init

chatpdf.py

@@ -0,0 +1,190 @@
+# -*- coding: utf-8 -*-
+"""
+@author:XuMing([email protected])
+@description: 
+"""
+import logging
+
+from similarities import Similarity
+from textgen import ChatGlmModel, LlamaModel
+from transformers import pipeline
+from loguru import logger
+
+PROMPT_TEMPLATE = """\
+基于以下已知信息，简洁和专业的来回答用户的问题。
+如果无法从中得到答案，请说 "根据已知信息无法回答该问题" 或 "没有提供足够的相关信息"，不允许在答案中添加编造成分，答案请使用中文。
+
+已知内容:
+{context_str}
+
+问题:
+{query_str}
+"""
+
+
+class ChatPDF:
+    def __init__(
+            self,
+            sim_model_name_or_path: str = "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2",
+            gen_model_type: str = "chatglm",
+            gen_model_name_or_path: str = "THUDM/chatglm-6b-int4",
+            lora_model_name_or_path: str = None,
+
+    ):
+        self.sim_model = Similarity(model_name_or_path=sim_model_name_or_path)
+        self.model_type = gen_model_type
+
+        if gen_model_type == "chatglm":
+            self.gen_model = ChatGlmModel(gen_model_type, gen_model_name_or_path, lora_name=lora_model_name_or_path)
+        elif gen_model_type == "llama":
+            self.gen_model = LlamaModel(gen_model_type, gen_model_name_or_path, lora_name=lora_model_name_or_path)
+        elif gen_model_type == "t5":
+            self.gen_model = pipeline('text2text-generation', model=gen_model_name_or_path, device=0)
+        else:
+            raise ValueError('gen_model_type must be chatglm or llama.')
+        self.history = None
+        self.pdf_path = None
+
+    def load_pdf_file(self, pdf_path: str):
+        """Load a PDF file."""
+        if pdf_path.endswith('.pdf'):
+            corpus = self.extract_text_from_pdf(pdf_path)
+        elif pdf_path.endswith('.docx'):
+            corpus = self.extract_text_from_docx(pdf_path)
+        elif pdf_path.endswith('.md'):
+            corpus = self.extract_text_from_markdown(pdf_path)
+        else:
+            corpus = self.extract_text_from_txt(pdf_path)
+        self.sim_model.add_corpus(corpus)
+        self.pdf_path = pdf_path
+
+    @staticmethod
+    def extract_text_from_pdf(file_path: str):
+        """Extract text content from a PDF file."""
+        import PyPDF2
+        contents = []
+        with open(file_path, 'rb') as f:
+            pdf_reader = PyPDF2.PdfReader(f)
+            for page in pdf_reader.pages:
+                page_text = page.extract_text().strip()
+                raw_text = [text.strip() for text in page_text.splitlines() if text.strip()]
+                new_text = ''
+                for text in raw_text:
+                    new_text += text
+                    if text[-1] in ['.', '!', '?', '。', '！', '？', '…', ';', '；', ':', '：', '”', '’', '）', '】', '》', '」',
+                                    '』', '〕', '〉', '》', '〗', '〞', '〟', '»', '"', "'", ')', ']', '}']:
+                        contents.append(new_text)
+                        new_text = ''
+                if new_text:
+                    contents.append(new_text)
+        return contents
+
+    @staticmethod
+    def extract_text_from_txt(file_path: str):
+        """Extract text content from a TXT file."""
+        contents = []
+        with open(file_path, 'r', encoding='utf-8') as f:
+            contents = [text.strip() for text in f.readlines() if text.strip()]
+        return contents
+
+    @staticmethod
+    def extract_text_from_docx(file_path: str):
+        """Extract text content from a DOCX file."""
+        import docx
+        document = docx.Document(file_path)
+        contents = [paragraph.text.strip() for paragraph in document.paragraphs if paragraph.text.strip()]
+        return contents
+
+    @staticmethod
+    def extract_text_from_markdown(file_path: str):
+        """Extract text content from a Markdown file."""
+        import markdown
+        from bs4 import BeautifulSoup
+        with open(file_path, 'r', encoding='utf-8') as f:
+            markdown_text = f.read()
+        html = markdown.markdown(markdown_text)
+        soup = BeautifulSoup(html, 'html.parser')
+        contents = [text.strip() for text in soup.get_text().splitlines() if text.strip()]
+        return contents
+
+    @staticmethod
+    def _add_source_numbers(lst):
+        """Add source numbers to a list of strings."""
+        return [f'[{idx + 1}]\t "{item}"' for idx, item in enumerate(lst)]
+
+    def _generate_answer(self, query_str, context_str, history=None, max_length=1024):
+        """Generate answer from query and context."""
+        if self.model_type == "t5":
+            response = self.gen_model(query_str, max_length=max_length, do_sample=True)[0]['generated_text']
+            return response, history
+        prompt = PROMPT_TEMPLATE.format(context_str=context_str, query_str=query_str)
+        response, out_history = self.gen_model.chat(prompt, history, max_length=max_length)
+        return response, out_history
+
+    def chat(self, query_str, history=None, max_length=1024):
+        if self.model_type == "t5":
+            response = self.gen_model(query_str, max_length=max_length, do_sample=True)[0]['generated_text']
+            logger.debug(response)
+            return response, history
+
+        response, out_history = self.gen_model.chat(query_str, history, max_length=max_length)
+        return response, out_history
+
+    def query(
+            self,
+            query,
+            topn: int = 5,
+            max_length: int = 1024,
+            max_input_size: int = 1024,
+            use_history: bool = False
+    ):
+        """Query from corpus."""
+
+        sim_contents = self.sim_model.most_similar(query, topn=topn)
+
+        reference_results = []
+        for query_id, id_score_dict in sim_contents.items():
+            for corpus_id, s in id_score_dict.items():
+                reference_results.append(self.sim_model.corpus[corpus_id])
+        if not reference_results:
+            return '没有提供足够的相关信息', reference_results
+        reference_results = self._add_source_numbers(reference_results)
+
+        context_str = '\n'.join(reference_results)[:(max_input_size - len(PROMPT_TEMPLATE))]
+
+        if use_history:
+            response, out_history = self._generate_answer(query, context_str, self.history, max_length=max_length)
+            self.history = out_history
+        else:
+
+            response, out_history = self._generate_answer(query, context_str)
+
+        return response, out_history, reference_results
+
+    def save_index(self, index_path=None):
+        """Save model."""
+        if index_path is None:
+            index_path = '.'.join(self.pdf_path.split('.')[:-1]) + '_index.json'
+        self.sim_model.save_index(index_path)
+
+    def load_index(self, index_path=None):
+        """Load model."""
+        if index_path is None:
+            index_path = '.'.join(self.pdf_path.split('.')[:-1]) + '_index.json'
+        self.sim_model.load_index(index_path)
+
+
+if __name__ == "__main__":
+    import sys
+
+    if len(sys.argv) > 2:
+        gen_model_name_or_path = sys.argv[1]
+    else:
+        print('Usage: python chatpdf.py <gen_model_name_or_path>')
+        gen_model_name_or_path = "THUDM/chatglm-6b-int4"
+    m = ChatPDF(gen_model_name_or_path=gen_model_name_or_path)
+    m.load_pdf_file(pdf_path='sample.pdf')
+    response = m.query('自然语言中的非平行迁移是指什么？')
+    print(response[0])
+    response = m.query('本文作者是谁？')
+    print(response[0])

requirements.txt

@@ -0,0 +1,11 @@
+gradio
+similarities>=1.0.5
+sentencepiece
+textgen
+markdown
+PyPDF2
+python-docx
+pandas
+protobuf
+cpm-kernels
+loguru

webui.py

@@ -0,0 +1,290 @@
+# -*- coding: utf-8 -*-
+"""
+@author:XuMing([email protected])
+@description:
+modified from https://github.com/imClumsyPanda/langchain-ChatGLM/blob/master/webui.py
+"""
+import gradio as gr
+import os
+import shutil
+from loguru import logger
+from chatpdf import ChatPDF
+import hashlib
+from typing import List
+
+pwd_path = os.path.abspath(os.path.dirname(__file__))
+
+CONTENT_DIR = os.path.join(pwd_path, "content")
+logger.info(f"CONTENT_DIR: {CONTENT_DIR}")
+VECTOR_SEARCH_TOP_K = 3
+MAX_INPUT_LEN = 2048
+
+embedding_model_dict = {
+    "text2vec-large": "GanymedeNil/text2vec-large-chinese",
+    "text2vec-base": "shibing624/text2vec-base-chinese",
+    "sentence-transformers": "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2",
+    "ernie-tiny": "nghuyong/ernie-3.0-nano-zh",
+    "ernie-base": "nghuyong/ernie-3.0-base-zh",
+
+}
+
+# supported LLM models
+llm_model_dict = {
+  
+    # "chatglm-6b": "E:\\sdwebui\\image2text_prompt_generator\\models\\chatglm-6b",
+    "chatglm-6b-int4": "THUDM/chatglm-6b-int4",
+    "chatglm-6b": "THUDM/chatglm-6b",
+    "chatglm-6b-int4-qe": "THUDM/chatglm-6b-int4-qe",
+    "llama-7b": "decapoda-research/llama-7b-hf",
+    "llama-13b": "decapoda-research/llama-13b-hf",
+    "t5-lamini-flan-783M": "MBZUAI/LaMini-Flan-T5-783M",
+}
+
+llm_model_dict_list = list(llm_model_dict.keys())
+embedding_model_dict_list = list(embedding_model_dict.keys())
+
+model = None
+
+
+def get_file_list():
+    if not os.path.exists("content"):
+        return []
+    return [f for f in os.listdir("content") if
+            f.endswith(".txt") or f.endswith(".pdf") or f.endswith(".docx") or f.endswith(".md")]
+
+
+def upload_file(file, file_list):
+    if not os.path.exists(CONTENT_DIR):
+        os.mkdir(CONTENT_DIR)
+    filename = os.path.basename(file.name)
+    shutil.move(file.name, os.path.join(CONTENT_DIR, filename))
+    # file_list首位插入新上传的文件
+    file_list.insert(0, filename)
+    return gr.Dropdown.update(choices=file_list, value=filename), file_list
+
+
+def parse_text(text):
+    """copy from https://github.com/GaiZhenbiao/ChuanhuChatGPT/"""
+    lines = text.split("\n")
+    lines = [line for line in lines if line != ""]
+    count = 0
+    for i, line in enumerate(lines):
+        if "```" in line:
+            count += 1
+            items = line.split('`')
+            if count % 2 == 1:
+                lines[i] = f'<pre><code class="language-{items[-1]}">'
+            else:
+                lines[i] = f'<br></code></pre>'
+        else:
+            if i > 0:
+                if count % 2 == 1:
+                    line = line.replace("`", "\`")
+                    line = line.replace("<", "&lt;")
+                    line = line.replace(">", "&gt;")
+                    line = line.replace(" ", "&nbsp;")
+                    line = line.replace("*", "&ast;")
+                    line = line.replace("_", "&lowbar;")
+                    line = line.replace("-", "&#45;")
+                    line = line.replace(".", "&#46;")
+                    line = line.replace("!", "&#33;")
+                    line = line.replace("(", "&#40;")
+                    line = line.replace(")", "&#41;")
+                    line = line.replace("$", "&#36;")
+                lines[i] = "<br>" + line
+    text = "".join(lines)
+    return text
+
+
+def get_answer(
+        query,
+        index_path,
+        history,
+        topn: int = VECTOR_SEARCH_TOP_K,
+        max_input_size: int = 1024,
+        chat_mode: str = "pdf"
+):
+    global model
+
+    if model is None:
+        return [None, "模型还未加载"], query
+    if index_path and chat_mode == "pdf":
+        if not model.sim_model.corpus_embeddings:
+            model.load_index(index_path)
+        response, empty_history, reference_results = model.query(query=query, topn=topn, max_input_size=max_input_size)
+
+        logger.debug(f"query: {query}, response with content: {response}")
+        for i in range(len(reference_results)):
+            r = reference_results[i]
+            response += f"\n{r.strip()}"
+        response = parse_text(response)
+        history = history + [[query, response]]
+    else:
+        # 未加载文件，仅返回生成模型结果
+        response, empty_history = model.chat(query, history)
+        response = parse_text(response)
+        history = history + [[query, response]]
+        logger.debug(f"query: {query}, response: {response}")
+    return history, ""
+
+
+def update_status(history, status):
+    history = history + [[None, status]]
+    logger.info(status)
+    return history
+
+
+def reinit_model(llm_model, embedding_model, history):
+    try:
+        global model
+        if model is not None:
+            del model
+        model = ChatPDF(
+            sim_model_name_or_path=embedding_model_dict.get(
+                embedding_model,
+                "sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2"
+            ),
+            gen_model_type=llm_model.split('-')[0],
+            gen_model_name_or_path=llm_model_dict.get(llm_model, "THUDM/chatglm-6b-int4"),
+            lora_model_name_or_path=None,
+        )
+
+        model_status = """模型已成功重新加载，请选择文件后点击"加载文件"按钮"""
+    except Exception as e:
+        model = None
+        logger.error(e)
+        model_status = """模型未成功重新加载，请重新选择后点击"加载模型"按钮"""
+    return history + [[None, model_status]]
+
+
+def get_file_hash(fpath):
+    return hashlib.md5(open(fpath, 'rb').read()).hexdigest()
+
+
+def get_vector_store(filepath, history, embedding_model):
+    logger.info(filepath, history)
+    index_path = None
+    file_status = ''
+    if model is not None:
+
+        local_file_path = os.path.join(CONTENT_DIR, filepath)
+
+        local_file_hash = get_file_hash(local_file_path)
+        index_file_name = f"{filepath}.{embedding_model}.{local_file_hash}.index.json"
+
+        local_index_path = os.path.join(CONTENT_DIR, index_file_name)
+
+        if os.path.exists(local_index_path):
+            model.load_index(local_index_path)
+            index_path = local_index_path
+            file_status = "文件已成功加载，请开始提问"
+
+        elif os.path.exists(local_file_path):
+            model.load_pdf_file(local_file_path)
+            model.save_index(local_index_path)
+            index_path = local_index_path
+            if index_path:
+                file_status = "文件索引并成功加载，请开始提问"
+            else:
+                file_status = "文件未成功加载，请重新上传文件"
+    else:
+        file_status = "模型未完成加载，请先在加载模型后再导入文件"
+
+    return index_path, history + [[None, file_status]]
+
+
+def reset_chat(chatbot, state):
+    return None, None
+
+
+block_css = """.importantButton {
+    background: linear-gradient(45deg, #7e0570,#5d1c99, #6e00ff) !important;
+    border: none !important;
+}
+.importantButton:hover {
+    background: linear-gradient(45deg, #ff00e0,#8500ff, #6e00ff) !important;
+    border: none !important;
+}"""
+
+webui_title = """
+# 🎉ChatPDF WebUI🎉
+Link in: [https://github.com/zhongpei/ChatPDF](https://github.com/zhongpei/ChatPDF)  Test for MBZUAI/LaMini-Flan-T5-783M
+"""
+
+init_message = """欢迎使用 ChatPDF Web UI，可以直接提问或上传文件后提问 """
+
+with gr.Blocks(css=block_css) as demo:
+    index_path, file_status, model_status = gr.State(""), gr.State(""), gr.State("")
+    file_list = gr.State(get_file_list())
+    gr.Markdown(webui_title)
+    with gr.Row():
+        with gr.Column(scale=2):
+            chatbot = gr.Chatbot([[None, init_message], [None, None]],
+                                 elem_id="chat-box",
+                                 show_label=False).style(height=700)
+            query = gr.Textbox(show_label=False,
+                               placeholder="请输入提问内容，按回车进行提交",
+                               ).style(container=False)
+            clear_btn = gr.Button('🔄Clear!', elem_id='clear').style(full_width=True)
+        with gr.Column(scale=1):
+            llm_model = gr.Radio(llm_model_dict_list,
+                                 label="LLM 模型",
+                                 value=list(llm_model_dict.keys())[0],
+                                 interactive=True)
+            embedding_model = gr.Radio(embedding_model_dict_list,
+                                       label="Embedding 模型",
+                                       value=embedding_model_dict_list[0],
+                                       interactive=True)
+
+            load_model_button = gr.Button("重新加载模型" if model is not None else "加载模型")
+
+            with gr.Row():
+                chat_mode = gr.Radio(choices=["chat", "pdf"], value="pdf", label="聊天模式")
+
+            with gr.Row():
+                topn = gr.Slider(1, 100, 20, step=1, label="最大搜索数量")
+                max_input_size = gr.Slider(512, 4096, MAX_INPUT_LEN, step=10, label="摘要最大长度")
+            with gr.Tab("select"):
+                with gr.Row():
+                    selectFile = gr.Dropdown(
+                        file_list.value,
+                        label="content file",
+                        interactive=True,
+                        value=file_list.value[0] if len(file_list.value) > 0 else None
+                    )
+                    # get_file_list_btn = gr.Button('🔄').style(width=10)
+            with gr.Tab("upload"):
+                file = gr.File(
+                    label="content file",
+                    file_types=['.txt', '.md', '.docx', '.pdf']
+                )
+            load_file_button = gr.Button("加载文件")
+
+    load_model_button.click(
+        reinit_model,
+        show_progress=True,
+        inputs=[llm_model, embedding_model, chatbot],
+        outputs=chatbot
+    )
+    # 将上传的文件保存到content文件夹下,并更新下拉框
+    file.upload(
+        upload_file,
+        inputs=[file, file_list],
+        outputs=[selectFile, file_list]
+    )
+    load_file_button.click(
+        get_vector_store,
+        show_progress=True,
+        inputs=[selectFile, chatbot, embedding_model],
+        outputs=[index_path, chatbot],
+    )
+    query.submit(
+        get_answer,
+        [query, index_path, chatbot, topn, max_input_size, chat_mode],
+        [chatbot, query],
+    )
+    clear_btn.click(reset_chat, [chatbot, query], [chatbot, query])
+
+demo.queue(concurrency_count=3).launch(
+    server_name='0.0.0.0', share=False, inbrowser=False
+)