From 3a5082f28a1cb6ef0c294e8062fbfb49d158b57f Mon Sep 17 00:00:00 2001
From: Uranus <109661872+UranusSeven@users.noreply.github.com>
Date: Wed, 27 Sep 2023 04:28:07 -0700
Subject: [PATCH] ENH: set `trust_remote_code` to true (#500)

---
 xinference/deploy/cmdline.py             | 2 +-
 xinference/model/llm/pytorch/baichuan.py | 3 +--
 xinference/model/llm/pytorch/chatglm.py  | 3 +--
 xinference/model/llm/pytorch/core.py     | 2 +-
 xinference/model/llm/pytorch/falcon.py   | 2 +-
 xinference/model/llm/vllm/core.py        | 2 +-
 6 files changed, 6 insertions(+), 8 deletions(-)

diff --git a/xinference/deploy/cmdline.py b/xinference/deploy/cmdline.py
index ab59884e02..9a5f24d276 100644
--- a/xinference/deploy/cmdline.py
+++ b/xinference/deploy/cmdline.py
@@ -385,7 +385,7 @@ def list_model_registrations(
 )
 @click.option(
     "--trust-remote-code",
-    default=False,
+    default=True,
     type=bool,
     help="Whether or not to allow for custom models defined on the Hub in their own modeling files.",
 )
diff --git a/xinference/model/llm/pytorch/baichuan.py b/xinference/model/llm/pytorch/baichuan.py
index f0e5672d4e..a08234216d 100644
--- a/xinference/model/llm/pytorch/baichuan.py
+++ b/xinference/model/llm/pytorch/baichuan.py
@@ -54,12 +54,11 @@ def _load_model(self, kwargs: dict):
         tokenizer = AutoTokenizer.from_pretrained(
             self.model_path,
             use_fast=self._use_fast_tokenizer,
-            trust_remote_code=True,
+            trust_remote_code=kwargs["trust_remote_code"],
             revision=kwargs["revision"],
         )
         model = AutoModelForCausalLM.from_pretrained(
             self.model_path,
-            trust_remote_code=True,
             **kwargs,
         )
         model.generation_config = GenerationConfig.from_pretrained(self.model_path)
diff --git a/xinference/model/llm/pytorch/chatglm.py b/xinference/model/llm/pytorch/chatglm.py
index 1754666fa7..b78407e9d7 100644
--- a/xinference/model/llm/pytorch/chatglm.py
+++ b/xinference/model/llm/pytorch/chatglm.py
@@ -51,12 +51,11 @@ def _load_model(self, kwargs: dict):
 
         tokenizer = AutoTokenizer.from_pretrained(
             self.model_path,
-            trust_remote_code=True,
+            trust_remote_code=kwargs["trust_remote_code"],
             revision=kwargs["revision"],
         )
         model = AutoModel.from_pretrained(
             self.model_path,
-            trust_remote_code=True,
             **kwargs,
         )
         return model, tokenizer
diff --git a/xinference/model/llm/pytorch/core.py b/xinference/model/llm/pytorch/core.py
index 4b997b6302..99af2afb66 100644
--- a/xinference/model/llm/pytorch/core.py
+++ b/xinference/model/llm/pytorch/core.py
@@ -87,7 +87,7 @@ def _sanitize_model_config(
         pytorch_model_config.setdefault("gptq_groupsize", -1)
         pytorch_model_config.setdefault("gptq_act_order", False)
         pytorch_model_config.setdefault("device", "auto")
-        pytorch_model_config.setdefault("trust_remote_code", False)
+        pytorch_model_config.setdefault("trust_remote_code", True)
         return pytorch_model_config
 
     def _sanitize_generate_config(
diff --git a/xinference/model/llm/pytorch/falcon.py b/xinference/model/llm/pytorch/falcon.py
index d245476a32..773fea83d7 100644
--- a/xinference/model/llm/pytorch/falcon.py
+++ b/xinference/model/llm/pytorch/falcon.py
@@ -51,12 +51,12 @@ def _load_model(self, kwargs: dict):
 
         tokenizer = AutoTokenizer.from_pretrained(
             self.model_path,
+            trust_remote_code=kwargs["trust_remote_code"],
             revision=kwargs["revision"],
         )
         model = AutoModelForCausalLM.from_pretrained(
             self.model_path,
             low_cpu_mem_usage=True,
-            trust_remote_code=True,
             **kwargs,
         )
         tokenizer.pad_token_id = 9
diff --git a/xinference/model/llm/vllm/core.py b/xinference/model/llm/vllm/core.py
index 70d98e508b..772e043f2b 100644
--- a/xinference/model/llm/vllm/core.py
+++ b/xinference/model/llm/vllm/core.py
@@ -116,7 +116,7 @@ def _sanitize_model_config(
         cuda_count = self._get_cuda_count()
 
         model_config.setdefault("tokenizer_mode", "auto")
-        model_config.setdefault("trust_remote_code", False)
+        model_config.setdefault("trust_remote_code", True)
         model_config.setdefault("tensor_parallel_size", cuda_count)
         model_config.setdefault("block_size", 16)
         model_config.setdefault("swap_space", 4)