lightray 0.2.0__tar.gz → 0.2.2__tar.gz

{lightray-0.2.0 → lightray-0.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: lightray
-Version: 0.2.0
+Version: 0.2.2
 Summary: Distribute a LightningCLI hyperparameter search with Ray Tune
 License: MIT
 Author: Ethan Jacob Marx
@@ -114,17 +114,19 @@ tune_callback:
 cpus_per_trial: 2
 gpus_per_trial: 1
 
-# lightning cli
+# lightning cli class to tune
 lightning_cli_cls: example.lightning.Cli
-lightning_config: /home/ethan.marx/projects/lightray/example/cli.yaml
 ```
 
-Then, launch the tuning job
+Then, launch the tuning job.
 
 ```console
-lightray --config tune.yaml 
+lightray --config tune.yaml -- --config cli.yaml
 ```
 
+Any arguments passed after "--" will automatically be forwarded to the specified `LightningCLI` class as if it were being run from the command line. This way, tuning configuration and training configuration can be kept separate. Also, you can easily utilize existing configuration files compatible with your `LightningCLI`.
+
+
 ## S3 Support
 In addition, there is automatic support for `s3` storage. Make sure you have set the `AWS_ENDPOINT_URL`, `AWS_ACCESS_KEY_ID`, and `AWS_SECRET_ACCESS_KEY` environment variables. 
 Then, simply pass the path to your bucket with the `s3` prefix (e.g. `s3://{bucket}/{folder}`) to the `run_config.storage_path` argument.
@@ -138,9 +140,9 @@ This is provided at `lightray.callbacks.LightRayReportCheckpointCallback`
 
 
 ## Remote cluster
-To connect to a remote ray cluster, pass the ip address and (using port `10001`) to `ray_init.address`:
+To connect to a remote ray cluster, pass the ip address (using port `10001`) to `ray_init.address`:
 
 ```console
-lightray --config tune.yaml --ray_init.address ray://{ip}:10001
+lightray --config tune.yaml --ray_init.address ray://{ip}:10001 -- --config cli.yaml
 ```
 

{lightray-0.2.0 → lightray-0.2.2}/README.md

@@ -91,17 +91,19 @@ tune_callback:
 cpus_per_trial: 2
 gpus_per_trial: 1
 
-# lightning cli
+# lightning cli class to tune
 lightning_cli_cls: example.lightning.Cli
-lightning_config: /home/ethan.marx/projects/lightray/example/cli.yaml
 ```
 
-Then, launch the tuning job
+Then, launch the tuning job.
 
 ```console
-lightray --config tune.yaml 
+lightray --config tune.yaml -- --config cli.yaml
 ```
 
+Any arguments passed after "--" will automatically be forwarded to the specified `LightningCLI` class as if it were being run from the command line. This way, tuning configuration and training configuration can be kept separate. Also, you can easily utilize existing configuration files compatible with your `LightningCLI`.
+
+
 ## S3 Support
 In addition, there is automatic support for `s3` storage. Make sure you have set the `AWS_ENDPOINT_URL`, `AWS_ACCESS_KEY_ID`, and `AWS_SECRET_ACCESS_KEY` environment variables. 
 Then, simply pass the path to your bucket with the `s3` prefix (e.g. `s3://{bucket}/{folder}`) to the `run_config.storage_path` argument.
@@ -115,8 +117,8 @@ This is provided at `lightray.callbacks.LightRayReportCheckpointCallback`
 
 
 ## Remote cluster
-To connect to a remote ray cluster, pass the ip address and (using port `10001`) to `ray_init.address`:
+To connect to a remote ray cluster, pass the ip address (using port `10001`) to `ray_init.address`:
 
 ```console
-lightray --config tune.yaml --ray_init.address ray://{ip}:10001
+lightray --config tune.yaml --ray_init.address ray://{ip}:10001 -- --config cli.yaml
 ```

{lightray-0.2.0 → lightray-0.2.2}/lightray/cli.py

@@ -1,6 +1,5 @@
 """
-A large portion of this was inspired by
-https://github.com/mauvilsa/ray-tune-cli/
+Heavily inspired by https://github.com/mauvilsa/ray-tune-cli/
 """
 
 import logging
@@ -10,7 +9,12 @@ from typing import Optional, Type
 
 import pyarrow
 import ray
-from jsonargparse import ActionConfigFile, ArgumentParser, capture_parser
+from jsonargparse import (
+    REMAINDER,
+    ActionConfigFile,
+    ArgumentParser,
+    capture_parser,
+)
 from lightning.pytorch.cli import LightningCLI
 from ray import train, tune
 from ray.tune.integration.pytorch_lightning import TuneCallback
@@ -18,7 +22,7 @@ from ray.tune.integration.pytorch_lightning import TuneCallback
 from lightray import fs, utils
 
 
-def cli():
+def cli(args=None):
     parser = ArgumentParser()
     parser.add_argument("--config", action=ActionConfigFile)
     parser.add_subclass_arguments(TuneCallback, "tune_callback")
@@ -31,7 +35,7 @@ def cli():
     parser.add_class_arguments(train.SyncConfig, "sync_config")
     parser.add_function_arguments(ray.init, "ray_init")
     parser.add_argument(
-        "lightning_cli_cls",
+        "--lightning_cli_cls",
         type=Type[LightningCLI],
         help="Lightning CLI class",
     )
@@ -43,11 +47,6 @@ def cli():
     )
     parser.add_argument("--gpus_per_trial", type=int, default=0)
     parser.add_argument("--cpus_per_trial", type=int, default=1)
-    parser.add_argument(
-        "lightning_config",
-        type=str,
-        help="Path to the lightning config file",
-    )
 
     parser.link_arguments(
         "run_config.checkpoint_config.checkpoint_score_attribute",
@@ -60,7 +59,14 @@ def cli():
         apply_on="parse",
     )
 
-    cfg = parser.parse_args()
+    parser.add_argument(
+        "lightning_args",
+        nargs=REMAINDER,
+        help='All arguments after the double dash "--"'
+        "are forwarded to the LightningCLI-based function",
+    )
+
+    cfg = parser.parse_args(args)
 
     log_format = "%(asctime)s - %(name)s - %(levelname)s - %(message)s"
     logging.basicConfig(
@@ -69,7 +75,8 @@ def cli():
         stream=sys.stdout,
     )
 
-    # get lightning cli parser and parse passed config
+    # get lightning cli parser and parse any arguments
+    # passed after "--" at the command line
     lightning_cli_cls = cfg.lightning_cli_cls
     lightning_parser = capture_parser(lightning_cli_cls)
     if lightning_parser._subcommands_action:
@@ -77,9 +84,10 @@ def cli():
             lightning_parser._subcommands_action._name_parser_map["fit"]
         )
 
-    lightning_cfg = lightning_parser.parse_args(
-        ["--config", cfg.lightning_config]
-    )
+    if len(cfg.lightning_args) > 1:
+        lightning_cfg = lightning_parser.parse_args(cfg.lightning_args[1:])
+    else:
+        lightning_cfg = lightning_parser.get_defaults()
 
     callbacks = [cfg.tune_callback]
     callbacks += lightning_cfg.get("trainer.callbacks") or []

{lightray-0.2.0 → lightray-0.2.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "lightray"
-version = "0.2.0"
+version = "0.2.2"
 description = "Distribute a LightningCLI hyperparameter search with Ray Tune"
 authors = ["Ethan Jacob Marx <ethan.marx@ligo.org>"]
 license = "MIT"