MaxBelitsky/cache-steering
KV Cache Steering for Inducing Reasoning in Small Language Models
This project helps you make small language models behave like larger, more sophisticated ones, particularly by enabling them to show their 'thinking process' (chain-of-thought reasoning) or adopt specific writing styles. It works by taking an existing language model and injecting 'steering vectors' derived from examples of desired behavior. The output is a modified version of the language model's response, now exhibiting reasoning or a particular style. This tool is for AI practitioners, researchers, and developers who want to enhance the capabilities of smaller language models without extensive retraining.
No commits in the last 6 months.
Use this if you want to modify the behavior or output style of an existing language model (e.g., to induce reasoning or a specific tone) without needing to retrain the model or alter its architecture.
Not ideal if you are looking to train a new language model from scratch or if your primary goal is to improve the factual accuracy or base knowledge of a model.
Stars
46
Forks
5
Language
Python
License
—
Category
Last pushed
Jul 24, 2025
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/llm-tools/MaxBelitsky/cache-steering"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
zjunlp/KnowledgeEditingPapers
Must-read Papers on Knowledge Editing for Large Language Models.
zjunlp/CaKE
[EMNLP 2025] Circuit-Aware Editing Enables Generalizable Knowledge Learners
zjunlp/unlearn
[ACL 2025] Knowledge Unlearning for Large Language Models
OFA-Sys/Ditto
A self-ailgnment method for role-play. Benchmark for role-play. Resources for "Large Language...
zjunlp/AutoSteer
[EMNLP 2025] AutoSteer: Automating Steering for Safe Multimodal Large Language Models