phantom-kv

Refusal removal as a loadable KV-cache graft — zero weight modification, no refusal-direction projection, fully reversible.

lordx64/phantom-kv

View on GitHub ↗

Refusal removal as a loadable KV-cache graft — zero weight modification, no refusal-direction projection, fully reversible.

phantom-kv grew +51.1% in the last 30 days, ranks 42nd of 80 in Interpretability & Alignment, last commit 61 days ago.

Star trajectory

59 days tracked
2026-09-23275 stars today2026-11-22
Momentum

+51.1% in 30d, above the Interpretability & Alignment median of +0.4%.

Maintenance

Last commit 61 days ago, fresher than the median of 183d.

How it compares in Interpretability & Alignment

phantom-kv
275
stars
+51.1% 30d
commit 61 days ago
Interpretability & Alignment median
290
stars
+0.4% 30d
commit ~183d ago
Leader · lmms-eval
4.5k
stars
+0.9% 30d
commit today

Ranked 42nd of 80 by stars.

Ranked by shared tags, then keywords, then stars
  1. I replicated Ng's RYS method and found that duplicating 3 specific layers in Qwen2.5-32B boosts reasoning by…

    +2 (+0.8%) in 30d · Committed 247d ago · Python

    Shares Interpretability & Alignment, Model Optimization

  2. Code for ACL 2024 paper "TruthX: Alleviating Hallucinations by Editing Large Language Models in Truthful…

    +0 (+0.0%) in 30d · Committed 970d ago · Python · GPL-3.0

    Shares Interpretability & Alignment, Model Optimization

  3. Find, rank and steer MLP neurons of GGUF models on llama.cpp; export control vectors; 3D web UI and MCP server

    Committed yesterday · Python · MIT

    Shares Interpretability & Alignment, Model Optimization

  4. End-to-end training of sparse deep neural networks with little-to-no performance loss.

    +2 (+0.6%) in 30d · Committed 1,395d ago · Python · Apache-2.0

    Shares Model Optimization, Interpretability & Alignment

  5. Perform vector ablation to local LLM models

    +7 (+5.7%) in 30d · Committed 177d ago · Python · MIT

    Shares Interpretability & Alignment, Model Optimization

  6. Interactive canvas for taking an LLM apart tensor by tensor: measured INT8/INT4/FP8 error, distributions…

    +1 (+1.3%) in 30d · Committed 80d ago · TypeScript · MIT

    Shares Interpretability & Alignment, Model Optimization

  7. Official implementation of Half-Quadratic Quantization (HQQ)

    +3 (+0.3%) in 30d · Committed 268d ago · Python · Apache-2.0

    Shares Model Optimization

  8. MindSpore + 🤗Huggingface: Run any Transformers/Diffusers model on MindSpore with seamless compatibility and…

    +1 (+0.1%) in 30d · Committed 82d ago · Python · Apache-2.0

    Shares Model Optimization

  9. Inspect, visualize and intervene on any operation in any neural network. Capture every activation and…

    Committed 5d ago · Python · Apache-2.0

    Shares Interpretability & Alignment

  10. The purpose of this repository is to make prototypes as case study in the context of proof of concept(PoC) and…

    +0 (+0.0%) in 30d · Committed 1,062d ago · Python · GPL-2.0

    Shares Interpretability & Alignment

Language
Python
License
MIT License
Contributors
1
Forks
43
Watchers
2
Open issues
1
Releases
1
Created
2026-09-17
Stars this week
+2

You've seen where phantom-kv stands today.

Get the moves before everyone else: the 5 Interpretability & Alignment repos climbing fastest, every Tuesday.

Free weekly digest. Double opt-in, unsubscribe anytime.

Maintain phantom-kv?

Claim this page to add your own call-to-action.

Claim this page