Merge branch 'rk/automl_nb_fix' of https://github.com/Azure/azureml-assets into rk/automl_nb_fix

Author: iamrk04

assets/training/finetune_acft_hf_nlp/components/pipeline_components/text_generation/spec.yaml

@@ -1,6 +1,6 @@
 $schema: https://azuremlschemas.azureedge.net/latest/pipelineComponent.schema.json
 name: text_generation_pipeline
-version: 0.0.66
+version: 0.0.67
 type: pipeline
 display_name: Text Generation Pipeline
 description: Pipeline component for text generation
@@ -566,7 +566,7 @@ jobs:
       auto_find_batch_size: '${{parent.inputs.auto_find_batch_size}}'
   text_generation_model_import:
     type: command
-    component: azureml:text_generation_model_import:0.0.64
+    component: azureml:text_generation_model_import:0.0.65
     compute: '${{parent.inputs.compute_model_import}}'
     resources:
       instance_type: '${{parent.inputs.instance_type_model_import}}'
@@ -578,7 +578,7 @@ jobs:
       system_properties: '${{parent.inputs.system_properties}}'
   text_generation_datapreprocess:
     type: command
-    component: azureml:text_generation_datapreprocess:0.0.64
+    component: azureml:text_generation_datapreprocess:0.0.65
     compute: '${{parent.inputs.compute_preprocess}}'
     resources:
       instance_type: '${{parent.inputs.instance_type_preprocess}}'
@@ -598,7 +598,7 @@ jobs:
       system_properties: '${{parent.inputs.system_properties}}'
   text_generation_finetune:
     type: command
-    component: azureml:text_generation_finetune:0.0.64
+    component: azureml:text_generation_finetune:0.0.65
     compute: '${{parent.inputs.compute_finetune}}'
     distribution:
       type: pytorch

assets/training/finetune_acft_hf_nlp/src/finetune/finetune.py

@@ -811,6 +811,25 @@ def setup_vllm(task_name: str, finetune_config: Dict[str, Any], base_model_metad
                         base image value is {removed_base_image}.")
     else:
         if base_model_metadata.get("azureml.base_image") is not None:
+            # FT environment for chat-completion task is updated to transformers 4.46.3 and
+            # FTed model tokenizers can't be loaded with transformers<4.45.0
+            # Hence FMI version < 61 can't be used for deployments and Inferencing.
+            # Currently force updating the image in finetune script.
+            # After models are migrated to mcr.microsoft.com/azureml/curated/foundation-model-inference:61
+            # remove the following code marked for removal, including comments
+            # start of removal code
+            if task_name == Tasks.CHAT_COMPLETION:
+                try:
+                    base_vllm_image = str(base_model_metadata.get("azureml.base_image"))
+                    base_vllm_container, base_vllm_image_version = base_vllm_image.split(":")
+                    if int(base_vllm_image_version) < 61:
+                        logger.info("Updating the vllm inference container version.")
+                        base_model_metadata["azureml.base_image"] = base_vllm_container + ":61"
+                except Exception:
+                    logger.info("Unable to fetch vllm inference container version, force updating the image.")
+                    base_model_metadata["azureml.base_image"] = \
+                        "mcr.microsoft.com/azureml/curated/foundation-model-inference:61"
+            # end of removal code
             logger.info(
                 "Adding inferencing base image {} for {} task.\
                 ".format(base_model_metadata.get("azureml.base_image"), task_name)