update handler and some files

handler.py

@@ -1,18 +1,31 @@
 from typing import Dict, List, Any
-from transformers import WhisperForConditionalGeneration, pipeline
+from transformers import (
+    AutomaticSpeechRecognitionPipeline,
+    WhisperForConditionalGeneration,
+    WhisperTokenizer,
+    WhisperProcessor
+)
 from peft import LoraConfig, PeftModel, LoraModel, LoraConfig, get_peft_model, PeftConfig
 
 class EndpointHandler():
     def __init__(self, path=""):
         # Preload all the elements you are going to need at inference.
-        peft_config = PeftConfig.from_pretrained(path)
-        self.model= WhisperForConditionalGeneration.from_pretrained(
+        language = "Chinese"
+        task = "transcribe" 
+        self.peft_config = PeftConfig.from_pretrained(path)
+        model= WhisperForConditionalGeneration.from_pretrained(
             peft_config.base_model_name_or_path
         )
-        self.model = PeftModel.from_pretrained(self.model, peft_model_id)
-        self.pipeline = pipeline(task= "automatic-speech-recognition", model=self.model)
-        self.pipeline.model.config.forced_decoder_ids = self.pipeline.tokenizer.get_decoder_prompt_ids(language="Chinese", task="transcribe")
-        self.pipeline.model.generation_config.forced_decoder_ids = self.pipeline.model.config.forced_decoder_ids
+        model = PeftModel.from_pretrained(model, peft_model_id)
+        tokenizer = WhisperTokenizer.from_pretrained(peft_config.base_model_name_or_path, language=language, task=task)
+        processor = WhisperProcessor.from_pretrained(peft_config.base_model_name_or_path, language=language, task=task)
+        feature_extractor = processor.feature_extractor
+        forced_decoder_ids = processor.get_decoder_prompt_ids(language=language, task=task)
+        self.pipeline = AutomaticSpeechRecognitionPipeline(model=model, tokenizer=tokenizer, feature_extractor=feature_extractor)
+
+        # self.pipeline = pipeline(task= "automatic-speech-recognition", model=self.model)
+        # self.pipeline.model.config.forced_decoder_ids = self.pipeline.tokenizer.get_decoder_prompt_ids(language="Chinese", task="transcribe")
+        # self.pipeline.model.generation_config.forced_decoder_ids = self.pipeline.model.config.forced_decoder_ids
 
     def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
         """
@@ -24,5 +37,6 @@ class EndpointHandler():
         """
 
         inputs = data.pop("inputs", data)
-        prediction = self.pipeline(inputs, return_timestamps=False)
-        return prediction
+        with torch.cuda.amp.autocast():
+            predicion = self.pipeline(inputs, generate_kwargs={"forced_decoder_ids": forced_decoder_ids}, max_new_tokens=255)["text"]
+        return {"prediction": prediction}

tokenizer_config.json

@@ -9,6 +9,7 @@
     "rstrip": false,
     "single_word": false
   },
+  "clean_up_tokenization_spaces": true,
   "dropout": 0.0,
   "eos_token": {
     "__type": "AddedToken",
@@ -23,7 +24,6 @@
   "pad_token": null,
   "processor_class": "WhisperProcessor",
   "return_attention_mask": false,
-  "special_tokens_map_file": null,
   "tokenizer_class": "WhisperTokenizer",
   "unk_token": {
     "__type": "AddedToken",