test: add more unit tests for sequence_length coverage

- _parse_context_length with invalid K value and non-numeric string
- sequence_length filter when no recipes have SequenceLength field
- sequence_length filter with FULL training type
- verify smallest sufficient sequence_length is selected
- verify no sequence_length uses first recipe (backward compat)

Author: guanweim

sagemaker-train/tests/unit/train/common_utils/test_finetune_utils.py

@@ -864,3 +864,274 @@ def test__get_fine_tuning_options_subscription_enabled_but_not_subscribed(self,
         # Should still have standard params, just not datamix ones
         assert "max_steps" in options._specs
         assert "customer_data_percent" not in options._specs
+
+    def test__create_serverless_config_with_sequence_length(self):
+        config = _create_serverless_config("model-arn", "SFT", TrainingType.LORA, accept_eula=True, sequence_length="8K")
+
+        assert config.sequence_length == "8K"
+        assert config.base_model_arn == "model-arn"
+
+    def test__create_serverless_config_without_sequence_length(self):
+        config = _create_serverless_config("model-arn", "SFT", TrainingType.LORA, accept_eula=True)
+
+        assert config.sequence_length is None
+
+    def test__parse_context_length_with_k_suffix(self):
+        assert _parse_context_length("8K") == 8192
+        assert _parse_context_length("32K") == 32768
+        assert _parse_context_length("128K") == 131072
+
+    def test__parse_context_length_with_lowercase(self):
+        assert _parse_context_length("8k") == 8192
+
+    def test__parse_context_length_with_integer(self):
+        assert _parse_context_length("4096") == 4096
+
+    def test__parse_context_length_with_none(self):
+        assert _parse_context_length(None) == 0
+
+    def test__parse_context_length_with_empty(self):
+        assert _parse_context_length("") == 0
+
+    @patch('sagemaker.train.common_utils.finetune_utils._get_hub_content_metadata')
+    def test__get_fine_tuning_options_filters_by_sequence_length(self, mock_get_hub_content):
+        mock_session = Mock()
+        mock_session.boto_session.region_name = "us-east-1"
+        mock_s3 = Mock()
+        mock_s3.get_object.return_value = {
+            "Body": Mock(read=Mock(return_value=b'{"max_length": {"default": 32768}}'))
+        }
+        mock_session.boto_session.client.return_value = mock_s3
+
+        mock_get_hub_content.return_value = {
+            'hub_content_arn': "arn:aws:sagemaker:us-east-1:123456789012:model/test-model",
+            'hub_content_document': {
+                "GatedBucket": False,
+                "RecipeCollection": [
+                    {
+                        "CustomizationTechnique": "SFT",
+                        "SmtjRecipeTemplateS3Uri": "s3://bucket/template-4k.json",
+                        "SmtjOverrideParamsS3Uri": "s3://bucket/params-4k.json",
+                        "Peft": True,
+                        "SequenceLength": "4K"
+                    },
+                    {
+                        "CustomizationTechnique": "SFT",
+                        "SmtjRecipeTemplateS3Uri": "s3://bucket/template-32k.json",
+                        "SmtjOverrideParamsS3Uri": "s3://bucket/params-32k.json",
+                        "Peft": True,
+                        "SequenceLength": "32K"
+                    }
+                ]
+            }
+        }
+
+        result = _get_fine_tuning_options_and_model_arn("test-model", "SFT", "LORA", mock_session, sequence_length="8K")
+
+        if result is not None:
+            options, model_arn, is_gated_model = result
+            # Should pick the 32K recipe (smallest >= 8K)
+            mock_s3.get_object.assert_called_once()
+            call_args = mock_s3.get_object.call_args[1]
+            assert "params-32k" in call_args["Key"]
+
+    @patch('sagemaker.train.common_utils.finetune_utils._get_hub_content_metadata')
+    def test__get_fine_tuning_options_raises_when_no_sufficient_context_length(self, mock_get_hub_content):
+        mock_session = Mock()
+        mock_session.boto_session.region_name = "us-east-1"
+
+        mock_get_hub_content.return_value = {
+            'hub_content_arn': "arn:aws:sagemaker:us-east-1:123456789012:model/test-model",
