cohere-ai/magikarp
Code for the paper "Fishing for Magikarp"
This project helps large language model (LLM) developers identify when specific tokens, or pieces of words, within their models haven't been adequately trained. By inputting an LLM, it generates detailed reports and visualizations that highlight these 'under-trained' tokens. This tool is designed for AI researchers and engineers who build and fine-tune large language models.
182 stars.
Use this if you are developing or working with a large language model and need to diagnose issues related to token training, ensuring your model's vocabulary is robustly represented.
Not ideal if you are an end-user of an LLM or focusing on application-level performance rather than deep model-level token diagnostics.
Stars
182
Forks
16
Language
Python
License
Apache-2.0
Category
Last pushed
Mar 12, 2026
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/nlp/cohere-ai/magikarp"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.