Doc: Document keywords of run.py (#443)

Closes #442

run.py

@@ -57,6 +57,8 @@
 # Apply a patch in a local repository to an issue specified as Markdown file and run a custom installer script in the container
 python run.py --model_name "gpt4" --data_path "/path/to/my_issue.md" --repo_path "/path/to/my/local/repo" --environment_setup "/path/to/setup.sh" --config_file "config/default_from_url.yaml" --apply_patch_locally
 ```
+
+**For more information**: https://princeton-nlp.github.io/SWE-agent/usage/cl_tutorial/
 """
 
 handler = RichHandler(show_time=False, show_path=False)
@@ -96,8 +98,11 @@ class ScriptArguments(FlattenedAccess, FrozenSerializable):
     environment: EnvironmentArguments
     agent: AgentArguments
     actions: ActionsArguments
-    instance_filter: str = ".*"  # Only run instances that completely match this regex
-    skip_existing: bool = True  # Skip instances with existing trajectories
+    # Only run instances that completely match this regex
+    instance_filter: str = ".*"
+    # Skip instances with existing trajectories
+    skip_existing: bool = True
+    # Suffix for the run name (used for example in trajectory directory naming)
     suffix: str = ""
     # Raise unhandled exceptions during the run (useful for debugging)
     raise_exceptions: bool = False

sweagent/agent/models.py

@@ -32,12 +32,19 @@
 class ModelArguments(FrozenSerializable):
     """Arguments configuring the model and its behavior."""
 
+    # Name of the model to use
     model_name: str
+    # Cost limit for every instance (task)
     per_instance_cost_limit: float = 0.0
+    # Total cost limit
     total_cost_limit: float = 0.0
+    # Sampling temperature
     temperature: float = 1.0
+    # Sampling top-p
     top_p: float = 1.0
+    # Path to replay file when using the replay model
     replay_path: str = None
+    # Host URL when using Ollama model
     host_url: str = "localhost:11434"
 
 

sweagent/environment/swe_env.py

@@ -59,16 +59,27 @@ class EnvironmentArguments(FrozenSerializable):
     # (`json`, `jsonl`) or directory. To run over single issue: github issue url or path to markdown file
     # with problem statement or problem statement as text prefixed with `text://`.
     data_path: str
-    image_name: str
+    # Name of the docker image to use for the environment. Defaults to sweagent/swe-agent:latest
+    image_name: str = "sweagent/swe-agent:latest"
+    # When running over SWE-bench issues: Specify the split to use.
     split: str = "dev"
     # Specify a branch name or a commit hash to checkout before running the task.
     # Only used when running over a single problem statement/issue.
     base_commit: str | None = None
+    # Use a persistent container with this name
     container_name: str | None = None
+    # Try to install the environment before running the task.
     install_environment: bool = True
-    timeout: int = 35
+    # No effect, kept for backwards compatibility.
+    timeout: int | None = None
+    # Enable environment logger.
     verbose: bool = False
+    # Do not use attempt to use a repository mirror from https://github.com/swe-bench.
     no_mirror: bool = False
+    # Cache task images to speed up task initialization. This means that the environment will be saved as a
+    # docker image for every repository and base commit combination. This uses quite a bit of disk space
+    # but speeds up task initialization significantly when running over multiple issues from the same repository
+    # (or using different models for the same issues).
     cache_task_images: bool = False
     # Custom environment setup. Currently only used when data_path points to a single issue.
     # This needs to be either a string pointing to a yaml file (with yaml, yml file extension)
@@ -78,6 +89,10 @@ class EnvironmentArguments(FrozenSerializable):
     # Only used when running on single issue. Path to local repository or github repository.
     repo_path: str = ""
 
+    def __post_init__(self):
+        if self.timeout is not None:
+            logger.warning("The 'timeout' argument is deprecated and has no effect.")
+
 
 class EnvHook:
     def on_init(self): ...
@@ -137,8 +152,6 @@ def __init__(self, args: EnvironmentArguments):
         self.image_name = args.image_name
         self._reset_container()
 
-        # Set timeout
-        self.timeout = self.args.timeout
         self.idx = 0
         self.clean_multi_line_functions = lambda x: x
         self.hooks = []