(self, pretrained_model_name_or_path, cache_dir=None, do_sample=False, max_length=4096)
| 9 | class PULSEEvaluator(Evaluator): |
| 10 | |
| 11 | def __init__(self, pretrained_model_name_or_path, cache_dir=None, do_sample=False, max_length=4096): |
| 12 | super(PULSEEvaluator, self).__init__() |
| 13 | |
| 14 | self.tokenizer = AutoTokenizer.from_pretrained( |
| 15 | pretrained_model_name_or_path=pretrained_model_name_or_path, |
| 16 | cache_dir=cache_dir, |
| 17 | trust_remote_code=True |
| 18 | ) |
| 19 | self.model = AutoModelForCausalLM.from_pretrained( |
| 20 | pretrained_model_name_or_path=pretrained_model_name_or_path, |
| 21 | cache_dir=cache_dir, |
| 22 | device_map='auto', |
| 23 | low_cpu_mem_usage=True, |
| 24 | torch_dtype=torch.float16, |
| 25 | trust_remote_code=True |
| 26 | ) |
| 27 | self.model = self.model.eval() |
| 28 | self.do_sample = do_sample |
| 29 | self.max_length = max_length |
| 30 | print(f'Memory footprint: {self.model.get_memory_footprint() / 1e6:.2f} MB') |
| 31 | |
| 32 | def format_prompt(self, prompt): |
| 33 | return f'Instructions: You are Helper, a large language model full of intelligence. Respond conversationally.</s>User: {prompt}</s> Helper: ' |
nothing calls this directly
no outgoing calls
no test coverage detected