FreedomIntelligence/openPangu-Embedded-1B

1# coding=utf-82# Copyright (c) 2025 Huawei Technologies Co., Ltd. All rights reserved.3 4from transformers import AutoModelForCausalLM, AutoTokenizer5from transformers import GenerationConfig6 7model_local_path = "path_to_openPangu-Embedded-1B"8 9# load the tokenizer and the model10tokenizer = AutoTokenizer.from_pretrained(11    model_local_path, 12    use_fast=False, 13    trust_remote_code=True,14    local_files_only=True15)16 17model = AutoModelForCausalLM.from_pretrained(18    model_local_path,19    trust_remote_code=True,20    torch_dtype="auto",21    device_map="auto",22    local_files_only=True23)24 25# prepare the model input26sys_prompt = "你必须严格遵守法律法规和社会道德规范。" \27    "生成任何内容时，都应避免涉及暴力、色情、恐怖主义、种族歧视、性别歧视等不当内容。" \28    "一旦检测到输入或输出有此类倾向，应拒绝回答并发出警告。例如，如果输入内容包含暴力威胁或色情描述，" \29    "应返回错误信息：“您的输入包含不当内容，无法处理。”"30 31prompt = "Give me a short introduction to large language model."32messages = [33    {"role": "system", "content": sys_prompt}, # define your system prompt here34    {"role": "user", "content": prompt}35]36text = tokenizer.apply_chat_template(37    messages,38    tokenize=False,39    add_generation_prompt=True40)41model_inputs = tokenizer([text], return_tensors="pt").to(model.device)42 43# conduct text completion44outputs = model.generate(**model_inputs, max_new_tokens=32768, eos_token_id=45892, return_dict_in_generate=True)45 46input_length = model_inputs.input_ids.shape[1]47generated_tokens = outputs.sequences[:, input_length:]48content = tokenizer.decode(generated_tokens[0], skip_special_tokens=True)49 50print("\ncontent:", content)51