Spaces:
Sleeping
Sleeping
import streamlit as st | |
import random | |
import openai | |
import os | |
from PIL import Image | |
from transformers import pipeline | |
if "model" not in st.session_state: | |
st.session_state.model = pipeline("image-to-text", model="Salesforce/blip-image-captioning-large") | |
if "result" not in st.session_state: | |
st.session_state.result = "" | |
if "essay" not in st.session_state: | |
st.session_state.essay = "" | |
# 初始化图片描述生成模型 | |
# image_to_text = pipeline("image-to-text", model="Salesforce/blip-image-captioning-large") | |
# Set OpenAI API key | |
OPENAI_API_KEY = st.secrets["OPENAI_API_KEY"] | |
if not OPENAI_API_KEY: | |
st.error("OPENAI_API_KEY not set in environment variables!") | |
raise SystemExit | |
openai.api_key = OPENAI_API_KEY | |
st.set_page_config(layout="wide", initial_sidebar_state="collapsed") | |
def split_image(image, output_dir='.'): | |
# 获取图片尺寸 | |
width, height = image.size | |
# 计算每一部分的宽度 | |
part_width = width // 3 | |
# 存储分割后图片的路径和描述 | |
results = [] | |
# 分割图片并保存 | |
for i in range(3): | |
left = i * part_width | |
right = left + part_width if i != 2 else width | |
part_img = image.crop((left, 0, right, height)) | |
file_path = os.path.join(output_dir, f'part_{i + 1}.png') | |
part_img.save(file_path) | |
# 生成图片描述 | |
caption = st.session_state.model(part_img) | |
results.append((file_path, caption)) | |
return results | |
def add_numbering_after_dot(input_string): | |
sentences = input_string.split(".") | |
numbered_sentences = [] | |
sentence_count = 1 | |
for sentence in sentences: | |
sentence = sentence.strip() | |
if sentence: | |
numbered_sentences.append(f"{sentence}.【{sentence_count}】") | |
sentence_count += 1 | |
return " ".join(numbered_sentences) | |
def add_paragraph_numbering(input_text): | |
paragraphs = input_text.split("\n\n") | |
numbered_text = "" | |
for i, paragraph in enumerate(paragraphs, start=1): | |
numbered_text += f"{i}. {paragraph}\n\n" | |
return numbered_text | |
def get_completion_from_messages(messages, | |
model="gpt-3.5-turbo-16k", | |
temperature=0, max_tokens=1000): | |
response = openai.ChatCompletion.create( | |
model=model, | |
messages=messages, | |
temperature=temperature, | |
max_tokens=max_tokens, | |
) | |
return response.choices[0].message["content"] | |
student_essay = '''One Saturday, Jack’s parents were sitting in the sofa and reading a magazic. “ | |
I want to go Sanya on holiday!” said Jack in a cheerful manner. “Let’s go!” | |
A few minutes later they arrived to the air plane station. | |
Suddenly, the sky turned black and it stared to rain. “Oh, it’s a bad weather to go travelling!’ said Jack. | |
One hours later, they arrived the Sanya, | |
the sky was sunny and they played on the beach. “What a wonderful day!” said Jack.''' | |
example = """ | |
student essay: One day, dad was reading a magazine where there have lots of information in it. | |
“Would you like to go on holiday in San Ya?” asked dad. He thinks San Ya is a warm city in China, | |
which is the perfect place to spend holiday. Then, he bought three tickets. | |
It rains dogs and cats when they were going to the plane. “What a bad weather!” said Tom. | |
One days ago, they arrived at there on Saturday. | |
A local person, who is the photographer, helped they took a photo. They had a happy holiday. | |
image content: a cartoon of a family sitting on a couch reading a book,a cartoon drawing of a group of people climbing up a flight of stairs | |
,a cartoon drawing of a family standing on the beach | |
you should return: | |
Total mark: 15 points | |
------------------------------------------------------------------------------------ | |
Content | |
5 points | |
- Each picture involves content: ✔️ | |
- Each image contains 2-3 sentences; if sub-clauses are used, description can be reduced to 1-2 sentences for each image: ✔️ | |
- Writing is relevant with sufficient information conveyed: ✔️ | |
------------------------------------------------------------------------------------ | |
Organization | |
5 points | |
- Structured writing with clear introduction, body, and conclusion. The first two sentences provide clear context regarding time, place, individuals, and actions: ✔️ | |
- Sentences are 90% logically connected: ✔️ | |
- Incorporates 2-3 different linkage techniques such as conjunctions (because, but, so) and adverbs (unexpectedly, fortunately): ✔️ | |
- Uses at least 5 linkage techniques, with a mixture of 2-3 different methods: ✔️ | |
------------------------------------------------------------------------------------ | |
Language | |
5 points | |
Dimension 1: Complexity | |
- About 20% (15-20%) of words are bi-syllabic or multi-syllabic: ✔️ | |
- Includes 3 words of B1 level or above: ✔️ | |
- Covers 2-3 types of sub-clauses: adverbial and relative: ✔️ | |
- At least 3 compound sentence structures: ✔️ | |
Dimension 2: Diversity | |
- Diverse word usage without obvious repetitive grammar structures: ✔️ | |
Language Dimension 3: Accuracy | |
- Word spelling errors (e.g., "the-they") and grammatical errors (e.g., "...Where there were lots of pleasant beaches...; I wanted to travelling to Sanya, Because"): ❌ (Vocabulary richness compensates for this shortcoming) | |
""" | |
example_2 = """ | |
student essay: | |
There is a Jack’s birthday. There are he, his brother, mum and dad. His brother is giving him a ball to football. Next day, Jack and his brother are playing football match. Jack score a goal. | |
image content: cartoon of a family sitting around a table with a birthday cake,cartoon of a man giving a birthday cake to a woman, | |
cartoon of a man kicking a soccer ball into a goal | |
you should return: | |
Total mark: 9 points | |
------------------------------------------------------------------------------------ | |
Content | |
3 points: | |
- Each picture involves content: ✔️ | |
- Each picture has at least one sentence: ✔️ | |
- The content is relevant, but the information derived from the image is not exhaustive: ❌ | |
Organization | |
3 points: | |
- Complete structure with a clear introduction, development, and conclusion: ✔️ | |
- 50% of the sentences have logical connections: ✔️ | |
- Contains one to two types of cohesive devices, such as pronouns (e.g., he, his brother) and temporal relations (e.g., next day). The number of cohesive devices used ranges between 2-4 for one type: ✔️ | |
Language | |
3 points: | |
*Dimension 1: Complexity* | |
- Bi-syllabic and multi-syllabic words make up 9%: ✔️ | |
- All words are common, everyday vocabulary: ✔️ | |
- No complex sentences or compound sentences: ❌ | |
*Dimension 2: Diversity* | |
- Repetition of words: "Brother" appears 3 times, "football" appears 2 times: ❌ | |
*Dimension 3: Accuracy* | |
- No word spelling mistakes: ✔️ | |
- Did not use past tense: ❌ | |
- Grammatical errors in four places: "There is a Jack’s birthday", "There are he", "to football", "Jack score": ❌ | |
""" | |
system_message = f''' | |
你是一个剑桥KET作文老师。 | |
请你认真批改学生在【】符号中的作文 | |
首先你会给出你的评分,在评分文章时,你会从以下方面考虑: | |
Content | |
Organization | |
Language | |
记住KET的Band Score都是整数 | |
total = Content Band Score + Organization Band Score + Language Band Score。 | |
以下是评分的要求: | |
Band 5:Content:一个高质量的5级作文要确保每一幅图的内容都被涉及,并且每幅图都至少有2-3个句子描述(如果使用从句,描述可以缩减到1-2句)。整体的写作内容需要与主题紧密相关,并确保信息得到充分的传达。 | |
Organization:作文需要有一个完整的结构,包括明确的开头、经过和结尾,且第一句应清晰地交代时间、地点、人物和行为。句子之间的逻辑连接需要达到90%以上,使用的衔接手段需要多样,涵盖两到三种不同类型(如连词、指代和特定的副词)。而且,衔接手段的使用不能少于5次,且不能都是同一种手段。 | |
Language:复杂度:这要求作文中双音节和多音节词的占比为15-20%,至少有3-5个非日常生活词汇,包括2-3种不同类型的从句(如宾语从句、状语从句和定语从句),以及不少于3句的复合句式。 | |
丰富度:作文中的单词和句子结构应该变化丰富,相同的单词和句式最多只能出现1次(人名除外)。 | |
准确度:作文中的单词拼写和句式使用错误不能超过3个,并特别注意描述语言应使用过去时态,而对话部分则没有时态限制。 | |
Band 4:Content:1.每幅图的内容都涉及 2.每幅图至少有1个句子 3.写作内容基本相关,1幅图细节不充分 | |
Organization:1. 结构完整(明确的开头、经过和结尾)2.句与句之间有70%的衔接逻辑 3.包含两种衔接手段:连词、指代或副词 4.衔接手段数量为3-5个(涵盖两个种类) | |
Language:维度1:复杂度 1.双音节、多音节词占比为10-15% 2.有1-2个非日常生活词汇 3.1-2种从句:宾从、状从或定从 4.复合句式2-3句 | |
维度2:丰富度 相同单词、句式结构最多复现2次 | |
维度3:准确度 单词拼写错误、句式错误:3-5个(描述语言使用过去时态,对话语言不限时态) | |
Band 3:Content:1.每幅图内容基本涉及 2.每幅图至少有1个句子 3.写作内容稍偏离主题,至少有1幅图细节不充分 | |
Organization:1.结构完整(明确的开头、经过和结尾)2.句与句之间有50%的衔接逻辑 3.包含一到两种衔接手段:连词、指代或副词4.衔接手段数量为2-4个(一个种类) | |
Language:维度1:复杂度 1. 双音节、多音节词占比为5-10% 2.单词均为日常生活词汇 3.1种从句:宾从、状从或定从 4.复合句式1句 | |
维度2:丰富度 相同单词、句式复现2次以上 | |
维度3:准确度 单词拼写错误、句式错误:6-8个(语言错误少于6个,可不写从句) | |
Band 2:Content:1.图片内容没有完全覆盖或图片信息有误 2.某一幅图片1个句子的描述都没有 3.写作内容稍偏离主题,至少2幅图片细节不充分 | |
Organization:1. 结构不完整(有开头没经过或者有开头没结尾)2.句与句之间有一些逻辑(30%) 3.包含一种衔接手段:连词、指代或副词 4.衔接手段数量为1-2个 | |
Language:维度1:复杂度 1.双音节、多音节词占比少于5% 2.单词均为日常生活词汇 3.无从句或复合句 | |
维度2:准确度 单词拼写错误、句式错误:5-8个 | |
Band 1:Content:1.图片信息缺漏、有误 2.两幅图片1个句子的描述都没有 3.写作内容与图片不相干 | |
Organization:1.故事结构不完整 2.句与句之间几乎无逻辑(10%) 3.包含一种衔接手段:连词、指代或副词 4.衔接手段数量为1个 | |
Language:维度1:复杂度 1.双音节、多音节词占比少于3% 2.单词均为日常生活词汇 3.无从句或复合句 | |
维度2:准确度 单词拼写错误、句式错误:10个以上 | |
Band 0:Content:写作内容与图片毫无关联 | |
Organization:1. 缺乏故事结构 2.句与句之间毫无逻辑(0%) 3.无衔接手段 | |
Language:1. 无双音节、多音节单词 2.无完整句型 | |
把回答都整理成markdown格式。不要给出具体建议和修改文章。 | |
这是一个具体的例子:{example} | |
这是另一个例子:{example_2} | |
''' | |
st.markdown(""" | |
<h2 style='color: black;'> | |
阅思乐KET作文批改 demo 0.0.2 | |
</h2> | |
""", | |
unsafe_allow_html=True | |
) | |
st.markdown( | |
""" | |
<style> | |
/* 修改标题的大小 */ | |
.streamlit-container h1 { | |
font-size: 2.5em !important; | |
} | |
</style> | |
""", | |
unsafe_allow_html=True, | |
) | |
col1, col2 = st.columns([3,3]) | |
with col1: | |
uploaded_image = st.file_uploader("上传作文图片", type=["jpg", "png", "jpeg"]) | |
if uploaded_image: | |
st.image(uploaded_image, caption="已上传的图片。", use_column_width=True) | |
st.session_state.essay = st.text_area( | |
" 输入你的作文", | |
height= 400 | |
) | |
push = False | |
if st.button("开始打分"): | |
push = True | |
with col2: | |
st.markdown( | |
""" | |
<style> | |
/* 创建一个固定大小的容器,并允许滚动 */ | |
.scrollable-container { | |
max-height: 400px; /* 容器的高度 */ | |
overflow-y: auto; /* 允许垂直滚动 */ | |
} | |
</style> | |
""", | |
unsafe_allow_html=True, | |
) | |
if student_essay and push and uploaded_image: | |
with st.spinner('处理图片...'): | |
image = Image.open(uploaded_image).convert("RGB") | |
output_dir = '.' | |
st.session_state.result = split_image(image, output_dir) | |
progress_placeholder = st.empty() | |
progress_text = "开始打分" | |
my_bar = progress_placeholder.progress(0, text=progress_text) | |
query = str(st.session_state.essay) | |
user_message = query | |
messages = [ | |
{'role':'system', | |
'content': system_message}, | |
{'role':'user', | |
'content': f"remember your settings student essay:【{query}】, image content:{st.session_state.result}"},] | |
response = get_completion_from_messages(messages) | |
my_bar.progress(50, text="开始提供相关素材") | |
system_message_4 = ''' | |
你是一个KET老师,根据学生提供的文章,提供一些素材和观点的地道英文表达及中文释义; | |
在每句英文的后面的括号里给出中文释义 | |
并把回答都整理成markdown格式。 | |
这是一个具体的例子: | |
返回: | |
❤ 此类文章可能会用到的高分素材和观点,提供给你参考 ❤ | |
- it was a scorching summer day. | |
- ... along which there were a wide variety of shops. | |
- It was terrible to run on such a scorching/burning day. | |
- The moment he stopped for a short break, he had sweated heavily. (一般过去时与过去完成时连用) | |
- The reason why he went into a convenient store was that he wanted to get something icy to drink. | |
- ...sweated heavily; Jimmy asked the shopkeeper politely | |
- ...replied Jimmy in a cheerful manner. | |
学生:''' | |
user_message = query | |
messages = [ | |
{'role':'system', | |
'content': system_message_4}, | |
{'role':'user', | |
'content': f"请记住你的设定【{query}】"},] | |
response3 = get_completion_from_messages(messages) | |
my_bar.progress(100, text="打分完成") | |
st.markdown('<div class="scrollable-container">', unsafe_allow_html=True) | |
with st.container(): | |
st.markdown(response) | |
st.divider() | |
with st.container(): | |
st.markdown(""" | |
<h4 style='color: black;'> | |
素材建议 | |
</h4> | |
""", | |
unsafe_allow_html=True | |
) | |
st.write(response3) | |
my_bar.empty() | |
st.markdown('</div>', unsafe_allow_html=True) |