Release v0.6.0

.github/workflows/docker/env

@@ -2,7 +2,7 @@ TRINITY_NODE1_GPU_0=4
 TRINITY_NODE1_GPU_1=5
 TRINITY_NODE2_GPU_0=6
 TRINITY_NODE2_GPU_1=7
-TRINITY_DOCKER_IMAGE=trinity-rft-unittest:20260622
+TRINITY_DOCKER_IMAGE=trinity-rft-unittest:20260626
 TRINITY_MOUNT_DIR=/mnt1/checkpoints
 TRINITY_HF_ENDPOINT=https://hf-mirror.com
 TRINITY_PYPI_INDEX_URL=http://mirrors.cloud.aliyuncs.com/pypi/simple/

README.md

@@ -32,6 +32,7 @@ Trinity-RFT provides functionalities for users with different backgrounds and ob
 
 ## 🚀 News
 
+* [2026-06] [[Release Notes]](https://github.com/agentscope-ai/Trinity-RFT/releases/tag/v0.6.0) Trinity-RFT v0.6.0 released: SGLang support, optimized fully async weights synchronization and scheduling to reduce bubbles, improved MoE training stability, and verl 0.8.0 upgrade.
 * [2026-06] 🎉 Two papers accepted by ICML 2026: [Learn-to-Ask](https://github.com/agentscope-ai/Trinity-RFT/tree/main/examples/learn_to_ask), and [LLM-RL entropy dynamics](https://github.com/agentscope-ai/Trinity-RFT/tree/main/examples/entropy).
 * [2026-04] [[Release Notes]](https://github.com/agentscope-ai/Trinity-RFT/releases/tag/v0.5.2) Trinity-RFT v0.5.2 released: support for Qwen3.5 series, bug fixes and optimizations.
 * [2026-03] 🤖 Trinity-RFT empowers the training of CoPaw-Flash, building a small agent model better suited for localized scenarios. Feel free to try CoPaw-Flash on [CoPaw](https://github.com/agentscope-ai/CoPaw); models are also available on [ModelScope](https://www.modelscope.cn/organization/AgentScope) and [HuggingFace](https://huggingface.co/agentscope-ai) ([News](https://mp.weixin.qq.com/s/-BXNU_PMi6QJuwSB5BqTbQ)).

README_zh.md

@@ -41,6 +41,7 @@ Trinity-RFT 面向不同背景和目标的用户提供相应功能：
 
 ## 🚀 新闻
 
+* [2026-06] [[发布说明]](https://github.com/agentscope-ai/Trinity-RFT/releases/tag/v0.6.0) Trinity-RFT v0.6.0 发布：支持 SGLang，优化全异步参数同步以及调度策略以减少空泡，优化 MoE 模型训练稳定性，升级到 verl 0.8.0。
 * [2026-06] 🎉 两篇论文被 ICML 2026 接收：[Learn-to-Ask](https://github.com/agentscope-ai/Trinity-RFT/tree/main/examples/learn_to_ask) 和 [LLM-RL 熵动力学](https://github.com/agentscope-ai/Trinity-RFT/tree/main/examples/entropy)。
 * [2026-04] [[发布说明]](https://github.com/agentscope-ai/Trinity-RFT/releases/tag/v0.5.2) Trinity-RFT v0.5.2 发布：支持 Qwen3.5 系列，修复 Bug 并进行多项优化。
 * [2026-03] 🤖 Trinity-RFT 助力 CoPaw-Flash 训练，打造更懂本地化场景的智能体小模型。欢迎到 [CoPaw](https://github.com/agentscope-ai/CoPaw) 试用 CoPaw-Flash，模型下载请见 [ModelScope](https://www.modelscope.cn/organization/AgentScope) 和 [HuggingFace](https://huggingface.co/agentscope-ai)（[新闻](https://mp.weixin.qq.com/s/-BXNU_PMi6QJuwSB5BqTbQ)）。

docker/Dockerfile

@@ -40,7 +40,7 @@ RUN pip install uv && uv venv /opt/venv --python=python3.12
 
 # Install  Trinity-RFT
 RUN . /opt/venv/bin/activate && \
-    uv pip install -e .[mm,dev,tinker,data,agent] && \
+    uv pip install -e .[mm,dev,tinker,agent] && \
     uv pip install vllm==0.23.0
 
 # Install flash attention
@@ -72,8 +72,11 @@ RUN . /opt/venv/bin/activate && MAX_JOBS=${BUILD_JOBS} \
     git+https://github.com/NVIDIA/apex.git
 
 # Install SGLang
-RUN . /opt/venv/bin/activate && uv pip install sglang==0.5.13 --prerelease=allow && \
-    uv pip install transformers==5.12.1
+RUN . /opt/venv/bin/activate && uv pip install sglang==0.5.13 --prerelease=allow
+
+# Install some other dependencies
+RUN . /opt/venv/bin/activate && uv pip install py-data-juicer==1.5.2 && \
+    uv pip install transformers==5.12.1 fsspec==2026.4.0 numpy==2.4.6
 
 # Set Env variables
 # ENV LD_LIBRARY_PATH=/usr/lib/x86_64-linux-gnu:/usr/local/nvidia/lib:/usr/local/nvidia/lib64:/usr/local/cuda/lib64