corpus_for_tokenisation.py

# -*- coding: utf-8 -*-
"""Corpus for tokenisation.ipynb

Automatically generated by Colab.

Original file is located at
    https://colab.research.google.com/drive/1_maItYuOpWMe8YAaz4nzcKT32Rvye144
"""

continue_seamless = True
# keep this true to run the whole "Tokenisation Whole (Experimenation)" block at once
# run one of 0.5b, or 1b block

"""# Tokenisation Whole (Experimenation) - 0.5b"""

!pip install zstandard

!pip install datasets

from huggingface_hub import hf_hub_download
from huggingface_hub import list_repo_files
from datasets import load_dataset

# Commented out IPython magic to ensure Python compatibility.
!mkdir SlimPajama-627B
# %cd SlimPajama-627B
!mkdir validation

# Commented out IPython magic to ensure Python compatibility.
# %cd validation

!git init
!git remote add -f origin https://huggingface.co/datasets/cerebras/SlimPajama-627B
!git config core.sparseCheckout true
!echo "validation/*" >> .git/info/sparse-checkout
!git pull origin main

# Commented out IPython magic to ensure Python compatibility.
# %cd /content/SlimPajama-627B/validation/validation

!pwd

!pip install chardet

import os
import json
import chardet
from tqdm.auto import tqdm
from collections import Counter

!ls -lh /content/SlimPajama-627B/validation/validation

!ls -lh /content/SlimPajama-627B/validation/validation/chunk1

!ls -l /content/SlimPajama-627B/validation/validation/chunk1 | grep -v '^d' | wc -l

import os
import json
import zstandard as zstd
from tqdm import tqdm

# Define the path to the validation dataset
dataset_dir = "/content/SlimPajama-627B/validation/validation"

# Initialize an empty list to store the extracted text
corpus = []

# Function to read .jsonl.zst files
def read_jsonl_zst(file_path):
    with open(file_path, "rb") as f:
        dctx = zstd.ZstdDecompressor()
        with dctx.stream_reader(f) as reader:
            decompressed_data = reader.read()  # Read all decompressed data
            for line in decompressed_data.splitlines():  # Split into lines
                try:
                    data = json.loads(line.decode("utf-8"))  # Decode and parse JSON
                    if isinstance(data, dict) and "text" in data:
                        yield data["text"]
                except json.JSONDecodeError:
                    print(f"Skipping malformed JSON in {file_path}")

# Recursively iterate over all files in the dataset directory
for root, _, files in tqdm(os.walk(dataset_dir)):
    for file in tqdm(files):
        file_path = os.path.join(root, file)

        # Process only .jsonl.zst files
        if file.endswith(".jsonl.zst"):
            corpus.extend(read_jsonl_zst(file_path))

# The corpus now contains all extracted text
print(f"Extracted {len(corpus)} text entries.")

corpus[0]

corpus[1][-50:]

corpus[2]

corpus[len(corpus)//2][-50:]

corpus[len(corpus)//2+1]

corpus[len(corpus)//2+2]

len(corpus)

first_half_corpus = " ".join(corpus[:len(corpus)//2])
second_half_corpus = " ".join(corpus[len(corpus)//2:])

net_corpus = " ".join(corpus)
# net_corpus = first_half_corpus + " " + second_half_corpus

net_corpus[:500]

len(net_corpus)

!pwd

# Commented out IPython magic to ensure Python compatibility.
# %cd /content

with open("whole_corpus_0.5b_val.txt", "w") as file:
    file.write(net_corpus)

with open("first_half_corpus_0.5b_val.txt", "w") as file:
    file.write(first_half_corpus)

with open("second_half_corpus_0.5b_val.txt", "w") as file:
    file.write(second_half_corpus)

with open("whole_corpus_ind_texts_0.5b_val.json", "w") as file:
    json.dump(corpus, file, indent=4)

with open("first_half_corpus_ind_texts_0.5b_val.json", "w") as file:
    json.dump(corpus[:len(corpus)//2], file, indent=4)

with open("second_half_corpus_ind_texts_0.5b_val.json", "w") as file:
    json.dump(corpus[len(corpus)//2:], file, indent=4)

# Commented out IPython magic to ensure Python compatibility.
# %cd /content/SlimPajama-627B/validation/validation

char_counts = Counter(net_corpus)

len(char_counts)

sorted_char_counts = dict(sorted(char_counts.items(), key=lambda item: item[1], reverse=True))

total_char_occ = 0
for i in sorted_char_counts:
  total_char_occ += sorted_char_counts[i]
total_char_occ

sorted_char_counts[' ']

sorted_char_counts

len(net_corpus)

import regex
import re

def split_text_into_words(text):
    # Unicode-aware splitting: removes punctuation, spaces, and special characters
    words = regex.split(r'[^\p{L}\p{N}_]+', text)

    # Remove empty strings
    return [word for word in words if word]

sp_ch_sep_toks = split_text_into_words(net_corpus)
len(sp_ch_sep_toks)

word_counts = Counter(sp_ch_sep_toks)

len(word_counts)

word_counts = word_counts | sorted_char_counts

len(word_counts)

word_counts

sorted_word_counts = dict(sorted(word_counts.items(), key=lambda x: x[1], reverse=True))
sorted_word_counts

max_len = 0
for word in word_counts:
  if len(word) > max_len:
    max_len = len(word)
print(max_len)

len(sorted_word_counts)

word_len_count_map = {}
for word in word_counts:
  if len(word) not in word_len_count_map:
    word_len_count_map[len(word)] = word_counts[word]
  else:
    word_len_count_map[len(word)] += word_counts[word]

# Sort in descending order based on the second element
sorted_word_len_count_map = dict(sorted(word_len_count_map.items(), key=lambda x: x[1], reverse=True))
sorted_word_len_count_map

len(sorted_word_len_count_map)

sum = 0
for i in sorted_word_len_count_map:
  sum += sorted_word_len_count_map[i]
sum

lengths = {}
for word in list(sorted_word_counts.keys())[:131_072]:
  if len(word) not in lengths:
    lengths[len(word)] = [sorted_word_counts[word],sorted_word_counts[word],1]
  else:
    lengths[len(word)][0] += sorted_word_counts[word]
    lengths[len(word)][2] += 1
    if sorted_word_counts[word] > lengths[len(word)][1]:
      lengths[len(word)][1] = sorted_word_counts[word]
lengths

len(lengths)

sorted_lengths_max_ind_tok_counts = dict(sorted(lengths.items(), key=lambda item: item[1][1], reverse=True))
sorted_lengths_max_ind_tok_counts

from collections import defaultdict

def count_char_ngrams_in_words(word_list, max_n=14):
    """
    Count character-level n-grams for each word in a list.

    Args:
        word_list (list): List of words to analyze
        max_n (int): Maximum size of n-grams to count (default: 14)

    Returns:
        dict: Nested dictionary with n as outer key and
              inner dictionaries containing n-grams and their counts
    """
    # Initialize nested defaultdict to store counts
    ngram_counts = defaultdict(lambda: defaultdict(int))

    # Process each word separately
    for word in tqdm(word_list):
        word_length = len(word)

        # For each possible n-gram size (from 1 to max_n or word length)
        for n in range(2, min(max_n + 1, word_length)):
            # Use sliding window within each word
            for i in range(word_length - n + 1):
                # Extract the character n-gram
                ngram = word[i:i+n]

                # Increment the count for this n-gram
                ngram_counts[n][ngram] += 1

    # Convert defaultdict to regular dict for return
    return {n: dict(counts) for n, counts in ngram_counts.items()}

n_gram = count_char_ngrams_in_words(sp_ch_sep_toks,max_n=15)

threshold = 56_515
count_above_threshold = 0
for n in n_gram:
  for i in n_gram[n]:
    if n_gram[n][i] > threshold:
      count_above_threshold += 1
count_above_threshold

threshold = 141_993
count_above_threshold = 0
for n in n_gram:
  for i in n_gram[n]:
    if n_gram[n][i] > threshold:
      count_above_threshold += 1
count_above_threshold

