diff --git a/tests/entrypoints/llm/test_encode.py b/tests/entrypoints/llm/test_encode.py
index 41163809237e9..3906ad766e0b6 100644
--- a/tests/entrypoints/llm/test_encode.py
+++ b/tests/entrypoints/llm/test_encode.py
@@ -105,3 +105,10 @@ def test_multiple_pooling_params(llm: LLM):
     # pooling_params is None, default params should be applied
     outputs = llm.encode(PROMPTS, pooling_params=None)
     assert len(PROMPTS) == len(outputs)
+
+
+@pytest.mark.skip_global_cleanup
+def test_right_side_truncation(llm: LLM):
+    # Embeddings models should truncate the end of the prompt
+    tokenizer = llm.get_tokenizer()
+    assert tokenizer.truncation_side == "right"
diff --git a/vllm/config.py b/vllm/config.py
index 65cb0d85f172a..4ffc13a05e026 100644
--- a/vllm/config.py
+++ b/vllm/config.py
@@ -357,6 +357,10 @@ def __init__(self,
         supported_tasks, task = self._resolve_task(task, self.hf_config)
         self.supported_tasks = supported_tasks
         self.task: Final = task
+        if self.task in ("draft", "generate"):
+            self.truncation_side = "left"
+        else:
+            self.truncation_side = "right"
 
         self.pooler_config = self._init_pooler_config(override_pooler_config)
         self.logits_processor_pattern = logits_processor_pattern
diff --git a/vllm/transformers_utils/tokenizer_group/__init__.py b/vllm/transformers_utils/tokenizer_group/__init__.py
index d400276796996..09569c564a58d 100644
--- a/vllm/transformers_utils/tokenizer_group/__init__.py
+++ b/vllm/transformers_utils/tokenizer_group/__init__.py
@@ -24,7 +24,8 @@ def init_tokenizer_from_configs(model_config: ModelConfig,
                        max_input_length=None,
                        tokenizer_mode=model_config.tokenizer_mode,
                        trust_remote_code=model_config.trust_remote_code,
-                       revision=model_config.tokenizer_revision)
+                       revision=model_config.tokenizer_revision,
+                       truncation_side=model_config.truncation_side)
 
     return get_tokenizer_group(parallel_config.tokenizer_pool_config,
                                **init_kwargs)