jeff
Small fine-tuned Qwen3.5 and Gemma 4 models for zero-shot classification: given a situation description and a list of options, they return a calibrated probability for each option in a single forward pass.
-
7.7
VectifyAI/PageIndex
A vectorless RAG engine that builds a hierarchical tree index of a document and retrieves by LLM reasoning over that tree, the way a human navigates a report. Aimed at long professional PDFs such as financial, legal…
-
7.3
PostHog/jeeves
Jeeves is a reasoning model built on Qwen3.5-9B (LoRA + pointer head), trained with SFT and CISPO, that thinks before answering and returns calibrated probabilities for yes/no, multiple-choice and rating questions via…
-
7.3
Niko1221/Strata
A C++ local inference engine that runs the 125B MoE model Qwen3.8-Flash-Next on a regular PC with one NVIDIA GPU (12-24 GB) and 64 GB RAM, exposing an OpenAI/Anthropic-compatible API on localhost.
-
5.9
ninjahawk/livenerf
A benchmark for tracking whether a frontier model quietly degrades after release: it runs a frozen question panel daily and statistically measures drift in accuracy and token counts against the launch-week baseline.
-
5.7
deepseek-ai/DeepGEMM-Ascend
A port of DeepGEMM to Huawei Ascend: GEMM kernels (BF16, FP8, FP4, MQA logits, MegaMoE) with an API compatible with DeepGEMM, targeting peak NPU performance.