Light-R1-14B-DS

An open-source 14B-parameter mathematical model, trained using reinforcement learning, with excellent performance.

CommonProductProductivityReinforcement LearningMathematical Model

Light-R1-14B-DS is an open-source mathematical model developed by Qihoo 360 Technology Co., Ltd. Trained using reinforcement learning based on DeepSeek-R1-Distill-Qwen-14B, it achieved high scores of 74.0 and 60.2 on the AIME24 and AIME25 mathematics competition benchmarks, respectively, surpassing many 32B parameter models. It successfully implemented reinforcement learning on an already long-chain reasoning fine-tuned model under a lightweight budget, providing the open-source community with a powerful mathematical model tool. Its open-source nature promotes the application of natural language processing in education, particularly in mathematical problem-solving, offering researchers and developers valuable research foundations and practical tools.

Reinforcement learning-based long-chain reasoning training enhances mathematical problem-solving capabilities.
Open-source model facilitates secondary development and research by researchers and developers.
Excellent performance in mathematical benchmark tests such as AIME24 and AIME25
with high accuracy.
Supports efficient training under a lightweight budget
reducing computational costs.
Provides detailed training logs and technical reports for easy understanding and reproducibility.

This model is suitable for researchers and developers in natural language processing
especially those focusing on mathematical problem-solving
educational applications
and reinforcement learning. It provides an excellent reference for teams aiming for high-performance model training on a lightweight budget
enabling quick adoption and research and development.

Researchers can utilize this model to research and improve mathematical problem-solving algorithms.
Developers can build educational applications based on this model to help students better solve mathematical problems.
Businesses can apply this model to intelligent customer service systems to improve the ability to answer math-related questions.

1. Visit the Hugging Face website and locate the Light-R1-14B-DS model page.
2. Download the model files and related resources
including training logs and technical reports.
3. Load the model using a supported framework
such as PyTorch or TensorFlow.

Visit

Light-R1-14B-DS Visit Over Time

Monthly Visits

29742941

Bounce Rate

44.20%

Page per Visit

5.9

Visit Duration

00:04:44

Light-R1-14B-DS Visit Trend

Light-R1-14B-DS Visit Geography

Light-R1-14B-DS Traffic Sources

Light-R1-14B-DS Alternatives

Light-R1-14B-DS — An open-source 14B-parameter mathematical model, trained using reinforcement learning, with excellent performance.

Best AI Websites & Tools

Light-R1-14B-DS

Light-R1-14B-DS Visit Over Time

Light-R1-14B-DS Visit Trend

Light-R1-14B-DS Visit Geography

Light-R1-14B-DS Traffic Sources

Light-R1-14B-DS Alternatives

Light-R1-14B-DS — An open-source 14B-parameter mathematical model, trained using reinforcement learning, with excellent performance.

Light-R1 — Light-R1 is an open-source project focusing on long-chain reasoning (Long COT), providing a training method from scratch through curriculum-style SFT, DPO, and RL.

Steiner-32b-preview — Steiner is a reasoning model trained on synthetic data, designed to explore multiple reasoning paths and verify them autonomously.

SWE-RL — Enhancing the reasoning capabilities of large language models in open-source software evolution through reinforcement learning.

R1-V — Enhances the generalization capabilities of visual language models at a low cost of less than $3.

Tülu 3 405B — Tülu 3 405B is a large-scale open-source language model enhanced through reinforcement learning.

DeepSeek-R1-Distill-Qwen-1.5B — DeepSeek-R1-Distill-Qwen-1.5B is an efficient inference open-source language model suitable for various natural language processing tasks.

DeepSeek-R1-Distill-Qwen-7B — DeepSeek-R1-Distill-Qwen-7B is an open-source reasoning model focusing on mathematics, coding, and reasoning tasks.

DeepSeek-R1-Distill-Llama-8B — DeepSeek-R1-Distill-Llama-8B is a high-performance open-source language model suitable for text generation and inference tasks.

DeepSeek-R1-Distill-Qwen-32B — DeepSeek-R1-Distill-Qwen-32B is a high-performance open-source language model suitable for various text generation tasks.

DeepSeek-R1-Distill-Llama-70B — DeepSeek-R1-Distill-Llama-70B is a large language model optimized using reinforcement learning, focusing on reasoning and conversational capabilities.

DeepSeek-R1 — DeepSeek-R1 is a high-performance inference model supporting various languages and tasks, suitable for both research and commercial applications.

self-adaptive-llms — A real-time adaptive framework for unseen tasks using large language models.

PRIME-RL — PRIME enhances the reasoning abilities of language models through implicit reward-driven online reinforcement learning.

HuatuoGPT-o1 — A large language model for complex reasoning in the medical field

Unitree RL GYM — Unitree robot platform for reinforcement learning

agibot_x1_train — Modular humanoid robot for reinforcement learning training

Qwen2.5-Math — World-leading open-source large language model for mathematics

MuKoe — An open-source implementation of MuZero, a distributed AI framework

Same — A tool that can replicate any webpage UI interface and generate code suggestions.

Sesame CSM — A model for generating conversational speech, supporting high-quality speech generation from text and audio input.

RagaAI — RagaAI Catalyst is a platform for observing, evaluating, and debugging AI agents, helping developers optimize AI workflows and deploy securely.

open-mcp-client — An open-source multi-cloud platform client that supports LangGraph agent and frontend application development.

IMM — Inductive Moment Matching is a novel generative model for high-quality image generation.

BashBuddy — BashBuddy lets you enter commands naturally without worrying about parameters or syntax.

R1-Omni — R1-Omni is a full-modality emotion recognition model incorporating reinforcement learning, focusing on improving the interpretability of multimodal emotion recognition.

nanobrowser — An open-source Chrome extension for AI-powered web automation, supporting multi-agent workflows.

l1m — A proxy API for extracting structured data from text and images, implemented based on LLMs.

NotaGen — NotaGen is a model for symbolic music generation, employing a large language model training paradigm and focusing on generating high-quality classical music scores.

Proxy Lite — Proxy Lite is an open-source 3B parameter visual language model (VLM) focused on web automation tasks.