Tencent-Hunyuan/GradLoc

Implementation of GradLoc from the Tencent Hunyuan blog "Stabilizing RLVR via Token-level Gradient Diagnosis and Layerwise Clipping".

/ 100

Emerging

This tool helps machine learning engineers and researchers diagnose why their Reinforcement Learning from Vision and Language (RLVR) models sometimes fail during training. It takes the RLVR model's training process as input and outputs a precise identification of which specific data "tokens" are causing instability by creating problematic gradient spikes. This allows practitioners to understand and fix the root causes of training collapse.

Use this if you are a machine learning engineer or researcher experiencing unexplained training failures or instability in your RLVR models and need a way to pinpoint the exact data inputs responsible for these issues.

Not ideal if you are working with machine learning models outside of the RLVR domain or if you are not comfortable applying code patches to an existing framework.

Reinforcement Learning Vision and Language Models Machine Learning Debugging Model Training Stability Gradient Analysis

No Package No Dependents

Maintenance 10 / 25

Adoption 9 / 25

Maturity 11 / 25

Community 15 / 25

How are scores calculated?

Stars

Forks

Language

Python

License

—

Higher-rated alternatives

jncraton/languagemodels

Explore large language models in 512MB of RAM

microsoft/unilm

Large-scale Self-supervised Pre-training Across Tasks, Languages, and Modalities

haizelabs/verdict

Inference-time scaling for LLMs-as-a-judge.

albertan017/LLM4Decompile

Reverse Engineering: Decompiling Binary Code with Large Language Models

bytedance/Sa2VA

Official Repo For Pixel-LLM Codebase

Explore Transformer Models

All categories Trending Transformer directory Insights