fix json format issue for "transformerpipeline" (#63)

Author: depenglee1707

llmserve/backend/llm/pipelines/default_pipeline.py

@@ -57,8 +57,6 @@ def preprocess(self, prompts: List[str], **generate_kwargs):
         try:
             prompt_text_bak = prompt_text
             logger.info(f"call json.loads")
-            # for p in prompt_text:
-            #     logger.info(f"{p}")
             prompt_text = [json.loads(prompt, strict=False) for prompt in prompt_text]
             logger.info(f"call tokenizer.apply_chat_template")
             prompt_text = [self.tokenizer.apply_chat_template(prompt_obj, tokenize=False, add_generation_prompt=True) for prompt_obj in prompt_text]

llmserve/backend/llm/pipelines/default_transformers_pipeline.py

@@ -144,7 +144,7 @@ def preprocess(self, prompts: List[str], **generate_kwargs):
         if isinstance(self.pipeline, transformers.pipelines.text_generation.TextGenerationPipeline):
             try:
                 prompt_text_bak = prompt_text
-                prompt_text = [json.loads(prompt) for prompt in prompt_text]
+                prompt_text = [json.loads(prompt, strict=False) for prompt in prompt_text]
                 prompt_text = [self.tokenizer.apply_chat_template(prompt_obj, tokenize=False, add_generation_prompt=True) for prompt_obj in prompt_text]
             except:
                 logger.info("Seems no chat template from user or the model donot has a 'chat template'")