Views
No views yet
learning_rate=1e-5
lr_scheduler_type=cosine
max_length=2048
warmup_ratio=0.05
batch_size=64
epoch=101# !/usr/bin/env python
2# -*- coding:utf-8 -*-
3# ==================================================================
4# [Author] : xiaofeng
5# [Descriptions] :
6# ==================================================================
7
8from transformers import AutoTokenizer, AutoModelForCausalLM
9import transformers
10import torch
11
12
13llama3_jinja = """{% if messages[0]['role'] == 'system' %}
14 {% set offset = 1 %}
15{% else %}
16 {% set offset = 0 %}
17{% endif %}
18
19{{ bos_token }}
20{% for message in messages %}
21 {% if (message['role'] == 'user') != (loop.index0 % 2 == offset) %}
22 {{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}
23 {% endif %}
24
25 {{ '<|start_header_id|>' + message['role'] + '<|end_header_id|>\n\n' + message['content'] | trim + '<|eot_id|>' }}
26{% endfor %}
27
28{% if add_generation_prompt %}
29 {{ '<|start_header_id|>' + 'assistant' + '<|end_header_id|>\n\n' }}
30{% endif %}"""
31
32
33dtype = torch.bfloat16
34
35model_dir = "XiaofengAlg/Technology-llama3_1_8B_instruct"
36model = AutoModelForCausalLM.from_pretrained(
37 model_dir,
38 device_map="cuda",
39 torch_dtype=dtype,
40)
41
42tokenizer = AutoTokenizer.from_pretrained(model_dir)
43tokenizer.chat_template = llama3_jinja # update template
44
45message = [
46 {"role": "system", "content": "You are a helpful assistant"},
47 {
48 "role": "user",
49 "content": "请详细描述科技研究如何改变了我们的教育系统。",
50 },
51]
52prompt = tokenizer.apply_chat_template(
53 message, tokenize=False, add_generation_prompt=True
54)
55print(prompt)
56inputs = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt")
57prompt_length = len(inputs[0])
58print(f"prompt_length:{prompt_length}")
59
60generating_args = {
61 "do_sample": True,
62 "temperature": 1.0,
63 "top_p": 0.5,
64 "top_k": 15,
65 "max_new_tokens": 512,
66}
67
68
69generate_output = model.generate(input_ids=inputs.to(model.device), **generating_args)
70
71response_ids = generate_output[:, prompt_length:]
72response = tokenizer.batch_decode(
73 response_ids, skip_special_tokens=True, clean_up_tokenization_spaces=True
74)[0]
75
76
77"""
78科技研究对我们的教育系统产生了深远的影响。首先，科技研究使得教育变得更加普及。通过互联网和数字化技术，学生可以在任何时间、任何地点接受教育，这大大增加了教育的可获取性。其次，科技研究也使得教育变得更加个性化。通过大数据和人工智能等技术，教育系统可以根据每个学生的学习情况和需求，提供定制化的教学方案。此外，科技研究还促进了教育的互动性。通过虚拟现实、增强现实等技术，学生可以更好地参与到学习中来，提高学习的趣味性和效果。总的来说，科技研究正在不断地推动教育系统的发展，使教育更加普及、个性化和互动。
79"""
80print(f"response:{response}")
81
821@misc{shi2024industryinstruction,
2 title = {IndustryInstruction},
3 author = {Xiaofeng Shi and Lulu Zhao and Hua Zhou and Donglin Hao and Yonghua Lin},
4 year = {2024},
5 publisher = {Hugging Face},
6 doi = {10.57967/hf/3487},
7 url = {https://huggingface.co/datasets/BAAI/IndustryInstruction}
8}