+            'hub_content_document': {
+                "GatedBucket": False,
+                "RecipeCollection": [
+                    {
+                        "CustomizationTechnique": "SFT",
+                        "SmtjRecipeTemplateS3Uri": "s3://bucket/template-4k.json",
+                        "SmtjOverrideParamsS3Uri": "s3://bucket/params-4k.json",
+                        "Peft": True,
+                        "SequenceLength": "4K"
+                    }
+                ]
+            }
+        }
+
+        # Requesting 128K but only 4K available — should raise
+        with pytest.raises(ValueError, match="No recipes found with SequenceLength >= 128K"):
+            _get_fine_tuning_options_and_model_arn(
+                "test-model", "SFT", "LORA", mock_session, sequence_length="128K"
+            )
+
+    def test__parse_context_length_with_invalid_k_value(self):
+        assert _parse_context_length("abcK") == 0
+
+    def test__parse_context_length_with_non_numeric_string(self):
+        assert _parse_context_length("hello") == 0
+
+    @patch("sagemaker.train.common_utils.finetune_utils._get_hub_content_metadata")
+    def test__get_fine_tuning_options_raises_when_no_recipes_have_sequence_length(
+        self, mock_get_hub_content
+    ):
+        mock_session = Mock()
+        mock_session.boto_session.region_name = "us-east-1"
+
+        mock_get_hub_content.return_value = {
+            "hub_content_arn": "arn:aws:sagemaker:us-east-1:123456789012:model/test-model",
+            "hub_content_document": {
+                "GatedBucket": False,
+                "RecipeCollection": [
+                    {
+                        "CustomizationTechnique": "SFT",
+                        "SmtjRecipeTemplateS3Uri": "s3://bucket/template.json",
+                        "SmtjOverrideParamsS3Uri": "s3://bucket/params.json",
+                        "Peft": True,
+                    }
+                ],
+            },
+        }
+
+        with pytest.raises(ValueError, match="and sequence length"):
+            _get_fine_tuning_options_and_model_arn(
+                "test-model", "SFT", "LORA", mock_session, sequence_length="8K"
+            )
+
+    @patch("sagemaker.train.common_utils.finetune_utils._get_hub_content_metadata")
+    def test__get_fine_tuning_options_filters_by_sequence_length_full_training(
+        self, mock_get_hub_content
+    ):
+        mock_session = Mock()
+        mock_session.boto_session.region_name = "us-east-1"
+        mock_s3 = Mock()
+        mock_s3.get_object.return_value = {
+            "Body": Mock(read=Mock(return_value=b'{"max_length": {"default": 8192}}'))
+        }
+        mock_session.boto_session.client.return_value = mock_s3
+
+        mock_get_hub_content.return_value = {
+            "hub_content_arn": "arn:aws:sagemaker:us-east-1:123456789012:model/test-model",
+            "hub_content_document": {
+                "GatedBucket": False,
+                "RecipeCollection": [
+                    {
+                        "CustomizationTechnique": "SFT",
+                        "SmtjRecipeTemplateS3Uri": "s3://bucket/template-8k.json",
+                        "SmtjOverrideParamsS3Uri": "s3://bucket/params-8k.json",
+                        "SequenceLength": "8K",
+                    },
+                    {
+                        "CustomizationTechnique": "SFT",
+                        "SmtjRecipeTemplateS3Uri": "s3://bucket/template-32k.json",
+                        "SmtjOverrideParamsS3Uri": "s3://bucket/params-32k.json",
+                        "SequenceLength": "32K",
+                    },
+                ],
+            },
+        }
+
+        result = _get_fine_tuning_options_and_model_arn(
+            "test-model", "SFT", "FULL", mock_session, sequence_length="8K"
+        )
+
+        if result is not None:
+            options, model_arn, is_gated_model = result
+            mock_s3.get_object.assert_called_once()
+            call_args = mock_s3.get_object.call_args[1]
+            assert "params-8k" in call_args["Key"]
+
+    @patch("sagemaker.train.common_utils.finetune_utils._get_hub_content_metadata")
+    def test__get_fine_tuning_options_selects_smallest_sufficient_sequence_length(
+        self, mock_get_hub_content
