| | from typing import List, Dict, Any |
| | from transformers import AutoTokenizer, AutoModelForSeq2SeqLM |
| |
|
| | class EndpointHandler(): |
| | def __init__(self, path=""): |
| | self.model = model = AutoModelForSeq2SeqLM.from_pretrained(path) |
| | self.tokenizer = AutoTokenizer.from_pretrained(path) |
| | self.bos_instruction = "Definition: The output will be the aspects (both implicit and explicit) the corresponding opinion/describing terms and the sentiment polarity (positive, negative, neutral) of the opinion term . In cases where there are no aspects the output should be noaspectterm:none:none.\n Positive example 1-\n input: Faan 's got a great concept but a little rough on the delivery .\n output: delivery:rough:positive\n Positive example 2- \n input: I just wonder how you can have such a delicious meal for such little money .\n output: meal:delicious:positive, money:little:positive\n Now complete the following example-\n input: " |
| | self.eos_instruct = ' \noutput:' |
| | self.delim_instruct = '' |
| |
|
| | def preprocess(self, inputs: List[str]): |
| | processed_inputs = [] |
| | for text in inputs: |
| | processed_inputs.append(self.bos_instruction + text + self.delim_instruct + self.eos_instruct) |
| | return self.tokenizer(processed_inputs, return_tensors="pt", truncation=True, padding=True) |
| |
|
| | def postprocess(self, outputs): |
| | output = self.tokenizer.decode(outputs[0], skip_special_tokens=True) |
| | aspects = [] |
| | opinions = [] |
| | sentiments = [] |
| | for triplet in output.split(','): |
| | triplet = triplet.strip() |
| | if triplet == '': |
| | continue |
| | aos = triplet.split(':') |
| | aspects.append(aos[0]) |
| | opinions.append(aos[1]) |
| | sentiments.append(aos[2]) |
| | result = "" |
| | for i in range(len(aspects)): |
| | result += f"Aspect: {aspects[i]} - Opinion: {opinions[i]} - Sentiment: {sentiments[i]}\n" |
| | result = result.strip() |
| | return result |
| |
|
| | def __call__(self, data: Dict[str, Any]): |
| | inputs = data.pop('inputs', data) |
| | parameters = data.pop("parameters", None) |
| | model_inputs = self.preprocess(inputs) |
| | model_outputs = self.model.generate(model_inputs.input_ids) |
| | return self.postprocess(model_outputs) |
