attentionmech

# this needs a sqlite dump from nsight nsys utility
# for example on windows i ran this:
# "C:\Program Files\NVIDIA Corporation\Nsight Systems 2025.1.3\target-windows-x64\nsys.exe" profile --trace=cuda,nvtx --export sqlite --output profile_matmul you_cuda_program.exe
# once that dump is there run the follow python code on it basically it takes few key tables and animate a timeline

import sqlite3
import numpy as np
import matplotlib.pyplot as plt
from matplotlib.patches import Patch
from mpl_toolkits.mplot3d import Axes3D

attentionmech

Part I: Algebraic Structures & Topological Foundations

Group

• Analogy: Like a dance routine: steps can be combined, there's a neutral pose, and every move has an undo.
• Joke: Why did the group go to therapy? It had identity issues.
• Definition: A set $G$ with operation $*$ satisfying associativity, identity, and inverses.

Subgroup

attentionmech

Part I: Algebraic Structures & Topological Foundations

Groups and Group Theory

• Group: Abstract set with a binary operation satisfying associativity, identity, and invertibility. Captures symmetry and structure.
• Subgroup: A subset that is itself a group under the same operation.
• Normal Subgroup: Invariant under conjugation; key to forming quotient groups.
• Quotient Group: The result of identifying elements differing by a normal subgroup; often used in symmetry reduction.
• Group Homomorphism & Isomorphism: Structure-preserving maps between groups; isomorphisms show structural equivalence.
• Group Action: A group acting on a set reveals symmetry in geometry and algebra.

attentionmech

import numpy as np
import matplotlib.pyplot as plt
from sklearn.decomposition import PCA
from itertools import product

def generate_4d_honeycomb(grid_size=4, spacing=2.0):
    """Generate a 4D hexagonal-like lattice (offset in x/y, uniform in z/w)."""
    points = []
    edges = []
    point_index = {}

attentionmech

import torch
import numpy as np
import matplotlib.pyplot as plt
from matplotlib import colormaps
from matplotlib.animation import FuncAnimation
from mpl_toolkits.mplot3d import Axes3D
from transformers import GPT2LMHeadModel, GPT2Tokenizer
from sklearn.decomposition import PCA

# === CONFIG ===

attentionmech

import torch
import argparse
import numpy as np
import matplotlib.pyplot as plt
from matplotlib import cm
from matplotlib import colormaps
from mpl_toolkits.mplot3d import Axes3D
from transformers import GPT2LMHeadModel, GPT2Tokenizer
from sklearn.decomposition import PCA
from sklearn.manifold import TSNE

attentionmech

import numpy as np
import matplotlib.pyplot as plt
import os

# --- GELU Activation ---
def gelu(x):
    return 0.5 * x * (1 + np.tanh(np.sqrt(2 / np.pi) * (x + 0.044715 * x**3)))

def gelu_derivative(x):
    tanh_term = np.tanh(np.sqrt(2 / np.pi) * (x + 0.044715 * x**3))

attentionmech

import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d.art3d import Poly3DCollection
import matplotlib.colors as mcolors

# Timeline JSON-style data
timeline = [
    {"step": "kernel_launch", "objects": {"threads": 4096, "blocks": 16, "warps": 128}},
    {"step": "index_calculation", "objects": {"row_index": 4096, "col_index": 4096}},
    {"step": "tile_phase_0_shared_memory_load", "objects": {"global_reads": 8192, "shared_stores": 8192}},
    {"step": "tile_phase_0_sync_1", "objects": {"sync_barriers": 16}},

attentionmech

from nnsight import LanguageModel
import torch
import argparse
import os
import sys
import time

def clear_terminal():
    os.system('cls' if os.name == 'nt' else 'clear')

attentionmech