Install from Dockerfile

README.md

@@ -123,13 +123,30 @@ source .venv/bin/activate
 pip install -e .[dev]
 # for zsh
 pip install -e .\[dev\]
+
+# Install flash-attn after all dependencies are installed
+# Note: flash-attn will take a long time to compile, please be patient.
+pip install flash-attn -v
+# Try the following command if you encounter errors during installation
+# pip install flash-attn -v --no-build-isolation
 ```
 
+Installation from docker:
 
+We provided a dockerfile for Trinity-RFT (trinity)
 
-Installation with pip:
-(coming soon)
+```shell
+git clone https://github.com/modelscope/Trinity-RFT
+cd Trinity-RFT
 
+# build the docker image
+# Note: you can edit the dockerfile to customize the environment
+# e.g., use pip mirrors or set api key
+docker build -f scripts/docker/Dockerfile -t trinity-rft:latest .
+
+# run the docker image
+docker run -it --gpus all --shm-size="64g" --rm -v $PWD:/workspace -v <root_path_of_data_and_checkpoints>:/data trinity-rft:latest
+```
 
 
 ### Step 2: prepare dataset and model
@@ -263,18 +280,6 @@ Please refer to [this document](./docs/sphinx_doc/source/tutorial/trinity_progra
 This project is currently under active development, and we welcome contributions from the community!
 
 
-
-Installation for development:
-
-```shell
-# for bash
-pip install -e .[dev]
-# for zsh
-pip install -e .\[dev\]
-```
-
-
-
 Code style check:
 
 ```shell

docs/sphinx_doc/source/main.md

@@ -115,12 +115,31 @@ source .venv/bin/activate
 pip install -e .[dev]
 # for zsh
 pip install -e .\[dev\]
+
+# Install flash-attn after all dependencies are installed
+# Note: flash-attn will take a long time to compile, please be patient.
+pip install flash-attn -v
+# Try the following command if you encounter errors during installation
+# pip install flash-attn -v --no-build-isolation
 ```
 
 
+Installation from docker:
+
+We provided a dockerfile for Trinity-RFT (trinity)
+
+```shell
+git clone https://github.com/modelscope/Trinity-RFT
+cd Trinity-RFT
+
+# build the docker image
+# Note: you can edit the dockerfile to customize the environment
+# e.g., use pip mirrors or set api key
+docker build -f scripts/docker/Dockerfile -t trinity-rft:latest .
 
-Installation with pip:
-(coming soon)
+# run the docker image
+docker run -it --gpus all --shm-size="64g" --rm -v $PWD:/workspace -v <root_path_of_data_and_checkpoints>:/data trinity-rft:latest
+```
 
 
 
@@ -255,18 +274,6 @@ Please refer to [this document](tutorial/trinity_programming_guide.md).
 This project is currently under active development, and we welcome contributions from the community!
 
 
-
-Installation for development:
-
-```shell
-# for bash
-pip install -e .[dev]
-# for zsh
-pip install -e .\[dev\]
-```
-
-
-
 Code style check:
 
 ```shell

pyproject.toml

@@ -23,7 +23,6 @@ dependencies = [
     "ray==2.43.0",
     "vllm==0.8.3",
     "tensordict==0.6.2",
-    "redis",
     "wandb",
     "omegaconf",
     "sqlalchemy",

scripts/docker/Dockerfile

@@ -0,0 +1,44 @@
+# Build and run the docker image with the following command:
+#
+# cd <Trinity-RFT root dir>
+# docker build -f scripts/docker/Dockerfile -t trinity-rft:latest .
+# docker run -it --gpus all --shm-size="64g" --rm -v $PWD:/workspace -v <root_path_of_data_and_checkpoints>:/data trinity-rft:latest
+
+
+FROM nvcr.io/nvidia/cuda:12.4.1-cudnn-devel-ubuntu22.04
+
+WORKDIR /workspace
+
+RUN apt update && apt install -y \
+    build-essential \
+    curl \
+    git \
+    wget \
+    vim \
+    tmux \
+    python3 \
+    python3-pip \
+    python3-dev \
+    python3-packaging \
+    && rm -rf /var/lib/apt/lists/* \
+    && ln -sf /usr/bin/python3 /usr/bin/python
+
+
+# For Aliyun users: update pip mirror to aliyun to speed up pip install
+RUN pip config set global.index-url http://mirrors.cloud.aliyuncs.com/pypi/simple/ \
+   && pip config set global.trusted-host mirrors.cloud.aliyuncs.com
+
+# copy the Trinity-RFT dir into the workspace
+COPY . .
+
+RUN pip install --upgrade pip && pip install -e .[dev] && pip install flash-attn
+
+# Set Env variables
+
+# WANDB
+# ENV WANDB_API_KEY=
+# ENV WANDB_BASE_URL=
+
+# LLM API
+# ENV OPENAI_API_KEY=
+# ENV DASH_API_KEY=

trinity/common/verl_config.py

@@ -268,6 +268,7 @@ def synchronize_config(self, config: Config) -> None:
         self.trainer.nnodes = config.cluster.node_num - rollout_node_num
         self.actor_rollout_ref.model.path = config.model.model_path
         self.critic.model.path = config.model.critic_model_path
+        self.critic.model.tokenizer_path = config.model.critic_model_path
 
         if config.cluster.node_num == 1:
             # for single node scenarios, rollout and training are on the same node