| import torch |
| from typing import Dict, List, Any |
| from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline |
| from transformers.generation.utils import GenerationConfig |
|
|
| |
| dtype = torch.bfloat16 if torch.cuda.get_device_capability()[0] == 8 else torch.float16 |
|
|
| class EndpointHandler: |
| def __init__(self, path=""): |
| |
| self.model = AutoModelForCausalLM.from_pretrained(path, device_map="auto", torch_dtype=dtype, trust_remote_code=True) |
| self.model.generation_config = GenerationConfig.from_pretrained(path) |
| self.tokenizer = AutoTokenizer.from_pretrained(path, use_fast=False, trust_remote_code=True) |
|
|
| def __call__(self, data: Any) -> List[List[Dict[str, float]]]: |
| inputs = data.pop("inputs", data) |
| |
| messages = [{"role": "user", "content": inputs}] |
| response = self.model.chat(self.tokenizer, messages) |
| if torch.backends.mps.is_available(): |
| torch.mps.empty_cache() |
| return [{'generated_text': response}] |
|
|