Add pipeline as kwargs to hf models

genai_stack/model/hf.py

@@ -1,5 +1,6 @@
 from typing import Optional, Dict
 from langchain.llms import HuggingFacePipeline
+from transformers import pipeline
 
 from genai_stack.model.base import BaseModel, BaseModelConfig, BaseModelConfigModel
 
@@ -17,6 +18,8 @@ class HuggingFaceModelConfigModel(BaseModelConfigModel):
     """Key word arguments passed to the pipeline."""
     task: str = "text-generation"
     """Valid tasks: 'text2text-generation', 'text-generation', 'summarization'"""
+    pipeline: Optional[pipeline] = None
+    """If pipeline is passed, all other configs are ignored."""
 
 
 class HuggingFaceModelConfig(BaseModelConfig):
@@ -30,9 +33,14 @@ def _post_init(self, *args, **kwargs):
         self.model = self.load()
 
     def load(self):
-        model = HuggingFacePipeline.from_model_id(
-            model_id=self.config.model, task=self.config.task, model_kwargs=self.config.model_kwargs
-        )
+        if self.config.pipeline is not None:
+            model = self.config.pipeline
+        else:
+            model = HuggingFacePipeline.from_model_id(
+                model_id=self.config.model,
+                task=self.config.task,
+                model_kwargs=self.config.model_kwargs,
+            )
         return model
 
     def predict(self, prompt: str):