deploy: a71c1b6

Author: chenyushuo

_modules/functools.html

@@ -84,7 +84,8 @@
            <div itemprop="articleBody">
 
   <h1>All modules for which code is available</h1>
-<ul><li><a href="sqlalchemy/orm/attributes.html">sqlalchemy.orm.attributes</a></li>
+<ul><li><a href="functools.html">functools</a></li>
+<li><a href="sqlalchemy/orm/attributes.html">sqlalchemy.orm.attributes</a></li>
 <li><a href="sqlalchemy/orm/decl_api.html">sqlalchemy.orm.decl_api</a></li>
 <li><a href="sqlalchemy/orm/decl_base.html">sqlalchemy.orm.decl_base</a></li>
 <li><a href="sqlalchemy/orm/instrumentation.html">sqlalchemy.orm.instrumentation</a></li>
@@ -123,6 +124,10 @@ <h1>All modules for which code is available</h1>
 <li><a href="trinity/explorer/runner_pool.html">trinity.explorer.runner_pool</a></li>
 <li><a href="trinity/explorer/workflow_runner.html">trinity.explorer.workflow_runner</a></li>
 <li><a href="trinity/manager/config_manager.html">trinity.manager.config_manager</a></li>
+<li><a href="trinity/manager/config_registry/config_registry.html">trinity.manager.config_registry.config_registry</a></li>
+<li><a href="trinity/manager/config_registry/explorer_config_manager.html">trinity.manager.config_registry.explorer_config_manager</a></li>
+<li><a href="trinity/manager/config_registry/model_config_manager.html">trinity.manager.config_registry.model_config_manager</a></li>
+<li><a href="trinity/manager/config_registry/trainer_config_manager.html">trinity.manager.config_registry.trainer_config_manager</a></li>
 <li><a href="trinity/manager/manager.html">trinity.manager.manager</a></li>
 <li><a href="trinity/trainer/trainer.html">trinity.trainer.trainer</a></li>
 <li><a href="trinity/trainer/verl/core_algos.html">trinity.trainer.verl.core_algos</a></li>

_modules/index.html

@@ -0,0 +1,52 @@
+trinity.manager.config_registry
+===============================
+Submodules
+----------
+
+trinity.manager.config\_registry.buffer\_config\_manager module
+---------------------------------------------------------------
+
+.. automodule:: trinity.manager.config_registry.buffer_config_manager
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
+trinity.manager.config\_registry.config\_registry module
+--------------------------------------------------------
+
+.. automodule:: trinity.manager.config_registry.config_registry
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
+trinity.manager.config\_registry.explorer\_config\_manager module
+-----------------------------------------------------------------
+
+.. automodule:: trinity.manager.config_registry.explorer_config_manager
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
+trinity.manager.config\_registry.model\_config\_manager module
+--------------------------------------------------------------
+
+.. automodule:: trinity.manager.config_registry.model_config_manager
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
+trinity.manager.config\_registry.trainer\_config\_manager module
+----------------------------------------------------------------
+
+.. automodule:: trinity.manager.config_registry.trainer_config_manager
+   :members:
+   :undoc-members:
+   :show-inheritance:
+
+Module contents
+---------------
+
+.. automodule:: trinity.manager.config_registry
+   :members:
+   :undoc-members:
+   :show-inheritance:

_modules/trinity/manager/config_manager.html

@@ -1,5 +1,13 @@
 trinity.manager
 ===============
+Subpackages
+-----------
+
+.. toctree::
+   :maxdepth: 4
+
+   trinity.manager.config_registry
+
 Submodules
 ----------
 

