From e1a77d3710c78949abc6218ee30299392d57d271 Mon Sep 17 00:00:00 2001
From: Nanbo Liu <nliu@lambda002.america.striveworks.us>
Date: Wed, 27 Sep 2023 18:13:47 +0000
Subject: [PATCH] added model_kwargs

---
 runtimes/huggingface/mlserver_huggingface/common.py   | 1 +
 runtimes/huggingface/mlserver_huggingface/settings.py | 5 ++++-
 2 files changed, 5 insertions(+), 1 deletion(-)

diff --git a/runtimes/huggingface/mlserver_huggingface/common.py b/runtimes/huggingface/mlserver_huggingface/common.py
index 53b9d10f2..c32b5fb99 100644
--- a/runtimes/huggingface/mlserver_huggingface/common.py
+++ b/runtimes/huggingface/mlserver_huggingface/common.py
@@ -54,6 +54,7 @@ def load_pipeline_from_settings(
     hf_pipeline = pipeline(
         hf_settings.task_name,
         model=model,
+        model_kwargs=hf_settings.model_kwargs,
         tokenizer=tokenizer,
         device=hf_settings.device,
         batch_size=batch_size,
diff --git a/runtimes/huggingface/mlserver_huggingface/settings.py b/runtimes/huggingface/mlserver_huggingface/settings.py
index b6bc11b5d..b461080f3 100644
--- a/runtimes/huggingface/mlserver_huggingface/settings.py
+++ b/runtimes/huggingface/mlserver_huggingface/settings.py
@@ -61,7 +61,10 @@ class Config:
     """
     Name of the model that should be loaded in the pipeline.
     """
-
+    model_kwargs: Optional[dict] = None
+    """
+    model kwargs that should be loaded in the pipeline.
+    """
     pretrained_tokenizer: Optional[str] = None
     """
     Name of the tokenizer that should be loaded in the pipeline.