t_sum_below = 0
for i in sorted_lengths_max_ind_tok_counts:
  if sorted_lengths_max_ind_tok_counts[i][1] < threshold:
    t_sum_below += sorted_lengths_max_ind_tok_counts[i][2]
t_sum_below

t_sum_below

n_gram_tok = {}
for n in tqdm(n_gram):
  for i in n_gram[n]:
    n_gram_tok[i] = n_gram[n][i]
n_gram_tok

sorted_n_gram_tok_counts = dict(sorted(n_gram_tok.items(), key=lambda item: item[1], reverse=True))
sorted_n_gram_tok_counts

sorted_word_counts

n_gram_lengths = {}
for i in list(sorted_n_gram_tok_counts.keys())[:131_072]:
    if len(i) not in n_gram_lengths:
      n_gram_lengths[len(i)] = [sorted_n_gram_tok_counts[i],sorted_n_gram_tok_counts[i],1]
    else:
      n_gram_lengths[len(i)][0] += sorted_n_gram_tok_counts[i]
      n_gram_lengths[len(i)][2] += 1
      if sorted_n_gram_tok_counts[i] > n_gram_lengths[len(i)][1]:
        n_gram_lengths[len(i)][1] = sorted_n_gram_tok_counts[i]
n_gram_lengths

sorted_n_gram_lengths_tok_counts = dict(sorted(n_gram_lengths.items(), key=lambda item: item[1][1], reverse=True))
sorted_n_gram_lengths_tok_counts

sorted_lengths_max_ind_tok_counts

import regex
from collections import Counter

def count_special_char_ngrams(text, max_n):
    """
    Count occurrences of character-level ngrams (only special characters, excluding whitespace) with n > 1.

    Parameters:
        text (str): The input text.
        max_n (int): The highest n-gram length to consider.

    Returns:
        dict: A mapping of special character ngrams to their counts.
    """
    # Pattern to match sequences of special characters (not letters, digits, underscore, or whitespace)
    special_pattern = r'[^\p{L}\p{N}_\s]+'

    counts = Counter()

    # Iterate over contiguous blocks of non-whitespace special characters
    for match in tqdm(regex.finditer(special_pattern, text)):
        block = match.group()
        L = len(block)

        # Generate ngrams for lengths from 2 to max_n (or block length, whichever is smaller)
        for n in range(2, min(L, max_n) + 1):
            for i in range(L - n + 1):
                ngram = block[i:i+n]
                counts[ngram] += 1

    return dict(counts)

sp_tok_n_grams = count_special_char_ngrams(net_corpus, 1000)
sp_tok_n_grams

sorted_sp_tok_n_grams = dict(sorted(sp_tok_n_grams.items(), key=lambda x: x[1], reverse=True))
sorted_sp_tok_n_grams

len(sorted_sp_tok_n_grams)

sorted_sp_tok_n_grams_len_count_map = {}
for sp_tok in sorted_sp_tok_n_grams:
  if len(sp_tok) not in sorted_sp_tok_n_grams_len_count_map:
    sorted_sp_tok_n_grams_len_count_map[len(sp_tok)] = sorted_sp_tok_n_grams[sp_tok]
  else:
    sorted_sp_tok_n_grams_len_count_map[len(sp_tok)] += sorted_sp_tok_n_grams[sp_tok]
# sorted_sp_tok_n_grams_len_count_map

# Sort in descending order based on the second element
sorted_sorted_sp_tok_n_grams_len_count_map = dict(sorted(sorted_sp_tok_n_grams_len_count_map.items(), key=lambda x: x[1], reverse=True))
sorted_sorted_sp_tok_n_grams_len_count_map

len(sorted_sp_tok_n_grams_len_count_map)

sum = 0
for i in sorted_sp_tok_n_grams_len_count_map:
  sum += sorted_sp_tok_n_grams_len_count_map[i]
sum

