Update handler.py

handler.py

@@ -1,7 +1,7 @@
 import torch
 from typing import Dict, List, Any
 from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline
-import logging
+from transformers.generation.utils import GenerationConfig
 
 # get dtype
 dtype = torch.bfloat16 if torch.cuda.get_device_capability()[0] == 8 else torch.float16
@@ -9,29 +9,13 @@ dtype = torch.bfloat16 if torch.cuda.get_device_capability()[0] == 8 else torch.
 class EndpointHandler:
     def __init__(self, path=""):
         # load the model
-        self.tokenizer = AutoTokenizer.from_pretrained(path, trust_remote_code=True)
-        self.model = AutoModelForCausalLM.from_pretrained(path, device_map="auto", torch_dtype=dtype, trust_remote_code=True)
-        # create inference pipeline
-        self.pipeline = pipeline("text-generation", model=self.model, tokenizer=self.tokenizer)
+        this.tokenizer = AutoTokenizer.from_pretrained("baichuan-inc/Baichuan-13B-Chat", use_fast=False, trust_remote_code=True)
+        this.model = AutoModelForCausalLM.from_pretrained("baichuan-inc/Baichuan-13B-Chat", device_map="auto", torch_dtype=dtype, trust_remote_code=True)
+        this.model.generation_config = GenerationConfig.from_pretrained("baichuan-inc/Baichuan-13B-Chat")
 
     def __call__(self, data: Any) -> List[List[Dict[str, float]]]:
         inputs = data.pop("inputs", data)
-        parameters = data.pop("parameters", None)
-
-        # pass inputs with all kwargs in data
-        if parameters is not None:
-            prediction = self.pipeline(inputs, **parameters)
-        else:
-            prediction = self.pipeline(inputs)
-        logging.warn("---start---")
-        logging.warn(prediction)
-        logging.warn("---end---")
-
         # ignoring parameters! Default to configs in generation_config.json.
-        messages = [{"role": "user", "content": data.pop("inputs", data)}]
-        response = self.model.chat(self.tokenizer, messages)
-        logging.warn("---start chat response---")
-        logging.warn(response)
-        logging.warn("---end chat response---")
-
-        return [[{response: 1.0}]]
+        messages = [{"role": "user", "content": inputs}]
+        response = this.model.chat(this.tokenizer, messages)
+        return [{'generated_text': response}]