You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
An open-source reinforcement learning library by Alibaba, optimized for large-scale language models. Supporting distributed training, multi-task learning, and agent interaction for simpler and more efficient AI model training.
25
-
</div>
26
-
<divclassName={styles.buttons}>
27
-
<ButtonclassName={styles.btn}>{"Get Started >"}</Button>
ROLL (Reinforcement Learning Optimization for Large-scale Learning) is an open-source reinforcement learning framework by Alibaba, designed for large-scale language models. Built on Ray distributed architecture, supporting mainstream algorithms like PPO and GRPO, providing complete solutions from research to production.
28
+
<divclassName={styles.desc}>
29
+
An open-source reinforcement learning library by Alibaba, optimized for large-scale language models. Supporting distributed training, multi-task learning, and agent interaction for simpler and more efficient AI model training.
30
+
</div>
31
+
<divclassName={styles.buttons}>
32
+
<Buttonhref="/ROLL/docs/English/start/"target='_blank'className={styles.btn}>{"Get Started >"}</Button>
children: <divclassName={styles.content}>Ray-based distributed architecture supporting mainstream engines like vLLM, SGLang, Megatron-Core, seamlessly scaling from single machine to large GPU clusters</div>,
children: <divclassName={styles.content}>Support for multi-task joint training including math reasoning, code generation, and dialogue, with dynamic sampling rate and data weight adjustment</div>,
85
-
},
86
-
{
87
-
key: '3',
88
-
label: <divclassName={styles.label}>Extremely Easy to Use</div>,
89
-
children: <divclassName={styles.content}>Gym-style clean API design with modular architecture for flexible extension, one-click switching between different backend engines and algorithm configurations</div>,
ROLL (Reinforcement Learning Optimization for Large-scale Learning) is an open-source reinforcement learning framework by Alibaba, designed for large-scale language models. Built on Ray distributed architecture, supporting mainstream algorithms like PPO and GRPO, providing complete solutions from research to production.
<divclassName={styles.content}>ROLL framework provides comprehensive reinforcement learning support, from model training to agent deployment, every aspect is carefully optimized to make AI training more efficient</div>
<divclassName={styles.content}>Built on a Ray-based distributed architecture, it supports large-scale cluster training at the thousand-GPU level. Its innovative Rollout scheduler and AutoDeviceMapping module dramatically improve GPU resource utilization .</div>
<divclassName={styles.label}>Extreme Training Efficiency</div>
118
-
<divclassName={styles.content}>Integrates cutting-edge technologies like Megatron-Core, SGLang, and vLLM to significantly accelerate the model training and inference sampling processes .</div>
<divclassName={styles.content}>Comes with built-in mainstream RL algorithms like PPO and GRPO, and supports multi-task RL and agent interaction scenarios. Its effectiveness has been validated in numerous real-world business applications .</div>
<divclassName={styles.label}>Open Source and Accessible</div>
132
-
<divclassName={styles.content}>ROLL is open-sourced on GitHub (https://github.com/alibaba/ROLL) under the Apache License 2.0, backed by an active community and comprehensive documentation .</div>
<divclassName={styles.content}>Join our vibrant open source community, explore cutting-edge reinforcement learning technologies with global AI researchers, and jointly promote the future of LLM and RL</div>
144
-
</div>
145
-
<divclassName={styles.cards}>
146
-
<divclassName={styles.card}>
147
-
<divclassName={styles.label}>How to Contribute</div>
148
-
<div>
149
-
<p>Contribute algorithm implementations and performance optimizations</p>
150
-
<p>Share experimental results and best practices</p>
children: <divclassName={styles.content}>Ray-based distributed architecture supporting mainstream engines like vLLM, SGLang, Megatron-Core, seamlessly scaling from single machine to large GPU clusters</div>,
children: <divclassName={styles.content}>Support for multi-task joint training including math reasoning, code generation, and dialogue, with dynamic sampling rate and data weight adjustment</div>,
91
+
},
92
+
{
93
+
key: '3',
94
+
label: <divclassName={styles.label}>Extremely Easy to Use</div>,
95
+
children: <divclassName={styles.content}>Gym-style clean API design with modular architecture for flexible extension, one-click switching between different backend engines and algorithm configurations</div>,
<divclassName={styles.content}>ROLL framework provides comprehensive reinforcement learning support, from model training to agent deployment, every aspect is carefully optimized to make AI training more efficient</div>
<divclassName={styles.content}>Built on a Ray-based distributed architecture, it supports large-scale cluster training at the thousand-GPU level. Its innovative Rollout scheduler and AutoDeviceMapping module dramatically improve GPU resource utilization .</div>
<divclassName={styles.label}>Extreme Training Efficiency</div>
124
+
<divclassName={styles.content}>Integrates cutting-edge technologies like Megatron-Core, SGLang, and vLLM to significantly accelerate the model training and inference sampling processes .</div>
<divclassName={styles.content}>Comes with built-in mainstream RL algorithms like PPO and GRPO, and supports multi-task RL and agent interaction scenarios. Its effectiveness has been validated in numerous real-world business applications .</div>
<divclassName={styles.label}>Open Source and Accessible</div>
138
+
<divclassName={styles.content}>ROLL is open-sourced on GitHub (https://github.com/alibaba/ROLL) under the Apache License 2.0, backed by an active community and comprehensive documentation .</div>
<divclassName={styles.content}>Join our vibrant open source community, explore cutting-edge reinforcement learning technologies with global AI researchers, and jointly promote the future of LLM and RL</div>
150
+
</div>
151
+
<divclassName={styles.cards}>
152
+
<divclassName={styles.card}>
153
+
<divclassName={styles.label}>How to Contribute</div>
154
+
<div>
155
+
<p>Contribute algorithm implementations and performance optimizations</p>
156
+
<p>Share experimental results and best practices</p>
0 commit comments