diff --git a/.github/workflows/docker.yaml b/.github/workflows/docker.yaml
index 32393ec11e..5c928049f0 100644
--- a/.github/workflows/docker.yaml
+++ b/.github/workflows/docker.yaml
@@ -50,6 +50,7 @@ jobs:
           tags: |
             type=match,pattern=\d.\d.\d,enable=${{ github.event_name == 'release' }}
             type=raw,value=${{ github.event.inputs.version }},enable=${{ github.event_name == 'workflow_dispatch' }}
+            type=raw,value=latest,enable={{is_default_branch}}
       # This step uses the `docker/build-push-action` action to build the image, based on your repository's `Dockerfile`. If the build succeeds, it pushes the image to GitHub Packages.
       # It uses the `context` parameter to define the build's context as the set of files located in the specified path. For more information, see [Usage](https://github.com/docker/build-push-action#usage) in the README of the `docker/build-push-action` repository.
       # It uses the `tags` and `labels` parameters to tag and label the image with the output from the "meta" step.
diff --git a/.github/workflows/docker/docker-compose.yaml b/.github/workflows/docker/docker-compose.yaml
index a923b461eb..0889e57ffd 100644
--- a/.github/workflows/docker/docker-compose.yaml
+++ b/.github/workflows/docker/docker-compose.yaml
@@ -1,6 +1,6 @@
 services:
   trinity-node-1:
-    image: trinity-rft-unittest:20260126
+    image: trinity-rft-unittest:20260205
     cap_add:
       - SYS_PTRACE
     pull_policy: never
@@ -32,7 +32,7 @@ services:
             capabilities: [gpu]
 
   trinity-node-2:
-    image: trinity-rft-unittest:20260126
+    image: trinity-rft-unittest:20260205
     cap_add:
       - SYS_PTRACE
     pull_policy: never
diff --git a/README.md b/README.md
index 9ac0c04fed..afe36c6969 100644
--- a/README.md
+++ b/README.md
@@ -32,6 +32,7 @@ Trinity-RFT provides functionalities for users with different backgrounds and ob
 
 ## 🚀 News
 
+* [2026-02] [[Release Notes]](https://github.com/agentscope-ai/Trinity-RFT/releases/tag/v0.5.0) Trinity-RFT v0.5.0 released: colocate mode for single-GPU scenarios, trainer driven weight synchronization, automatic parallelism setting suggestion, and more.
 * [2026-01] 🎉 Three papers accepted by ICLR 2026: [CHORD](https://github.com/agentscope-ai/Trinity-RFT/tree/main/examples/mix_chord), [BOTS](https://github.com/agentscope-ai/Trinity-RFT/tree/main/examples/bots), and [Group-relative REINFORCE variants](https://github.com/agentscope-ai/Trinity-RFT/tree/main/examples/rec_gsm8k). Try out these new algorithms in Trinity-RFT!
 * [2026-01] [[Release Notes]](https://github.com/agentscope-ai/Trinity-RFT/releases/tag/v0.4.1) Trinity-RFT v0.4.1 released: upgraded verl to v0.7.0, Tinker backend supports OpenAI API, bug fixes.
 * [2026-01] Introducing [R3L](https://github.com/shiweijiezero/R3L): a systematic reflect-then-retry RL mechanism with efficient language-guided exploration and stable off-policy learning ([paper](https://arxiv.org/abs/2601.03715)).
@@ -165,7 +166,7 @@ Before installing, make sure your system meets the following requirements:
 
 - Python: version 3.10 to 3.12 (inclusive)
 - CUDA: version >= 12.8
-- GPUs: at least 2 GPUs
+- GPUs: At least one NVIDIA GPU with [compute capability](https://developer.nvidia.com/cuda/gpus) 8.0 or higher (e.g., RTX 30 series, A100, H100)
 
 **Recommended for first-time users:**
 
diff --git a/README_zh.md b/README_zh.md
index 09688b0f43..15a4ee26f7 100644
--- a/README_zh.md
+++ b/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
 
