revert benchmark fix (#8747)

PaddlePaddle · Jul 10, 2024 · a1dbc39 · a1dbc39
1 parent 3241120
commit a1dbc39
Showing 1 changed file with 8 additions and 15 deletions.
diff --git a/llm/predict/predictor.py b/llm/predict/predictor.py
@@ -849,8 +849,6 @@ def __init__(self, config: PredictorArgument, tokenizer: PretrainedTokenizer):
         self.free_list = [i for i in range(self.max_block_nums)][::-1]
         self.used_list = [[] for _ in range(config.batch_size)]
 
-        self.benchmark = config.benchmark
-
     def init_inputs(self, config: PredictorArgument):
         self.inputs = {}
 
@@ -967,22 +965,19 @@ def _get_rotary_position_embedding(self, position_ids, head_dim, rope_theta=1000
         return rot_emb
 
     def _preprocess(self, source):
-        if not self.benchmark and self.tokenizer.chat_template is not None:
+        if self.tokenizer.chat_template is not None:
             source = [source] if isinstance(source, str) else source
             source = [self.tokenizer.apply_chat_template(sentence, tokenize=False) for sentence in source]
 
         for i, text in enumerate(source):
-            add_special_tokens = self.tokenizer.chat_template is None or isinstance(
-                self.tokenizer, (ChatGLMv2Tokenizer, ChatGLMTokenizer)
-            )
-            add_special_tokens = add_special_tokens if not self.benchmark else False
             tokens = self.tokenizer(
                 text,
                 return_tensors="np",
                 padding=True,
                 max_length=self.config.src_length,
                 # if use chat_template, it will not add special_tokens
-                add_special_tokens=add_special_tokens,
+                add_special_tokens=self.tokenizer.chat_template is None
+                or isinstance(self.tokenizer, (ChatGLMv2Tokenizer, ChatGLMTokenizer)),
             )
             input_ids = tokens["input_ids"][0]
             length = len(input_ids)
@@ -1622,10 +1617,6 @@ def predict():
 
     predictor = create_predictor(predictor_args, model_args)
 
-    if predictor_args.benchmark:
-        benchmark(predictor, predictor_args, model_args)
-        return
-
     source_texts = []
     target_texts = []
     if model_args.data_file:
@@ -1669,12 +1660,14 @@ def predict():
                 out = {"src": source, "tgt": target, "output": output}
                 f.write(json.dumps(out, ensure_ascii=False) + "\n")
 
+    if predictor_args.benchmark:
+        benchmark(predictor, predictor_args, model_args)
+
 
 def benchmark(predictor, predictor_args, model_args):
     # Just construct a simple benchmark input. We pad input to the src_length.
-    benchmark_texts = [
-        predictor.tokenizer.pad_token * predictor_args.src_length for _ in range(predictor_args.batch_size)
-    ]
+    test_texts = "hello world, how are you?"
+    benchmark_texts = [test_texts + "<pad>" * predictor_args.src_length for _ in range(predictor_args.batch_size)]
 
     batch_benchmark_texts = batchfy_text(benchmark_texts, predictor_args.batch_size)
     print("***********Start Benchmark**********")