(self, prompt)
| 39 | |
| 40 | @retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6)) |
| 41 | def generate_text(self, prompt): |
| 42 | prompt = self.format_prompt(prompt) |
| 43 | inputs = self.tokenizer.apply_chat_template(conversation=prompt, tokenize=True, add_generation_prompt=True, return_tensors='pt').to(self.model.device) |
| 44 | outputs = self.model.generate( |
| 45 | inputs, |
| 46 | do_sample=self.do_sample |
| 47 | ) |
| 48 | response = self.tokenizer.decode(outputs[0][inputs.shape[1]:], skip_special_tokens=True) |
| 49 | |
| 50 | return response.strip() |
| 51 | |
| 52 | @retry(wait=wait_random_exponential(min=1, max=60), stop=stop_after_attempt(6)) |
| 53 | def count_tokens(self, prompt): |
nothing calls this directly
no test coverage detected