Restrict build options to only update model hyperparameters

.vscode/launch.json

@@ -18,26 +18,46 @@
       "request": "launch",
       "module": "machine.jobs.build_nmt_engine",
       "justMyCode": false,
-      "args": [
-        "--model-type",
-        "huggingface",
-        "--build-id",
-        "build1",
-        "--engine-id",
-        "engine1",
-        "--src-lang",
-        "spa_Latn",
-        "--trg-lang",
-        "eng_Latn",
-        "--clearml",
-        "--build-options",
-        "{\"max_steps\": 10}"
-      ]
+      "windows": {
+        "args": [
+          "--model-type",
+          "huggingface",
+          "--build-id",
+          "build1",
+          "--engine-id",
+          "engine1",
+          "--src-lang",
+          "spa_Latn",
+          "--trg-lang",
+          "eng_Latn",
+          "--build-options",
+          "{\\\"train_params\\\": {\\\"max_steps\\\": 10}}"
+        ]
+      },
+      "linux": {
+        "args": [
+          "--model-type",
+          "huggingface",
+          "--build-id",
+          "build1",
+          "--engine-id",
+          "engine1",
+          "--src-lang",
+          "spa_Latn",
+          "--trg-lang",
+          "eng_Latn",
+          "--build-options",
+          "{\"train_params\": {\"max_steps\": 10}}"
+        ]
+      }
     },
     {
-      "name": "Debug Unit Test",
+      "name": "Python: Debug Tests",
       "type": "python",
-      "request": "test",
+      "request": "launch",
+      "program": "${file}",
+      "purpose": ["debug-test"],
+      "console": "integratedTerminal",
       "justMyCode": false
     }
   ]

.vscode/settings.json

@@ -1,13 +1,12 @@
 {
   "editor.formatOnSave": true,
   "editor.codeActionsOnSave": {
-    "source.organizeImports": true
+    "source.organizeImports": "explicit"
   },
   "python.testing.unittestEnabled": false,
   "python.testing.pytestEnabled": true,
   "[python]": {
     "editor.defaultFormatter": "ms-python.black-formatter",
     "editor.formatOnSave": true
-  },
-  "python.formatting.provider": "none"
-}
+  }
+}

machine/jobs/build_nmt_engine.py

@@ -45,20 +45,20 @@ def clearml_progress(status: ProgressStatus) -> None:
         logger.info("NMT Engine Build Job started")
 
         SETTINGS.update(args)
-        try:
-            SETTINGS.build_options = json.loads(args["build_options"])
-        except ValueError as e:
-            raise ValueError("Build options could not be parsed: Invalid JSON") from e
-        except TypeError as e:
-            raise TypeError(f"Build options could not be parsed: {e}") from e
-        if SETTINGS.build_options:
-            SETTINGS.update(SETTINGS.build_options)
+        model_type = cast(str, SETTINGS.model_type).lower()
+        if "build_options" in SETTINGS:
+            try:
+                build_options = json.loads(cast(str, SETTINGS.build_options))
+            except ValueError as e:
+                raise ValueError("Build options could not be parsed: Invalid JSON") from e
+            except TypeError as e:
+                raise TypeError(f"Build options could not be parsed: {e}") from e
+            SETTINGS.update({model_type: build_options})
         SETTINGS.data_dir = os.path.expanduser(cast(str, SETTINGS.data_dir))
 
         logger.info(f"Config: {SETTINGS.as_dict()}")
 
         shared_file_service = ClearMLSharedFileService(SETTINGS)
-        model_type = cast(str, SETTINGS.model_type).lower()
         nmt_model_factory: NmtModelFactory
         if model_type == "huggingface":
             from .huggingface.hugging_face_nmt_model_factory import HuggingFaceNmtModelFactory
@@ -87,7 +87,7 @@ def main() -> None:
     parser.add_argument("--src-lang", required=True, type=str, help="Source language tag")
     parser.add_argument("--trg-lang", required=True, type=str, help="Target language tag")
     parser.add_argument("--clearml", default=False, action="store_true", help="Initializes a ClearML task")
-    parser.add_argument("--build-options", default="{}", type=str, help="Build configurations")
+    parser.add_argument("--build-options", default=None, type=str, help="Build configurations")
     args = parser.parse_args()
 
     run({k: v for k, v in vars(args).items() if v is not None})

machine/jobs/huggingface/hugging_face_nmt_model_factory.py

@@ -26,12 +26,13 @@ def __init__(self, config: Any, shared_file_service: SharedFileService) -> None:
         args = config.huggingface.train_params.to_dict()
         args["output_dir"] = str(self._model_dir)
         args["overwrite_output_dir"] = True
-        if "max_steps" in self._config:
-            if self._config.max_steps > 50000:
-                raise ValueError("max_steps must be less than or equal to 50000")
-            args["max_steps"] = self._config.max_steps
+        # Use "max_steps" from root for backward compatibility
+        if "max_steps" in self._config.huggingface:
+            args["max_steps"] = self._config.huggingface.max_steps
         parser = HfArgumentParser(cast(Any, Seq2SeqTrainingArguments))
         self._training_args = cast(Seq2SeqTrainingArguments, parser.parse_dict(args)[0])
+        if self._training_args.max_steps > 50000:
+            raise ValueError("max_steps must be less than or equal to 50000")
         if (
             not config.clearml
             and self._training_args.report_to is not None

machine/jobs/nmt_engine_build_job.py

@@ -86,7 +86,7 @@ def run(
             writer = stack.enter_context(self._shared_file_service.open_target_pretranslation_writer())
             current_inference_step = 0
             phase_progress(ProgressStatus.from_step(current_inference_step, inference_step_count))
-            batch_size = self._config["batch_size"]
+            batch_size = self._config["pretranslation_batch_size"]
             for pi_batch in batch(src_pretranslations, batch_size):
                 if check_canceled is not None:
                     check_canceled()

machine/jobs/settings.yaml

@@ -1,8 +1,7 @@
 default:
   model_type: huggingface
-  max_steps: 20000
   data_dir: ~/machine
-  batch_size: 1024
+  pretranslation_batch_size: 1024
   huggingface:
     parent_model_name: facebook/nllb-200-distilled-1.3B
     train_params:
@@ -16,6 +15,7 @@ default:
       gradient_checkpointing: true
       fp16: true
       save_strategy: no
+      max_steps: 20000
     generate_params:
       device: 0
       num_beams: 2
@@ -31,8 +31,9 @@ development:
     generate_params:
       num_beams: 1
 staging:
-  max_steps: 10
   huggingface:
     parent_model_name: facebook/nllb-200-distilled-600M
+    train_params:
+      max_steps: 10
     generate_params:
-      num_beams: 1
+      num_beams: 1

tests/jobs/test_nmt_engine_build_job.py

@@ -34,7 +34,7 @@ def test_cancel(decoy: Decoy) -> None:
 
 class _TestEnvironment:
     def __init__(self, decoy: Decoy) -> None:
-        config = {"src_lang": "es", "trg_lang": "en", "batch_size": 100}
+        config = {"src_lang": "es", "trg_lang": "en", "pretranslation_batch_size": 100}
         self.source_tokenizer_trainer = decoy.mock(cls=Trainer)
         self.target_tokenizer_trainer = decoy.mock(cls=Trainer)