diff --git a/README.md b/README.md index 793b87cdf8..ba04f8a239 100644 --- a/README.md +++ b/README.md @@ -39,6 +39,31 @@ Trinity-RFT provides functionalities for users with different backgrounds and ob * 📊 **Data engineers:** Create RFT datasets and build data pipelines for cleaning, augmentation, and human-in-the-loop scenarios [[tutorial]](https://modelscope.github.io/Trinity-RFT/en/main/tutorial/develop_operator.html) + +## 🚀 News + +* [2025-12] Trinity-RFT powers the medical and health business of "Taobao Shangou", enabling the AI agent to understand vague symptoms, proactively ask follow-up questions, and provide precise recommendations ([News](https://tech.china.com.cn/sx/20251201/411376.shtml)). +* [2025-11] [[Release Notes](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.3.3)] Trinity-RFT v0.3.3 released: bug fixes. +* [2025-11] Introducing [Learn-to-Ask](https://github.com/modelscope/Trinity-RFT/tree/main/examples/learn_to_ask): a framework for training proactive dialogue agents from offline expert data ([paper](https://arxiv.org/pdf/2510.25441)). +* [2025-11] Introducing [BOTS](https://github.com/modelscope/Trinity-RFT/tree/main/examples/bots): online RL task selection for efficient LLM fine-tuning ([paper](https://arxiv.org/pdf/2510.26374)). +* [2025-09] [Our paper](https://arxiv.org/pdf/2509.24203) reveals a novel off-policy interpretation for group-relative REINFORCE and its variants like GRPO and AsymRE ([implementation](https://github.com/modelscope/Trinity-RFT/tree/main/examples/rec_gsm8k)). +* [2025-08] Introducing [CHORD](https://github.com/modelscope/Trinity-RFT/tree/main/examples/mix_chord): dynamic SFT + RL integration for advanced LLM fine-tuning ([paper](https://arxiv.org/pdf/2508.11408)). + +
More... + +
+ + ## 🔨 Tutorials and Guidelines @@ -86,21 +111,25 @@ Trinity-RFT provides functionalities for users with different backgrounds and ob -## 🚀 News +## 🔧 Supported Algorithms + +We list some algorithms supported by Trinity-RFT in the following table. For more details, the concrete configurations are shown in the [Algorithm module](https://github.com/modelscope/Trinity-RFT/blob/main/trinity/algorithm/algorithm.py). You can also set up new algorithms by customizing different components, see [tutorial](https://modelscope.github.io/Trinity-RFT/en/main/tutorial/develop_algorithm.html). + +| Algorithm | Doc / Example | Source Code | Key Configurations | +|:-----------|:-----------|:---------------|:-----------| +| PPO [[Paper](https://arxiv.org/pdf/1707.06347)] | [[Doc](https://modelscope.github.io/Trinity-RFT/en/main/tutorial/example_reasoning_basic.html)] [[Countdown Example](https://github.com/modelscope/Trinity-RFT/tree/main/examples/ppo_countdown)] | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/ppo_policy_loss.py)] | `algorithm_type: ppo` | +| GRPO [[Paper](https://arxiv.org/pdf/2402.03300)] | [[Doc](https://modelscope.github.io/Trinity-RFT/en/main/tutorial/example_reasoning_basic.html)] [[GSM8K Example](https://github.com/modelscope/Trinity-RFT/tree/main/examples/grpo_gsm8k)]| [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/grpo_advantage.py)] | `algorithm_type: grpo` | +| CHORD 💡 [[Paper](https://arxiv.org/pdf/2508.11408)] | [[Doc](https://modelscope.github.io/Trinity-RFT/en/main/tutorial/example_mix_algo.html)] [[ToolACE Example](https://github.com/modelscope/Trinity-RFT/blob/main/examples/mix_chord/mix_chord_toolace.yaml)] | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/chord_policy_loss.py)] | `algorithm_type: mix_chord` | +| REC Series 💡 [[Paper](https://arxiv.org/pdf/2509.24203)] | [[GSM8K Example](https://github.com/modelscope/Trinity-RFT/tree/main/examples/rec_gsm8k)] | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/rec_policy_loss.py)] | `algorithm_type: rec` | +| RLOO [[Paper](https://arxiv.org/pdf/2402.14740)] | - | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/rloo_advantage.py)] | `algorithm_type: rloo` | +| REINFORCE++ [[Paper](https://arxiv.org/pdf/2501.03262)] | - | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/reinforce_advantage.py)] | `algorithm_type: reinforceplusplus` | +| GSPO [[Paper](https://arxiv.org/pdf/2507.18071)] | - | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/gspo_policy_loss.py)] | `algorithm_type: gspo` | +| TOPR [[Paper](https://arxiv.org/pdf/2503.14286)] | [[GSM8K Example](https://github.com/modelscope/Trinity-RFT/tree/main/examples/topr_gsm8k)] | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/topr_policy_loss.py)] | `algorithm_type: topr` | +| sPPO [[Paper](https://arxiv.org/pdf/2108.05828)] | [[GSM8K Example](https://github.com/modelscope/Trinity-RFT/tree/main/examples/sppo_gsm8k)] | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/sppo_loss_fn.py)] | `algorithm_type: sppo` | +| AsymRE [[Paper](https://arxiv.org/pdf/2506.20520)] | [[GSM8K Example](https://github.com/modelscope/Trinity-RFT/tree/main/examples/asymre_gsm8k)] | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/asymre_advantage.py)] | `algorithm_type: asymre` | +| CISPO [[Paper](https://arxiv.org/pdf/2506.13585)] | - | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/cispo_policy_loss.py)] | `algorithm_type: cispo` | +| SAPO [[Paper](https://arxiv.org/pdf/2511.20347)] | - | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/sapo_policy_loss.py)] | `algorithm_type: sapo` | -* [2025-11] [[Release Notes](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.3.3)] Trinity-RFT v0.3.3 released: bug fixes. -* [2025-11] Introducing [Learn-to-Ask](https://github.com/modelscope/Trinity-RFT/tree/main/examples/learn_to_ask): a framework for training proactive dialogue agents from offline expert data ([paper](https://arxiv.org/pdf/2510.25441)). -* [2025-11] Introducing [BOTS](https://github.com/modelscope/Trinity-RFT/tree/main/examples/bots): online RL task selection for efficient LLM fine-tuning ([paper](https://arxiv.org/pdf/2510.26374)). -* [2025-11] [[Release Notes](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.3.2)] Trinity-RFT v0.3.2 released: bug fixes and advanced task selection & scheduling. -* [2025-10] [[Release Notes](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.3.1)] Trinity-RFT v0.3.1 released: multi-stage training support, improved agentic RL examples, LoRA support, debug mode and new RL algorithms. -* [2025-09] [[Release Notes](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.3.0)] Trinity-RFT v0.3.0 released: enhanced Buffer, FSDP2 & Megatron support, multi-modal models, and new RL algorithms/examples. -* [2025-08] Introducing [CHORD](https://github.com/modelscope/Trinity-RFT/tree/main/examples/mix_chord): dynamic SFT + RL integration for advanced LLM fine-tuning ([paper](https://arxiv.org/pdf/2508.11408)). -* [2025-08] [[Release Notes](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.2.1)] Trinity-RFT v0.2.1 released. -* [2025-07] [[Release Notes](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.2.0)] Trinity-RFT v0.2.0 released. -* [2025-07] Technical report (arXiv v2) updated with new features, examples, and experiments: [link](https://arxiv.org/abs/2505.17826). -* [2025-06] [[Release Notes](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.1.1)] Trinity-RFT v0.1.1 released. -* [2025-05] [[Release Notes](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.1.0)] Trinity-RFT v0.1.0 released, plus [technical report](https://arxiv.org/abs/2505.17826). -* [2025-04] Trinity-RFT open sourced. --- diff --git a/README_zh.md b/README_zh.md index 423e76e103..02ca29c744 100644 --- a/README_zh.md +++ b/README_zh.md @@ -39,6 +39,31 @@ Trinity-RFT 面向不同背景和目标的用户提供相应功能： +## 🚀 新闻 + +* [2025-12] Trinity-RFT 助力淘宝闪购医药健康业务，让 AI 智能体能够理解模糊症状、主动询问后续问题，并提供精准推荐（[新闻](https://tech.china.com.cn/sx/20251201/411376.shtml)）。 +* [2025-11] [[发布说明](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.3.3)] Trinity-RFT v0.3.3 发布：修复若干 Bug。 +* [2025-11] 推出 [Learn-to-Ask](https://github.com/modelscope/Trinity-RFT/tree/main/examples/learn_to_ask)：利用离线专家数据，训练具备主动问询能力的对话智能体（[论文](https://arxiv.org/pdf/2510.25441)）. +* [2025-11] 推出 [BOTS](https://github.com/modelscope/Trinity-RFT/tree/main/examples/bots)：在线 RL 任务选择，实现高效 LLM 微调（[论文](https://arxiv.org/pdf/2510.26374)）。 +* [2025-09] 我们的 [论文](https://arxiv.org/pdf/2509.24203) 揭示了 group-relative REINFORCE 及其变种（如 GRPO 和 AsymRE）的 off-policy 解释（[代码](https://github.com/modelscope/Trinity-RFT/tree/main/examples/rec_gsm8k)）。 +* [2025-08] 推出 [CHORD](https://github.com/modelscope/Trinity-RFT/tree/main/examples/mix_chord)：动态 SFT + RL 集成，实现进阶 LLM 微调（[论文](https://arxiv.org/pdf/2508.11408)）。 + +
More... + +
+ + + ## 🔨 教程与指南 @@ -85,22 +110,26 @@ Trinity-RFT 面向不同背景和目标的用户提供相应功能： 系统架构 +## 🔨 算法支持 + +下表列出了 Trinity-RFT 支持的算法，更多算法请参考 [算法模块](https://github.com/modelscope/Trinity-RFT/blob/main/trinity/algorithm/algorithm.py)。您也可以通过自定义不同的模块来构建新算法，参见 [教程](https://modelscope.github.io/Trinity-RFT/zh/main/tutorial/develop_algorithm.html)。 + +| 算法 | 文档/示例 | 核心代码 | 关键配置 | +|:-----------|:-----------|:---------------|:-----------| +| PPO [[论文](https://arxiv.org/pdf/1707.06347)] | [[文档](https://modelscope.github.io/Trinity-RFT/zh/main/tutorial/example_reasoning_basic.html)] [[Countdown 例子](https://github.com/modelscope/Trinity-RFT/tree/main/examples/ppo_countdown)] | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/ppo_policy_loss.py)] | `algorithm_type: ppo` | +| GRPO [[论文](https://arxiv.org/pdf/2402.03300)] | [[文档](https://modelscope.github.io/Trinity-RFT/zh/main/tutorial/example_reasoning_basic.html)] [[GSM8K 例子](https://github.com/modelscope/Trinity-RFT/tree/main/examples/grpo_gsm8k)]| [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/grpo_advantage.py)] | `algorithm_type: grpo` | +| CHORD 💡 [[论文](https://arxiv.org/pdf/2508.11408)] | [[文档](https://modelscope.github.io/Trinity-RFT/zh/main/tutorial/example_mix_algo.html)] [[ToolACE 例子](https://github.com/modelscope/Trinity-RFT/blob/main/examples/mix_chord/mix_chord_toolace.yaml)] | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/chord_policy_loss.py)] | `algorithm_type: mix_chord` | +| REC Series 💡 [[论文](https://arxiv.org/pdf/2509.24203)] | [[GSM8K 例子](https://github.com/modelscope/Trinity-RFT/tree/main/examples/rec_gsm8k)] | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/rec_policy_loss.py)] | `algorithm_type: rec` | +| RLOO [[论文](https://arxiv.org/pdf/2402.14740)] | - | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/rloo_advantage.py)] | `algorithm_type: rloo` | +| REINFORCE++ [[论文](https://arxiv.org/pdf/2501.03262)] | - | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/reinforce_advantage.py)] | `algorithm_type: reinforceplusplus` | +| GSPO [[论文](https://arxiv.org/pdf/2507.18071)] | - | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/gspo_policy_loss.py)] | `algorithm_type: gspo` | +| TOPR [[论文](https://arxiv.org/pdf/2503.14286)] | [[GSM8K 例子](https://github.com/modelscope/Trinity-RFT/tree/main/examples/topr_gsm8k)] | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/topr_policy_loss.py)] | `algorithm_type: topr` | +| sPPO [[论文](https://arxiv.org/pdf/2108.05828)] | [[GSM8K 例子](https://github.com/modelscope/Trinity-RFT/tree/main/examples/sppo_gsm8k)] | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/sppo_loss_fn.py)] | `algorithm_type: sppo` | +| AsymRE [[论文](https://arxiv.org/pdf/2506.20520)] | [[GSM8K 例子](https://github.com/modelscope/Trinity-RFT/tree/main/examples/asymre_gsm8k)] | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/asymre_advantage.py)] | `algorithm_type: asymre` | +| CISPO [[论文](https://arxiv.org/pdf/2506.13585)] | - | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/cispo_policy_loss.py)] | `algorithm_type: cispo` | +| SAPO [[论文](https://arxiv.org/pdf/2511.20347)] | - | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/sapo_policy_loss.py)] | `algorithm_type: sapo` | -## 🚀 新闻 -* [2025-11] [[发布说明](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.3.3)] Trinity-RFT v0.3.3 发布：修复若干 Bug。 -* [2025-11] 推出 [Learn-to-Ask](https://github.com/modelscope/Trinity-RFT/tree/main/examples/learn_to_ask)：利用离线专家数据，训练具备主动问询能力的对话智能体（[论文](https://arxiv.org/pdf/2510.25441)）. -* [2025-11] 推出 [BOTS](https://github.com/modelscope/Trinity-RFT/tree/main/examples/bots)：在线 RL 任务选择，实现高效 LLM 微调（[论文](https://arxiv.org/pdf/2510.26374)）。 -* [2025-11] [[发布说明](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.3.2)] Trinity-RFT v0.3.2 发布：修复若干 Bug 并支持进阶的任务选择和调度。 -* [2025-10] [[发布说明](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.3.1)] Trinity-RFT v0.3.1 发布：多阶段训练支持、改进的智能体 RL 示例、LoRA 支持、调试模式和全新 RL 算法。 -* [2025-09] [[发布说明](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.3.0)] Trinity-RFT v0.3.0 发布：增强的 Buffer、FSDP2 & Megatron 支持，多模态模型，以及全新 RL 算法/示例。 -* [2025-08] 推出 [CHORD](https://github.com/modelscope/Trinity-RFT/tree/main/examples/mix_chord)：动态 SFT + RL 集成，实现进阶 LLM 微调（[论文](https://arxiv.org/pdf/2508.11408)）。 -* [2025-08] [[发布说明](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.2.1)] Trinity-RFT v0.2.1 发布。 -* [2025-07] [[发布说明](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.2.0)] Trinity-RFT v0.2.0 发布。 -* [2025-07] 技术报告（arXiv v2）更新，包含新功能、示例和实验：[链接](https://arxiv.org/abs/2505.17826)。 -* [2025-06] [[发布说明](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.1.1)] Trinity-RFT v0.1.1 发布。 -* [2025-05] [[发布说明](https://github.com/modelscope/Trinity-RFT/releases/tag/v0.1.0)] Trinity-RFT v0.1.0 发布，同时发布 [技术报告](https://arxiv.org/abs/2505.17826)。 -* [2025-04] Trinity-RFT 开源。 --- diff --git a/docs/sphinx_doc/source/main.md b/docs/sphinx_doc/source/main.md index 429a6d2c4c..65ff0af09d 100644 --- a/docs/sphinx_doc/source/main.md +++ b/docs/sphinx_doc/source/main.md @@ -66,6 +66,27 @@ Trinity-RFT provides functionalities for users with different backgrounds and ob +## 🔧 Supported Algorithms + +We list some algorithms supported by Trinity-RFT in the following table. For more details, the concrete configurations are shown in the [Algorithm module](https://github.com/modelscope/Trinity-RFT/blob/main/trinity/algorithm/algorithm.py). You can also set up new algorithms by customizing different components, see [tutorial](/tutorial/develop_algorithm.md). + +| Algorithm | Doc / Example | Source Code | Key Configurations | +|:-----------|:-----------|:---------------|:-----------| +| PPO [[Paper](https://arxiv.org/pdf/1707.06347)] | [[Doc](https://modelscope.github.io/Trinity-RFT/en/main/tutorial/example_reasoning_basic.html)] [[Countdown Example](https://github.com/modelscope/Trinity-RFT/tree/main/examples/ppo_countdown)] | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/ppo_policy_loss.py)] | `algorithm_type: ppo` | +| GRPO [[Paper](https://arxiv.org/pdf/2402.03300)] | [[Doc](https://modelscope.github.io/Trinity-RFT/en/main/tutorial/example_reasoning_basic.html)] [[GSM8K Example](https://github.com/modelscope/Trinity-RFT/tree/main/examples/grpo_gsm8k)]| [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/grpo_advantage.py)] | `algorithm_type: grpo` | +| CHORD 💡 [[Paper](https://arxiv.org/pdf/2508.11408)] | [[Doc](https://modelscope.github.io/Trinity-RFT/en/main/tutorial/example_mix_algo.html)] [[ToolACE Example](https://github.com/modelscope/Trinity-RFT/blob/main/examples/mix_chord/mix_chord_toolace.yaml)] | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/chord_policy_loss.py)] | `algorithm_type: mix_chord` | +| REC Series 💡 [[Paper](https://arxiv.org/pdf/2509.24203)] | [[GSM8K Example](https://github.com/modelscope/Trinity-RFT/tree/main/examples/rec_gsm8k)] | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/rec_policy_loss.py)] | `algorithm_type: rec` | +| RLOO [[Paper](https://arxiv.org/pdf/2402.14740)] | - | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/rloo_advantage.py)] | `algorithm_type: rloo` | +| REINFORCE++ [[Paper](https://arxiv.org/pdf/2501.03262)] | - | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/reinforce_advantage.py)] | `algorithm_type: reinforceplusplus` | +| GSPO [[Paper](https://arxiv.org/pdf/2507.18071)] | - | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/gspo_policy_loss.py)] | `algorithm_type: gspo` | +| TOPR [[Paper](https://arxiv.org/pdf/2503.14286)] | [[GSM8K Example](https://github.com/modelscope/Trinity-RFT/tree/main/examples/topr_gsm8k)] | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/topr_policy_loss.py)] | `algorithm_type: topr` | +| sPPO [[Paper](https://arxiv.org/pdf/2108.05828)] | [[GSM8K Example](https://github.com/modelscope/Trinity-RFT/tree/main/examples/sppo_gsm8k)] | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/sppo_loss_fn.py)] | `algorithm_type: sppo` | +| AsymRE [[Paper](https://arxiv.org/pdf/2506.20520)] | [[GSM8K Example](https://github.com/modelscope/Trinity-RFT/tree/main/examples/asymre_gsm8k)] | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/asymre_advantage.py)] | `algorithm_type: asymre` | +| CISPO [[Paper](https://arxiv.org/pdf/2506.13585)] | - | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/cispo_policy_loss.py)] | `algorithm_type: cispo` | +| SAPO [[Paper](https://arxiv.org/pdf/2511.20347)] | - | [[Code](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/sapo_policy_loss.py)] | `algorithm_type: sapo` | + + + ## Acknowledgements This project is built upon many excellent open-source projects, including: diff --git a/docs/sphinx_doc/source_zh/main.md b/docs/sphinx_doc/source_zh/main.md index 99d89a1ad2..e215f17b35 100644 --- a/docs/sphinx_doc/source_zh/main.md +++ b/docs/sphinx_doc/source_zh/main.md @@ -64,6 +64,27 @@ Trinity-RFT 面向不同背景和目标的用户提供相应功能： +## 🔨 算法支持 + +下表列出了 Trinity-RFT 支持的算法，更多算法请参考 [算法模块](https://github.com/modelscope/Trinity-RFT/blob/main/trinity/algorithm/algorithm.py)。您也可以通过自定义不同的模块来构建新算法，参见 [教程](/tutorial/develop_algorithm.md)。 + +| 算法 | 文档/示例 | 核心代码 | 关键配置 | +|:-----------|:-----------|:---------------|:-----------| +| PPO [[论文](https://arxiv.org/pdf/1707.06347)] | [[文档](https://modelscope.github.io/Trinity-RFT/zh/main/tutorial/example_reasoning_basic.html)] [[Countdown 例子](https://github.com/modelscope/Trinity-RFT/tree/main/examples/ppo_countdown)] | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/ppo_policy_loss.py)] | `algorithm_type: ppo` | +| GRPO [[论文](https://arxiv.org/pdf/2402.03300)] | [[文档](https://modelscope.github.io/Trinity-RFT/zh/main/tutorial/example_reasoning_basic.html)] [[GSM8K 例子](https://github.com/modelscope/Trinity-RFT/tree/main/examples/grpo_gsm8k)]| [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/grpo_advantage.py)] | `algorithm_type: grpo` | +| CHORD 💡 [[论文](https://arxiv.org/pdf/2508.11408)] | [[文档](https://modelscope.github.io/Trinity-RFT/zh/main/tutorial/example_mix_algo.html)] [[ToolACE 例子](https://github.com/modelscope/Trinity-RFT/blob/main/examples/mix_chord/mix_chord_toolace.yaml)] | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/chord_policy_loss.py)] | `algorithm_type: mix_chord` | +| REC Series 💡 [[论文](https://arxiv.org/pdf/2509.24203)] | [[GSM8K 例子](https://github.com/modelscope/Trinity-RFT/tree/main/examples/rec_gsm8k)] | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/rec_policy_loss.py)] | `algorithm_type: rec` | +| RLOO [[论文](https://arxiv.org/pdf/2402.14740)] | - | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/rloo_advantage.py)] | `algorithm_type: rloo` | +| REINFORCE++ [[论文](https://arxiv.org/pdf/2501.03262)] | - | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/reinforce_advantage.py)] | `algorithm_type: reinforceplusplus` | +| GSPO [[论文](https://arxiv.org/pdf/2507.18071)] | - | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/gspo_policy_loss.py)] | `algorithm_type: gspo` | +| TOPR [[论文](https://arxiv.org/pdf/2503.14286)] | [[GSM8K 例子](https://github.com/modelscope/Trinity-RFT/tree/main/examples/topr_gsm8k)] | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/topr_policy_loss.py)] | `algorithm_type: topr` | +| sPPO [[论文](https://arxiv.org/pdf/2108.05828)] | [[GSM8K 例子](https://github.com/modelscope/Trinity-RFT/tree/main/examples/sppo_gsm8k)] | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/sppo_loss_fn.py)] | `algorithm_type: sppo` | +| AsymRE [[论文](https://arxiv.org/pdf/2506.20520)] | [[GSM8K 例子](https://github.com/modelscope/Trinity-RFT/tree/main/examples/asymre_gsm8k)] | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/advantage_fn/asymre_advantage.py)] | `algorithm_type: asymre` | +| CISPO [[论文](https://arxiv.org/pdf/2506.13585)] | - | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/cispo_policy_loss.py)] | `algorithm_type: cispo` | +| SAPO [[论文](https://arxiv.org/pdf/2511.20347)] | - | [[代码](https://github.com/modelscope/Trinity-RFT/tree/main/trinity/algorithm/policy_loss_fn/sapo_policy_loss.py)] | `algorithm_type: sapo` | + + + ## 致谢 diff --git a/trinity/algorithm/algorithm.py b/trinity/algorithm/algorithm.py index 397336408d..93f76676f8 100644 --- a/trinity/algorithm/algorithm.py +++ b/trinity/algorithm/algorithm.py @@ -114,6 +114,52 @@ def default_config(cls) -> Dict: } +@ALGORITHM_TYPE.register_module("reinforceplusplus") +class ReinforcePlusPlusAlgorithm(AlgorithmType): + """Reinforce++ algorithm.""" + + use_critic: bool = False + use_reference: bool = True + compute_advantage_in_trainer: bool = True + can_balance_batch: bool = True + schema: str = "experience" + + @classmethod + def default_config(cls) -> Dict: + return { + "repeat_times": 2, + "advantage_fn": "reinforceplusplus", + "sample_strategy": "default", + "policy_loss_fn": "ppo", + "kl_penalty_fn": "none", + "kl_loss_fn": "k2", + "entropy_loss_fn": "default", + } + + +@ALGORITHM_TYPE.register_module("rloo") +class RLOOAlgorithm(AlgorithmType): + """RLOO algorithm.""" + + use_critic: bool = False + use_reference: bool = True + compute_advantage_in_trainer: bool = True + can_balance_batch: bool = True + schema: str = "experience" + + @classmethod + def default_config(cls) -> Dict: + return { + "repeat_times": 2, + "advantage_fn": "rloo", + "sample_strategy": "default", + "policy_loss_fn": "ppo", + "kl_penalty_fn": "none", + "kl_loss_fn": "k2", + "entropy_loss_fn": "default", + } + + @ALGORITHM_TYPE.register_module("opmd") class OPMDAlgorithm(AlgorithmType): """OPMD algorithm.""" @@ -250,6 +296,29 @@ def default_config(cls) -> Dict: } +@ALGORITHM_TYPE.register_module("gspo") +class GSPOAlgorithm(AlgorithmType): + """GSPO algorithm. See https://arxiv.org/pdf/2507.18071""" + + use_critic: bool = False + use_reference: bool = True + compute_advantage_in_trainer: bool = False + can_balance_batch: bool = True + schema: str = "experience" + + @classmethod + def default_config(cls) -> Dict: + return { + "repeat_times": 2, + "advantage_fn": "grpo", + "sample_strategy": "default", + "policy_loss_fn": "gspo", + "kl_penalty_fn": "none", + "kl_loss_fn": "k2", + "entropy_loss_fn": "default", + } + + @ALGORITHM_TYPE.register_module("sapo") class SAPOAlgorithm(AlgorithmType): """SAPO (Soft Adaptive Policy Optimization) algorithm.