Skip to content

11 GGUF Neural Network Visualization

Source: notebooks/11-gguf-neural-network-graphistry-vis-executed-e1.ipynb

Notebook focus

This page is a cell-by-cell walkthrough of the notebook, explaining the intent of each step and showing the exact code executed.

Cell-by-cell walkthrough

Cell 1 (Markdown)

11 GGUF Neural Network Visualization

Use GGUF metadata and GraphistryBuilders.embedding_knn() for quick visual exploration of model structure.

What you will learn: - Extract metadata from a GGUF file with gguf_report() - Build a k-nearest-neighbor graph from embedding vectors - Visualize token embedding neighborhoods

Requirements: llamatelemetry installed. A GGUF model dataset for the report.

Cell 2 (Markdown)

1) Install

Cell 3 (Code)

Summary: Installs required dependencies and runtime tools.

!pip -q install git+https://github.com/llamatelemetry/llamatelemetry.git@v0.1.1

Cell 4 (Markdown)

2) GGUF metadata report

Cell 5 (Code)

Summary: Imports core libraries: json, llamatelemetry. Works with GGUF models, quantization, or metadata.

import json
from llamatelemetry.api.gguf import gguf_report

model_path = "/kaggle/input/your-model/model.gguf"

report = gguf_report(model_path)
print(f"Report keys: {list(report.keys())}")
print(json.dumps(report, indent=2, default=str))

Cell 6 (Markdown)

3) Build an embedding kNN graph

embedding_knn() computes pairwise cosine distances and connects each node to its k nearest neighbors.

Cell 7 (Code)

Summary: Imports core libraries: llamatelemetry, numpy. Sets up Graphistry for graph visualization or analytics.

import numpy as np
from llamatelemetry.graphistry import GraphistryBuilders

# Simulated embeddings (replace with real token embeddings)
np.random.seed(42)
embeddings = np.random.randn(20, 16).tolist()
labels = [f"token-{i}" for i in range(20)]

nodes_df, edges_df = GraphistryBuilders.embedding_knn(
    embeddings,
    labels=labels,
    k=3,
    metric="cosine",
)

print(f"Nodes: {len(nodes_df)}")
print(nodes_df.head())
print(f"\nEdges: {len(edges_df)}")
print(edges_df.head())

Cell 8 (Markdown)

4) Inspect neighborhood structure

Cell 9 (Code)

Summary: Executes notebook-specific logic or data processing for this step.

# Show neighbors for a specific token
token_id = "token-0"
neighbors = edges_df[edges_df["source"] == token_id]
print(f"Neighbors of {token_id}:")
print(neighbors.to_string(index=False))

Cell 10 (Markdown)

5) Visualize (optional)

Cell 11 (Code)

Summary: Sets up Graphistry for graph visualization or analytics.

# from llamatelemetry.graphistry import GraphistrySession
# session = GraphistrySession.from_kaggle_secrets()
# g = session.connector.create_graph(edges_df, nodes_df=nodes_df)
# g.plot()