Warp
Run trillion-parameter models on a laptop by streaming weights from NVMe
Emerged finds new products worth a look, daily — what’s new today
Laplace's Scissors
Edits an open model's readout layer to fix wrong answers, keeping right ones
For teams running open-weight models, it offers a surgical alternative to fine-tuning: repair specific failures in the readout layer and export a normal checkpoint.
Laplace's Scissors works on the final readout matrix of a dense decoder model. It probes for weak output directions, diagnoses answers the model gets wrong, and applies a rotation that repairs them while holding a protected set of answers fixed, then exports the edited checkpoint. Written in Rust with Triton kernels for CUDA and HIP; needs a GPU and Hugging Face access. MIT. The README publishes no benchmarks yet, so effectiveness is yours to measure.
More like this
SentenceCoach
Fixes a learner's Mandarin with the smallest change and says why
STE80 Humanizer
An agent skill that edits LLM prose toward Simplified Technical English
Flide
Exact, shared-context inference server for small Jev-style decision models
Laya
Small encoder models that answer routing and triage questions in one pass
SemIf
Read a yes/no probability straight out of a small open model, no generated text