+    ):
+        mock_session = Mock()
+        mock_session.boto_session.region_name = "us-east-1"
+        mock_s3 = Mock()
+        mock_s3.get_object.return_value = {
+            "Body": Mock(read=Mock(return_value=b'{"max_length": {"default": 16384}}'))
+        }
+        mock_session.boto_session.client.return_value = mock_s3
+
+        mock_get_hub_content.return_value = {
+            "hub_content_arn": "arn:aws:sagemaker:us-east-1:123456789012:model/test-model",
+            "hub_content_document": {
+                "GatedBucket": False,
+                "RecipeCollection": [
+                    {
+                        "CustomizationTechnique": "SFT",
+                        "SmtjRecipeTemplateS3Uri": "s3://bucket/template-4k.json",
+                        "SmtjOverrideParamsS3Uri": "s3://bucket/params-4k.json",
+                        "Peft": True,
+                        "SequenceLength": "4K",
+                    },
+                    {
+                        "CustomizationTechnique": "SFT",
+                        "SmtjRecipeTemplateS3Uri": "s3://bucket/template-16k.json",
+                        "SmtjOverrideParamsS3Uri": "s3://bucket/params-16k.json",
+                        "Peft": True,
+                        "SequenceLength": "16K",
+                    },
+                    {
+                        "CustomizationTechnique": "SFT",
+                        "SmtjRecipeTemplateS3Uri": "s3://bucket/template-128k.json",
+                        "SmtjOverrideParamsS3Uri": "s3://bucket/params-128k.json",
+                        "Peft": True,
+                        "SequenceLength": "128K",
+                    },
+                ],
+            },
+        }
+
+        result = _get_fine_tuning_options_and_model_arn(
+            "test-model", "SFT", "LORA", mock_session, sequence_length="8K"
+        )
+
+        if result is not None:
+            options, model_arn, is_gated_model = result
+            # Should pick 16K (smallest >= 8K), not 128K
+            mock_s3.get_object.assert_called_once()
+            call_args = mock_s3.get_object.call_args[1]
+            assert "params-16k" in call_args["Key"]
+
+    @patch("sagemaker.train.common_utils.finetune_utils._get_hub_content_metadata")
+    def test__get_fine_tuning_options_without_sequence_length_uses_first_recipe(
+        self, mock_get_hub_content
+    ):
+        """Verify that when no sequence_length is provided, existing behavior is unchanged."""
+        mock_session = Mock()
+        mock_session.boto_session.region_name = "us-east-1"
+        mock_s3 = Mock()
+        mock_s3.get_object.return_value = {
+            "Body": Mock(read=Mock(return_value=b'{"max_length": {"default": 4096}}'))
+        }
+        mock_session.boto_session.client.return_value = mock_s3
+
+        mock_get_hub_content.return_value = {
+            "hub_content_arn": "arn:aws:sagemaker:us-east-1:123456789012:model/test-model",
+            "hub_content_document": {
+                "GatedBucket": False,
+                "RecipeCollection": [
+                    {
+                        "CustomizationTechnique": "SFT",
+                        "SmtjRecipeTemplateS3Uri": "s3://bucket/template-first.json",
+                        "SmtjOverrideParamsS3Uri": "s3://bucket/params-first.json",
+                        "Peft": True,
+                        "SequenceLength": "4K",
+                    },
+                    {
+                        "CustomizationTechnique": "SFT",
+                        "SmtjRecipeTemplateS3Uri": "s3://bucket/template-second.json",
+                        "SmtjOverrideParamsS3Uri": "s3://bucket/params-second.json",
+                        "Peft": True,
+                        "SequenceLength": "32K",
+                    },
+                ],
+            },
+        }
+
+        result = _get_fine_tuning_options_and_model_arn("test-model", "SFT", "LORA", mock_session)
+
+        if result is not None:
+            options, model_arn, is_gated_model = result
+            # Without sequence_length, should pick the first matching recipe
+            mock_s3.get_object.assert_called_once()
+            call_args = mock_s3.get_object.call_args[1]
+            assert "params-first" in call_args["Key"]