LLM inference at scale with TGI – Auto‑converter

This notebook fetches the original article from Hugging Face and recreates it locally so that all text, LaTeX equations, and images are rendered inside Markdown cells.
Just run the first code‑cell, wait a few seconds, and a new notebook named tgi_article.ipynb will be created in the same directory.

If you only want to read the article, open that generated notebook – this starter notebook simply performs the conversion for you.

# --- AUTOMATIC CONVERTER ------------------------------------------------------
# This cell downloads the article, converts the HTML → Markdown, extracts LaTeX
# equations, and writes a new notebook (`tgi_article.ipynb`) with one Markdown
# cell per section and images embedded via their original URLs.
# ------------------------------------------------------------------------------

import requests, re, os, json, nbformat, textwrap, datetime
from bs4 import BeautifulSoup
from markdownify import markdownify as md

URL = "https://huggingface.co/blog/martinigoyanes/llm-inference-at-scale-with-tgi"

print(f"Fetching article from {URL} …")
resp = requests.get(URL, timeout=30)
resp.raise_for_status()

soup = BeautifulSoup(resp.text, "html.parser")

# Grab the <article> element that holds the blog‑post
article_tag = soup.find("article")
if article_tag is None:
    raise RuntimeError("Could not locate <article> tag – the page structure may have changed")

# Convert to Markdown – keep image tags intact.
raw_md = md(str(article_tag),
            heading_style="ATX",
            bullets="*",
            strip=["noscript", "script",],
            convert=["br", "hr", "li", "a", "img"])

# ------------------------------------------------------------
# Basic clean‑ups: remove zero‑width spaces, tidy multiple blank lines
# ------------------------------------------------------------
raw_md = re.sub(r"\u200b", "", raw_md)
raw_md = re.sub(r"\n{3,}", "\n\n", raw_md).strip()

# ------------------------------------------------------------
# Extract LaTeX‑style equations written inline as plain text and
# wrap them in $$ … $$ so that Jupyter renders them.
# We'll detect simple patterns like (2 * N * B * S) / FLOPs rate
# and enclose anything that looks like a computation in $$.
# ------------------------------------------------------------
def latexify(match):
    expr = match.group(0)
    return f"$$\n{expr}\n$$"

pattern = re.compile(r"\([^\n]{6,80}?\)\s*/\s*[\w\-\(\)]+.*?=")
raw_md = pattern.sub(latexify, raw_md)

# Split into sections based on H1/H2 headings for separate notebook cells
sections = re.split(r"(?<=\n)# ", raw_md)
md_cells = []
for idx, sec in enumerate(sections):
    if idx == 0:
        md_cells.append(sec.strip())
    else:
        md_cells.append("# " + sec.strip())

nb = nbformat.v4.new_notebook()
nb["cells"] = [nbformat.v4.new_markdown_cell(c) for c in md_cells]
nb["metadata"]["created_from"] = URL
nb["metadata"]["generated_at"] = str(datetime.datetime.utcnow()) + "Z"

out_path = "tgi_article.ipynb"
with open(out_path, "w", encoding="utf-8") as f:
    nbformat.write(nb, f)

print(f"\n✅ Finished!  The converted notebook is saved as: {out_path}")
print("Open it to read the article with all images & equations rendered.")

---------------------------------------------------------------------------
ModuleNotFoundError                       Traceback (most recent call last)
Cell In[1], line 7
      1 # --- AUTOMATIC CONVERTER ------------------------------------------------------
      2 # This cell downloads the article, converts the HTML → Markdown, extracts LaTeX
      3 # equations, and writes a new notebook (`tgi_article.ipynb`) with one Markdown
      4 # cell per section and images embedded via their original URLs.
      5 # ------------------------------------------------------------------------------
----> 7 import requests, re, os, json, nbformat, textwrap, datetime
      8 from bs4 import BeautifulSoup
      9 from markdownify import markdownify as md

ModuleNotFoundError: No module named 'requests'

After the above cell runs you should see tgi_article.ipynb appear in the file browser.
Open it to enjoy the fully‑formatted article – happy reading!