_modules/trinity/manager/config_registry/config_registry.html

@@ -286,6 +286,126 @@ trinity run --config <your_yaml_file>
 
 ---
 
+## Adding New Config Entries for the Config Generator (Advanced)
+
+### Step 0: Understanding Streamlit
+
+Before adding new parameters to the Config Generator page, it is essential to familiarize yourself with the relevant API and mechanisms of [Streamlit](https://docs.streamlit.io/develop/api-reference). This project primarily utilizes various input components from Streamlit and employs `st.session_state` to store user-input parameters.
+
+### Step 1: Implement New Config Entries
+
+To illustrate the process of creating a new parameter setting for the Config Generator page, we will use `train_batch_size` as an example.
+
+1. Determine the appropriate scope for the parameter. Currently, parameters are categorized into four files:
+   - `trinity/manager/config_registry/buffer_config_manager.py`
+   - `trinity/manager/config_registry/explorer_config_manager.py`
+   - `trinity/manager/config_registry/model_config_manager.py`
+   - `trinity/manager/config_registry/trainer_config_manager.py`
+
+   In this case, `train_batch_size` should be placed in the `buffer_config_manager.py` file.
+
+2. Create a parameter setting function using Streamlit. The function name must follow the convention of starting with 'set_', and the remainder of the name becomes the config name.
+
+3. Decorate the parameter setting function with the `CONFIG_GENERATORS.register_config` decorator. This decorator requires the following information:
+   - Default value of the parameter
+   - Visibility condition (if applicable)
+   - Additional config parameters (if needed)
+
+```{note}
+The `CONFIG_GENERATORS.register_config` decorator automatically passes `key=config_name` as an argument to the registered configuration function. Ensure that your function accepts this keyword argument.
+```
+
+For `train_batch_size`, we will use the following settings:
+- Default value: 96
+- Visibility condition: `lambda: st.session_state["trainer_gpu_num"] > 0`
+- Additional config: `{"_train_batch_size_per_gpu": 16}`
+
+
+Here's the complete code for the `train_batch_size` parameter:
+
+```python
+@CONFIG_GENERATORS.register_config(
+    default_value=96,
+    visible=lambda: st.session_state["trainer_gpu_num"] > 0,
+    other_configs={"_train_batch_size_per_gpu": 16},
+)
+def set_train_batch_size(**kwargs):
+    key = kwargs.get("key")
+    trainer_gpu_num = st.session_state["trainer_gpu_num"]
+    st.session_state[key] = (
+        st.session_state["_train_batch_size_per_gpu"] * st.session_state["trainer_gpu_num"]
+    )
+
+    def on_change():
+        st.session_state["_train_batch_size_per_gpu"] = max(
+            st.session_state[key] // st.session_state["trainer_gpu_num"], 1
+        )
+
+    st.number_input(
+        "Train Batch Size",
+        min_value=trainer_gpu_num,
+        step=trainer_gpu_num,
+        help=_str_for_train_batch_size(),
+        on_change=on_change,
+        **kwargs,
+    )
+```
+
+If the parameter requires validation, create a check function. For `train_batch_size`, we need to ensure it is divisible by `trainer_gpu_num`. If not, a warning should be displayed, and the parameter should be added to `unfinished_fields`.
+
+Decorate the check function with the `CONFIG_GENERATORS.register_check` decorator:
+
+```python
+@CONFIG_GENERATORS.register_check()
+def check_train_batch_size(unfinished_fields: set, key: str):
+    if st.session_state[key] % st.session_state["trainer_gpu_num"] != 0:
+        unfinished_fields.add(key)
+        st.warning(_str_for_train_batch_size())
+```
+
+```{note}
+The `CONFIG_GENERATORS.register_check` decorator automatically receives `key=config_name` and `unfinished_fields=self.unfinished_fields` as arguments. Ensure your function accepts these keyword arguments.
+```
+
+### Step 2: Integrating New Parameters into `config_manager.py`
+
+To successfully integrate new parameters into the `config_manager.py` file, please adhere to the following procedure:
+
+1. Parameter Categorization:
+   Determine the appropriate section for the new parameter based on its functionality. The config generator page is structured into two primary modes:
+   - Beginner Mode: Comprises "Essential Configs" and "Important Configs" sections.
+   - Expert Mode: Includes "Model", "Buffer", "Explorer and Synchronizer", and "Trainer" sections.
+
+2. Parameter Addition:
+   Incorporate the new parameter into the relevant section using the `self.get_configs` method within the `ConfigManager` class.
+
+   Example:
+   ```python
+   class ConfigManager:
+       def _expert_buffer_part(self):
+           self.get_configs("total_epochs", "train_batch_size")
+   ```
+
+3. YAML File Integration:
+   Locate the appropriate position for the new parameter within the YAML file structure. This should be done in the `generate_config` function and its associated sub-functions.
+
+4. Parameter Value Assignment:
+   Utilize `st.session_state` to retrieve the parameter value from the config generator page and assign it to the corresponding field in the YAML.
+
+   Example:
+   ```python
+   class ConfigManager:
+       def _gen_buffer_config(self):
+           buffer_config = {
+               "batch_size": st.session_state["train_batch_size"],
+               # Additional configuration parameters
+           }
+   ```
+
+By meticulously following these steps, you can ensure that new parameters are successfully added to the Config Generator page and properly integrated into the configuration system. This process maintains the integrity and functionality of the configuration management framework.
+
+---
+
 ## Check Code Style
 
 Before submitting the code, make sure it passes the code style check. Follow these steps: