Hacker and Geeker's Way

import re

mmd_path = "attention_is_all_you_need.mmd"
# 打开文件并读取内容
with open(mmd_path, "r") as file:
    content = file.read()

# 使用正则表达式匹配表格内容和表格后一行内容
pattern = r"\\begin{table}(.*?)\\end{table}\n(.*?)\n"
matches = re.findall(pattern, content, re.DOTALL)

tables = []
# 添加匹配结果
for match in matches:
    tables.append(f"{match[0]}{match[1]}")

from llama_index.core import VectorStoreIndex
from llama_index.core.schema import TextNode

question = "when layer type is Convolutional, what is the Maximum Path Length?"
print(f"question: {question}")

nodes = [TextNode(text=t) for t in tables]
vector_index = VectorStoreIndex(nodes)
query_engine = vector_index.as_query_engine(similarity_top_k=2)
response = query_engine.query(question)
print(f"answer: {response}")
print("Source nodes: ")
for node in response.source_nodes:
    print(f"node text: {node.text}")

from apryse_sdk import *

PDFNet.Initialize("your_trial_key")

file_name = "demo"
input_filename = f"{file_name}.pdf"
output_dir = "output"

PDFNet.AddResourceSearchPath("./Lib")

htmlOutputOptions = HTMLOutputOptions()
htmlOutputOptions.SetContentReflowSetting(HTMLOutputOptions.e_reflow_full)

Convert.ToHtml(input_filename, f"{output_dir}/{file_name}.html", htmlOutputOptions)

import os
import pickle
from pathlib import Path
from llama_index.readers.file import FlatReader
from llama_index.core.node_parser import UnstructuredElementNodeParser

reader = FlatReader()
demo_file = reader.load_data(Path("demo.html"))
node_parser = UnstructuredElementNodeParser()

pkl_file = "demo.pkl"
if not os.path.exists(pkl_file):
    raw_nodes = node_parser.get_nodes_from_documents(demo_file)
    pickle.dump(raw_nodes, open(pkl_file, "wb"))
else:
    raw_nodes = pickle.load(open(pkl_file, "rb"))

base_nodes, node_mappings = node_parser.get_base_nodes_and_mappings(raw_nodes)

from llama_index.core.schema import IndexNode, TextNode

example_index_nodes = [b for b in base_nodes if isinstance(b, IndexNode)]
example_index_node = example_index_nodes[1]
print(
    f"\n--------\n{example_index_node.get_content(metadata_mode='all')}\n--------\n"
)
print(f"\n--------\nIndex ID: {example_index_node.index_id}\n--------\n")
print(
    f"\n--------\n{node_mappings[example_index_node.index_id].get_content()}\n--------\n"
)

from llama_index.core import VectorStoreIndex
from llama_index.core.retrievers import RecursiveRetriever
from llama_index.core.query_engine import RetrieverQueryEngine

vector_index = VectorStoreIndex(base_nodes)
vector_retriever = vector_index.as_retriever(similarity_top_k=1)
vector_query_engine = vector_index.as_query_engine(similarity_top_k=1)

recursive_retriever = RecursiveRetriever(
    "vector",
    retriever_dict={"vector": vector_retriever},
    node_dict=node_mappings,
    verbose=True,
)
query_engine = RetrieverQueryEngine.from_args(recursive_retriever)
question = "In the comparison of performance metrics for different AI models across various tasks. What is the performance metric of the model 'Qwen-VL-Plus' in task 'MMMU'? Tell me the exact number."
response = query_engine.query(question)
print(f"answer: {str(response)}")

# 显示结果
Retrieving with query id None: In the comparison of performance metrics for different AI models across various tasks. What is the performance metric of the model 'Qwen-VL-Plus' in task 'MMMU'? Tell me the exact number.
Retrieved node with id, entering: 41edc9a6-30ed-44cf-967e-685f7dfce8df
Retrieving with query id 41edc9a6-30ed-44cf-967e-685f7dfce8df: In the comparison of performance metrics for different AI models across various tasks. What is the performance metric of the model 'Qwen-VL-Plus' in task 'MMMU'? Tell me the exact number.

answer: The performance metric of the model 'Qwen-VL-Plus' in the task 'MMMU' is 45.2%.

models = [
    "Other BestOpen-source LVLM",
    "Gemini Pro",
    "Gemini Ultra",
    "GPT-4V",
    "Qwen-VL-Plus",
    "Qwen-VL-Max",
]
metrics = ["DocVQA", "ChartQA", "AI2D", "TextVQA", "MMMU", "MathVista", "MM-Bench-CN"]
questions = []
for model in models:
    for metric in metrics:
        questions.append(
            f"In the comparison of performance metrics for different AI models across various tasks. What is the performance metric of the model '{model}' in task '{metric}'? Tell me the exact number."
        )

actual_metrics = [
    81.6, 68.4, 73.7, 76.1, 45.9, 36.7, 72.4,
    88.1, 74.1, 73.9, 74.6, 47.9, 45.2, 74.3,
    90.9, 80.8, 75.9, 82.3, 59.4, 53, 0,
    88.4, 78.5, 78.2, 78, 56.8, 49.9, 73.9,
    91.4, 78.1, 75.9, 78.9, 45.2, 43.3, 68,
    93.1, 79.8, 79.3, 79.5, 51.4, 51, 75.1,
]

actual_answers = dict(zip(questions, actual_metrics))

result = {}
for q in questions:
  response = query_engine.query(q)
  answer = str(response)
  result[q] = str(actual_answers[q]) in answer
  print(f"question: {q}\nresponse: {answer}\nactual:{actual_answers[q]}\nresult:{result[q]}\n\n")

# 计算准确率
correct = sum(result.values())
total = len(result)
print(f"Percentage of True values: {correct / total * 100}%")

from llama_parse import LlamaParse

parser_gpt4o = LlamaParse(
    result_type="markdown",
    api_key="<llama_parse_api_key>",
    gpt4o_mode=True,
    gpt4o_api_key="<openai_api_key>"
)

pdf_file = "demo.pdf"
pkl_file = "demo.pkl"
if not os.path.exists(pkl_file):
    documents_gpt4o = parser_gpt4o.load_data(pdf_file)
    pickle.dump(documents_gpt4o, open(pkl_file, "wb"))
else:
    documents_gpt4o = pickle.load(open(pkl_file, "rb"))

def get_nodes(docs):
    """Split docs into nodes, by separator."""
    nodes = []
    for doc in docs:
        doc_chunks = doc.text.split("\n---\n")
        for doc_chunk in doc_chunks:
            node = TextNode(
                text=doc_chunk,
                metadata=deepcopy(doc.metadata),
            )
            nodes.append(node)

    return nodes

nodes = get_nodes(documents_gpt4o)
vector_index = VectorStoreIndex(nodes)
query_engine = vector_index.as_query_engine(similarity_top_k=2)
question = "In the comparison of performance metrics for different AI models across various tasks. What is the performance metric of the model 'Qwen-VL-Plus' in task 'MMMU'? Tell me the exact number."
response = query_engine.query(question)
print(f"answer: {str(response)}")

# 显示结果
answer: The performance metric of the model 'Qwen-VL-Plus' in the task 'MMMU' is 45.2%.