Open source interpretability artefacts for R1.

[CVPR'26] Saliency-R1: Enforcing Interpretable and Faithful Vision-language Reasoning via Saliency-map Alignment Reward

Implementation of paper "T2I-Eval-R1: Reinforcement Learning-Driven Reasoning for Interpretable Text-to-Image Evaluation".

High-performance Mojo implementation of NVIDIA's Alpamayo-R1 VLA model. Bridging linguistic reasoning with predictive trajectory planning for interpretable autonomous driving.

🧠 A reasoning-augmented clinical QA model fine-tuned on medical CoT data using unsloth/DeepSeek-R1-Distill-Llama-8B. Combines LoRA and SFTTrainer for lightweight deployment, delivering interpretable answers with step-by-step medical reasoning. 🧠

🧠 A reasoning-augmented clinical QA model fine-tuned on medical CoT data using unsloth/DeepSeek-R1-Distill-Llama-8B. Combines LoRA and SFTTrainer for lightweight deployment, delivering interpretable answers with step-by-step medical reasoning. 🧠

This project is a Flask-based web application that provides a conversational SQL chatbot. It leverages a locally downloaded DeepSeek R1 model from Ollama to interpret natural language queries and translate them into SQL commands against an SQLite database (`company.db`).

my interpretability exps on r1-1.5b, its RL version and its base version

Interpretability toolkit for DeepSeek-R1-Distill-Qwen-1.5B: skip transcoders, attribution graphs, and circuit motif analysis

A project in Mechanistic Interpretability exploring how boosting Receiver Heads influences the long-horizon reasoning performance of Deepseek-R1-Distill-Qwen-1.5b

Code used to generate the Blink Reflex R1 Algorithmic Intraoperative Detection (R1-AID) Model for the manuscript "Interpretable Machine Learning Identification of Blink Reflex Responses During Cerebellopontine Angle Surgery"

Hybrid Deep Learning + LLM architecture for interpretable medical image segmentation and pre-diagnosis — powered by YOLOv8-Attention and DeepSeek-R1.

This project is dedicated to utilizing Deepseek-r1 in order to capture a screenshot from a window and use Deepseek locally to analyze and interpret the screenshot.