+* [2026-02] [[发布说明]](https://github.com/agentscope-ai/Trinity-RFT/releases/tag/v0.5.0) Trinity-RFT v0.5.0 发布：单 GPU 场景下的 colocate 模式，trainer 驱动的权重同步，自动并行设置建议等新功能。
 * [2026-01] 🎉 三篇论文被 ICLR 2026 接收：[CHORD](https://github.com/agentscope-ai/Trinity-RFT/tree/main/examples/mix_chord)、[BOTS](https://github.com/agentscope-ai/Trinity-RFT/tree/main/examples/bots) 和 [Group-relative REINFORCE 系列变种](https://github.com/agentscope-ai/Trinity-RFT/tree/main/examples/rec_gsm8k)。在 Trinity-RFT 中尝试这些新算法吧！
 * [2026-01] [[发布说明]](https://github.com/agentscope-ai/Trinity-RFT/releases/tag/v0.4.1) Trinity-RFT v0.4.1 发布：升级 verl 至 v0.7.0，Tinker 后端支持 OpenAI API，修复若干 Bug。
 * [2026-01] 推出 [R3L](https://github.com/shiweijiezero/R3L)：基于反思-重试的强化学习机制，由自然语言反馈引导高效探索，并达成稳定的 off-policy 学习（[论文](https://arxiv.org/abs/2601.03715)）。
@@ -171,7 +172,7 @@ Trinity-RFT 面向不同背景和目标的用户提供相应功能：
 
 - **Python**：版本 3.10 至 3.12（含）
 - **CUDA**：版本 >= 12.8
-- **GPU**：至少 2 块 GPU
+- **GPU**： 至少一块 [compute capability](https://developer.nvidia.com/cuda/gpus) 为 8.0 或更高的 NVIDIA GPU（例如 RTX 30 系列、A100、H100）
 
 ## 源码安装（推荐）
 
diff --git a/docs/sphinx_doc/source/tutorial/trinity_installation.md b/docs/sphinx_doc/source/tutorial/trinity_installation.md
index 96b2b9a0b2..2a32dbf76f 100644
--- a/docs/sphinx_doc/source/tutorial/trinity_installation.md
+++ b/docs/sphinx_doc/source/tutorial/trinity_installation.md
@@ -11,7 +11,7 @@ Make sure your system meets these requirements:
 
 - **Python**: 3.10 – 3.12
 - **CUDA**: 12.8 or higher
-- **GPUs**: At least 2 available
+- **GPUs**: At least one NVIDIA GPU with [compute capability](https://developer.nvidia.com/cuda/gpus) 8.0 or higher (e.g., RTX 30 series, A100, H100)
 
 ### If you don’t have GPUs or prefer not to use them
 
@@ -84,24 +84,6 @@ uv sync --extra vllm --extra dev --extra flash_attn
 
 ---
 
-## Via PyPI
-
-If you just want to use the package without modifying the code:
-
-```bash
-pip install trinity-rft
-pip install flash-attn==2.8.1
-```
-
-Or with `uv`:
-
-```bash
-uv pip install trinity-rft
-uv pip install flash-attn==2.8.1
-```
-
----
-
 ## Using Docker
 
 You can download the Trinity-RFT Docker image from Github Container Registry or build it locally.
@@ -150,6 +132,24 @@ docker run -it \
 
 ---
 
+## Via PyPI
+
+If you just want to use the package without modifying the code:
+
+```bash
+pip install trinity-rft
+pip install flash-attn==2.8.1  --no-build-isolation
+```
+
+Or with `uv`:
+
+```bash
+uv pip install trinity-rft
+uv pip install flash-attn==2.8.1  --no-build-isolation
+```
+
+---
+
 ```{note}
 For training with **Megatron-LM**, please refer to {ref}`Megatron-LM Backend <Megatron-LM>`.
 ```
diff --git a/docs/sphinx_doc/source_zh/tutorial/trinity_installation.md b/docs/sphinx_doc/source_zh/tutorial/trinity_installation.md
index 6304ba35cf..5787a72649 100644
--- a/docs/sphinx_doc/source_zh/tutorial/trinity_installation.md
+++ b/docs/sphinx_doc/source_zh/tutorial/trinity_installation.md
@@ -11,7 +11,7 @@
 
 - **Python**：3.10 – 3.12
 - **CUDA**：12.8 或更高版本
-- **GPU**：至少 2 块可用
+- **GPU**：至少一块 [compute capability](https://developer.nvidia.com/cuda/gpus) 为 8.0 或更高的 NVIDIA GPU（例如 RTX 30 系列、A100、H100）
 
