Tmp refactor train func for ray

spacy/cli/train.py

@@ -156,9 +156,28 @@ def train_cli(
         with init_tok2vec.open("rb") as file_:
             weights_data = file_.read()
 
+    if num_workers and num_workers > 1:
+        from _ray_async_utils import distributed_setup_and_train
+        distributed_setup_and_train(
+            use_gpu,
+            num_workers,
+            strategy,
+            ray_address,
+            {
+                "config": config_path,
+                "train": train_path,
+                "dev": dev_path,
+                "output": output_path
+            }
+        )
+    else:
+        nlp, config = load_nlp_and_config(config_path)
+        corpus = Corpus(train_path, dev_path, limit=config["training"]["limit"])
+
         train_args = dict(
-        config_path=config_path,
+            nlp=nlp,
-        data_paths={"train": train_path, "dev": dev_path},
+            config=config,
+            corpus=corpus,
             output_path=output_path,
             raw_text=raw_text,
             tag_map=tag_map,
@@ -166,10 +185,6 @@ def train_cli(
             omit_extra_lookups=omit_extra_lookups
         )
 
-    if num_workers and num_workers > 1:
-        from spacy_ray import distributed_setup_and_train
-        distributed_setup_and_train(use_gpu, num_workers, strategy, ray_address, train_args)
-    else:
         if use_gpu >= 0:
             msg.info(f"Using GPU: {use_gpu}")
             require_gpu(use_gpu)
@@ -177,20 +192,8 @@ def train_cli(
             msg.info("Using CPU")
         train(**train_args)
 
-def train(
+
-    config_path: Path,
+def load_nlp_and_config(config_path):
-    data_paths: Dict[str, Path],
-    raw_text: Optional[Path] = None,
-    output_path: Optional[Path] = None,
-    tag_map: Optional[Path] = None,
-    weights_data: Optional[bytes] = None,
-    omit_extra_lookups: bool = False,
-    disable_tqdm: bool = False,
-    remote_optimizer = None,
-    randomization_index: int = 0
-) -> None:
-    msg.info(f"Loading config from: {config_path}")
-    # Read the config first without creating objects, to get to the original nlp_config
     config = util.load_config(config_path, create_objects=False)
     if config["training"].get("seed"):
         fix_random_seed(config["training"]["seed"])
@@ -199,16 +202,28 @@ def train(
         use_pytorch_for_gpu_memory()
     nlp_config = config["nlp"]
     config = util.load_config(config_path, create_objects=True)
+    nlp = util.load_model_from_config(nlp_config)
+    return nlp, config
+
+
+def train(
+    nlp,
+    config,
+    corpus,
+    raw_text: Optional[Path] = None,
+    output_path: Optional[Path] = None,
+    tag_map: Optional[Path] = None,
+    weights_data: Optional[bytes] = None,
+    omit_extra_lookups: bool = False,
+    disable_tqdm: bool = False,
+    randomization_index: int = 0,
+    num_workers=1
+) -> None:
+    msg.info(f"Loading config from: {config_path}")
+    # Read the config first without creating objects, to get to the original nlp_config
     training = config["training"]
     msg.info("Creating nlp from config")
-    nlp = util.load_model_from_config(nlp_config)
     optimizer = training["optimizer"]
-    # TODO: is there a cleaner way of doing this, instead of creating
-    # the optimizer twice? are there any problems when doing this?
-    if remote_optimizer:
-        optimizer = remote_optimizer
-    limit = training["limit"]
-    corpus = Corpus(data_paths["train"], data_paths["dev"], limit=limit)
     if "textcat" in nlp_config["pipeline"]:
         verify_textcat_config(nlp, nlp_config)
     if training.get("resume", False):
@@ -275,7 +290,6 @@ def train(
     )
     msg.info(f"Training. Initial learn rate: {optimizer.learn_rate}")
     print_row = setup_printer(training, nlp)
-
     tqdm_args = dict(total=training["eval_frequency"], leave=False, disable=disable_tqdm)
     try:
         progress = tqdm.tqdm(**tqdm_args)