[Bugfix] Properly propagate trust_remote_code settings (#10047)

Signed-off-by: Zifei Tong <zifeitong@gmail.com>

vllm/model_executor/models/chatglm.py

@@ -54,8 +54,9 @@ def mm_input_mapper_for_glmv(
     data: MultiModalData[object],
 ) -> Dict:
     model_config = ctx.model_config
-    tokenizer = cached_get_tokenizer(model_config.tokenizer,
-                                     trust_remote_code=True)
+    tokenizer = cached_get_tokenizer(
+        model_config.tokenizer,
+        trust_remote_code=model_config.trust_remote_code)
     if tokenizer is None:
         raise RuntimeError("No HuggingFace processor is available "
                            "to process the image object")

vllm/model_executor/models/molmo.py

@@ -844,8 +844,9 @@ def get_max_tokens(max_crops: int, crop_patches: int, left_margin: int,
 
 
 def get_max_molmo_image_tokens(ctx: InputContext) -> int:
-    processor = cached_get_processor(ctx.model_config.model,
-                                     trust_remote_code=True,
+    processor = cached_get_processor(
+        ctx.model_config.model,
+        trust_remote_code=ctx.model_config.trust_remote_code,
         revision=ctx.model_config.code_revision)
     image_processor = processor.image_processor
     max_llm_image_tokens = get_max_tokens(
@@ -870,8 +871,9 @@ def image_input_mapper_for_molmo(
 
 def dummy_data_for_molmo(ctx: InputContext, seq_len: int,
                          mm_counts: Mapping[str, int]):
-    processor = cached_get_processor(ctx.model_config.model,
-                                     trust_remote_code=True,
+    processor = cached_get_processor(
+        ctx.model_config.model,
+        trust_remote_code=ctx.model_config.trust_remote_code,
         revision=ctx.model_config.code_revision)
     image_processor = processor.image_processor
 
@@ -935,11 +937,11 @@ def input_processor_for_molmo(ctx: InputContext, inputs: DecoderOnlyInputs):
     multi_modal_data = inputs.get("multi_modal_data")
     image = None if multi_modal_data is None else multi_modal_data.get("image")
 
-    processor = cached_get_processor(ctx.model_config.model,
-                                     trust_remote_code=True,
-                                     revision=ctx.model_config.code_revision)
-
     model_config = ctx.model_config
+    processor = cached_get_processor(
+        ctx.model_config.model,
+        trust_remote_code=model_config.trust_remote_code,
+        revision=ctx.model_config.code_revision)
     tokenizer = cached_get_tokenizer(
         model_config.tokenizer,
         trust_remote_code=model_config.trust_remote_code)