 ### 如果您没有 GPU 或不希望使用 GPU
 
@@ -139,14 +139,14 @@ docker run -it \
 
 ```bash
 pip install trinity-rft
-pip install flash-attn==2.8.1
+pip install flash-attn==2.8.1 --no-build-isolation
 ```
 
 或使用 `uv`：
 
 ```bash
 uv pip install trinity-rft
-uv pip install flash-attn==2.8.1
+uv pip install flash-attn==2.8.1 --no-build-isolation
 ```
 
 ---
diff --git a/pyproject.toml b/pyproject.toml
index 411864a793..b3d2f237c8 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name =  "trinity-rft"
-version = "0.5.0.dev0"
+version = "0.5.0"
 authors = [
     {name="Trinity-RFT Team", email="trinity-rft@outlook.com"},
 ]
@@ -50,9 +50,9 @@ trinity = "trinity.cli.launcher:main"
 
 [project.optional-dependencies]
 vllm = [
-    "vllm>=0.10.2,<=0.14.1,!=0.12.0",
+    "vllm>=0.10.2,<=0.15.1,!=0.12.0",
     # v0.12.0 has a huge performance regression so we exclude it
-    # v0.10.2 is the most stable version, but we allow up to 0.14.1 for new features
+    # v0.10.2 is the most stable version, but we allow up to 0.15.1 for new features
 ]
 data = [
     "py-data-juicer>=1.4.3"
diff --git a/trinity/__init__.py b/trinity/__init__.py
index b29fbc4891..f9c4d8f061 100644
--- a/trinity/__init__.py
+++ b/trinity/__init__.py
@@ -1,4 +1,4 @@
 # -*- coding: utf-8 -*-
 """Trinity-RFT (Reinforcement Fine-Tuning)"""
 
-__version__ = "0.5.0.dev0"
+__version__ = "0.5.0"
diff --git a/trinity/common/models/vllm_model.py b/trinity/common/models/vllm_model.py
index fffc4e17e0..2372db6cc3 100644
--- a/trinity/common/models/vllm_model.py
+++ b/trinity/common/models/vllm_model.py
@@ -149,7 +149,10 @@ def __init__(
 
     async def _initialize_tokenizer(self):
         if self.tokenizer is None:
-            self.tokenizer = await self.async_llm.get_tokenizer()
+            if self.vllm_version >= parse_version("0.15.0"):
+                self.tokenizer = self.async_llm.get_tokenizer()
+            else:
+                self.tokenizer = await self.async_llm.get_tokenizer()
         self.tokenizer.truncation_side = "left"
 
     def _initialize_processor(self):
diff --git a/trinity/common/models/vllm_patch/worker_patch.py b/trinity/common/models/vllm_patch/worker_patch.py
index a0d6f37647..4d6412cc29 100644
--- a/trinity/common/models/vllm_patch/worker_patch.py
+++ b/trinity/common/models/vllm_patch/worker_patch.py
@@ -13,10 +13,10 @@
 def patch_vllm_prompt_logprobs(model_runner: GPUModelRunner):  # noqa: C901
     """Patch vLLM model runner to support prompt logprobs extraction."""
     version = get_vllm_version()
-    if version < parse_version("0.10.2") or version > parse_version("0.14.1"):
+    if version < parse_version("0.10.2") or version > parse_version("0.15.1"):
         raise ValueError(
             f"Unsupported vllm version: {vllm.__version__}. "
-            "This patch requires vllm version >= 0.10.2, <= 0.14.1."
+            "This patch requires vllm version >= 0.10.2, <= 0.15.1."
         )
     is_v0102 = version == parse_version("0.10.2")
 
@@ -150,7 +150,7 @@ def _get_prompt_logprobs_dict_v12(
 
         This is a monkey-patched version of `_get_prompt_logprobs_dict` from
         `vllm.v1.worker.gpu_model_runner.GPUModelRunner` (vLLM versions
-        0.12.0 to 0.14.1).
+        0.12.0 to 0.15.1).
 
         The original function does not apply temperature scaling to logits when
         calculating prompt logprobs, which can lead to incorrect logprob values