[Frontend] Use new Renderer for Completions and Tokenize API (#32863)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

tests/models/language/pooling/test_mm_classifier_conversion.py

@@ -96,7 +96,7 @@ def test_gemma_multimodal(
         dtype="bfloat16",
     ) as vllm_model:
         llm = vllm_model.get_llm()
-        prompts = llm.preprocess_chat(messages)
+        prompts = llm._preprocess_chat([messages])
 
         result = llm.classify(prompts)
         assert result[0].outputs.probs[0] > 0.95

tests/models/language/pooling/test_truncation_control.py

@@ -29,7 +29,8 @@ def test_smaller_truncation_size(
         model_name, runner="pooling", max_model_len=max_model_len
     ) as vllm_model:
         vllm_output = vllm_model.llm.embed(
-            input_str, truncate_prompt_tokens=truncate_prompt_tokens
+            input_str,
+            tokenization_kwargs=dict(truncate_prompt_tokens=truncate_prompt_tokens),
         )
 
     prompt_tokens = vllm_output[0].prompt_token_ids
@@ -44,7 +45,8 @@ def test_max_truncation_size(vllm_runner, model_name=MODEL_NAME, input_str=input
         model_name, runner="pooling", max_model_len=max_model_len
     ) as vllm_model:
         vllm_output = vllm_model.llm.embed(
-            input_str, truncate_prompt_tokens=truncate_prompt_tokens
+            input_str,
+            tokenization_kwargs=dict(truncate_prompt_tokens=truncate_prompt_tokens),
         )
 
     prompt_tokens = vllm_output[0].prompt_token_ids
@@ -64,7 +66,8 @@ def test_bigger_truncation_size(
         ) as vllm_model,
     ):
         llm_output = vllm_model.llm.embed(
-            input_str, truncate_prompt_tokens=truncate_prompt_tokens
+            input_str,
+            tokenization_kwargs=dict(truncate_prompt_tokens=truncate_prompt_tokens),
         )
 
         assert (

tests/models/language/pooling_mteb_test/mteb_embed_utils.py

@@ -187,7 +187,10 @@ def mteb_test_embed_models(
         head_dtype = model_config.head_dtype
 
         # Test embedding_size, isnan and whether to use normalize
-        vllm_outputs = vllm_model.embed(example_prompts, truncate_prompt_tokens=-1)
+        vllm_outputs = vllm_model.embed(
+            example_prompts,
+            tokenization_kwargs=dict(truncate_prompt_tokens=-1),
+        )
         outputs_tensor = torch.tensor(vllm_outputs)
         assert not torch.any(torch.isnan(outputs_tensor))
         embedding_size = model_config.embedding_size

tests/models/language/pooling_mteb_test/mteb_score_utils.py

@@ -79,9 +79,9 @@ class VllmMtebCrossEncoder(MtebCrossEncoderMixin):
         outputs = self.llm.score(
             queries,
             corpus,
-            truncate_prompt_tokens=-1,
             use_tqdm=False,
             chat_template=self.chat_template,
+            tokenization_kwargs={"truncate_prompt_tokens": -1},
         )
         scores = np.array(outputs)
         scores = scores[np.argsort(r)]