Add renderer-based prompt processing for embedding and classification endpoints (#24356)

Signed-off-by: sfeng33 <4florafeng@gmail.com>

tests/entrypoints/openai/test_truncation.py

@@ -73,17 +73,11 @@ async def test_zero_truncation_size(client: openai.AsyncOpenAI):
         "truncate_prompt_tokens": truncation_size
     }
 
-    with pytest.raises(openai.BadRequestError) as err:
-        await client.post(path="embeddings", cast_to=object, body={**kwargs})
+    response = await client.post(path="embeddings",
+                                 cast_to=object,
+                                 body={**kwargs})
 
-    assert err.value.status_code == 400
-    error_details = err.value.response.json()["error"]
-
-    assert error_details["type"] == "BadRequestError"
-    assert "This model's maximum context length is" in error_details["message"]
-    assert "tokens in the input for embedding generation" in error_details[
-        "message"]
-    assert "Please reduce the length of the input" in error_details["message"]
+    assert response["usage"]["prompt_tokens"] == truncation_size
 
 
 @pytest.mark.asyncio

tests/entrypoints/test_renderer.py

@@ -130,6 +130,23 @@ class TestRenderPrompt:
         assert call_args.kwargs["truncation"] is True
         assert call_args.kwargs["max_length"] == 50
 
+    @pytest.mark.asyncio
+    async def test_truncation_negative(self, renderer, mock_async_tokenizer):
+        # Test that negative truncation uses model's max_model_len
+        mock_async_tokenizer.return_value = MockTokenizerResult(
+            [101, 7592, 2088])  # Truncated to max_model_len
+        renderer.async_tokenizer_pool[
+            renderer.tokenizer] = mock_async_tokenizer
+
+        results = await renderer.render_prompt(prompt_or_prompts="Hello world",
+                                               max_length=200,
+                                               truncate_prompt_tokens=-1)
+
+        assert len(results) == 1
+        call_args = mock_async_tokenizer.call_args
+        assert call_args.kwargs["truncation"] is True
+        assert call_args.kwargs["max_length"] == 100  # model's max_model_len
+
     @pytest.mark.asyncio
     async def test_token_truncation_last_elements(self, renderer):
         # Test that token truncation keeps the last N elements