ChartTable-RAG / summarize.py
KhwaishK's picture
Update summarize.py
cdae2e1 verified
Raw History Blame Contribute Delete
3.25 kB
import time
from langchain_groq import ChatGroq
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
def create_text_and_table_summaries(text_chunks, tables):
prompt_text = """
You are an assistant tasked with summarizing tables and text.
Give a concise summary of the table or text.
Respond only with the summary, no additionnal comment.
Do not start your message by saying "Here is a summary" or anything like that.
Just give the summary as it is.
Table or text chunk: {element}
"""
prompt = ChatPromptTemplate.from_template(prompt_text)
# Summary chain
model = ChatGroq(temperature= 0.5, model="openai/gpt-oss-20b")
summarize_chain = prompt | model | StrOutputParser()
# Sequential processing to prevent 429 errors
text_summaries = []
print(f"⏳ Summarizing {len(text_chunks)} text chunks...")
for chunk in text_chunks:
text_summaries.append(summarize_chain.invoke(chunk))
time.sleep(2.0) # Pause to stay under token limits
table_summaries = []
tables_html = [table.metadata.text_as_html for table in tables]
print(f"⏳ Summarizing {len(tables)} tables...")
for html in tables_html:
table_summaries.append(summarize_chain.invoke(html))
time.sleep(2.0)
# # Summarize text
# text_summaries = summarize_chain.batch(text_chunks, {"max_concurrency": 3})
# # Summarize tables
# tables_html = [table.metadata.text_as_html for table in tables]
# table_summaries = summarize_chain.batch(tables_html, {"max_concurrency": 3})
return text_summaries, table_summaries
def create_image_summaries(images):
prompt_template = """Describe the image in detail. For context,
the image is part of a research paper explaining the transformers
architecture. Be specific about graphs, such as bar plots."""
messages = [
(
"user",
[
{"type": "text", "text": prompt_template},
{
"type": "image_url",
"image_url": {"url": "data:image/jpeg;base64,{image}"},
},
],
)
]
prompt = ChatPromptTemplate.from_messages(messages)
chain = prompt | ChatGroq(model_name="qwen/qwen3.8-27b") | StrOutputParser()
image_summaries = []
print(f"⏳ Analyzing {len(images)} images sequentially...")
for img in images:
try:
# Use invoke() instead of batch() for better control
summary = chain.invoke({"image": img})
image_summaries.append(summary)
time.sleep(3.0) # Vision models need more time to avoid rate limits
except Exception as e:
print(f"⚠️ Image skip: {e}")
image_summaries.append("Image analysis failed.")
return image_summaries
def run_summarization_pipeline(tables, text_chunks, images):
text_sums, table_sums = create_text_and_table_summaries(text_chunks, tables)
image_sums = create_image_summaries(images)
return text_sums, table_sums, image_sums