sp_token_ngram_lengths = {}
for sp_tok in list(sorted_sp_tok_n_grams.keys())[:131_072]:
  if len(sp_tok) not in sp_token_ngram_lengths:
    sp_token_ngram_lengths[len(sp_tok)] = [sorted_sp_tok_n_grams[sp_tok],sorted_sp_tok_n_grams[sp_tok],1]
  else:
    sp_token_ngram_lengths[len(sp_tok)][0] += sorted_sp_tok_n_grams[sp_tok]
    sp_token_ngram_lengths[len(sp_tok)][2] += 1
    if sorted_sp_tok_n_grams[sp_tok] > sp_token_ngram_lengths[len(sp_tok)][1]:
      sp_token_ngram_lengths[len(sp_tok)][1] = sorted_sp_tok_n_grams[sp_tok]
sp_token_ngram_lengths

def max_contiguous_occurrences(corpus):
    # Initialize a dictionary to store the maximum run length for each character
    max_runs = {}

    # Start tracking the current character and run length
    current_char = corpus[0]
    current_run = 1

    # Iterate over the corpus starting from the second character
    for ch in tqdm(corpus[1:]):
        if ch == current_char:
            # Increase the run if the same character continues
            current_run += 1
        else:
            if current_char not in max_runs:
                max_runs[current_char] = current_run
            else:
                max_runs[current_char] = max(max_runs[current_char], current_run)
            # Reset current_char and current_run for the new character
            current_char = ch
            current_run = 1

    if current_char not in max_runs:
        max_runs[current_char] = current_run
    else:
        max_runs[current_char] = max(max_runs[current_char], current_run)

    return max_runs

unique_char_list = sorted_char_counts.keys()

max_occ_map = max_contiguous_occurrences(net_corpus)

# Sort in descending order based on the second element
sorted_max_occ_map = dict(sorted(max_occ_map.items(), key=lambda x: x[1], reverse=True))
sorted_max_occ_map

sorted_word_counts

sorted_n_gram_tok_counts

sorted_sp_tok_n_grams

len(sorted_word_counts)

len(sorted_n_gram_tok_counts)

3809463 + 49495136

len(sorted_sp_tok_n_grams)

for i in tqdm(sorted_word_counts):
  if i in sorted_sp_tok_n_grams:
    print(i)

for i in tqdm(sorted_n_gram_tok_counts):
  if i in sorted_sp_tok_n_grams:
    print(i)

for i in tqdm(sorted_sp_tok_n_grams):
  if i in sorted_word_counts:
    print(i)

  if i in sorted_n_gram_tok_counts:
    print(i)

for i in tqdm(sorted_word_counts):
  if i not in sorted_n_gram_tok_counts:
    sorted_n_gram_tok_counts[i] = sorted_word_counts[i]
    continue

  sorted_n_gram_tok_counts[i] += sorted_word_counts[i]

sorted_n_gram_tok_counts = dict(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1], reverse=True))

sorted_n_gram_tok_counts

len(sorted_n_gram_tok_counts)

for i in tqdm(sorted_n_gram_tok_counts):
  if i in sorted_sp_tok_n_grams:
    print(i)

for i in tqdm(sorted_sp_tok_n_grams):
  if i in sorted_n_gram_tok_counts:
    print(i)

for i in list(sorted_n_gram_tok_counts.keys())[:131_072]:
  if len(set(list(i))) == 1 and len(i) > 3:
    print(i, len(i))

n_gram_lengths = {}
for i in list(sorted_n_gram_tok_counts.keys())[:131_072]:
    if len(i) not in n_gram_lengths:
      n_gram_lengths[len(i)] = [sorted_n_gram_tok_counts[i],sorted_n_gram_tok_counts[i],1]
    else:
      n_gram_lengths[len(i)][0] += sorted_n_gram_tok_counts[i]
      n_gram_lengths[len(i)][2] += 1
      if sorted_n_gram_tok_counts[i] > n_gram_lengths[len(i)][1]:
        n_gram_lengths[len(i)][1] = sorted_n_gram_tok_counts[i]
n_gram_lengths

for word in tqdm(sorted_n_gram_tok_counts):
  if word.strip(' \n').isnumeric() and sorted_n_gram_tok_counts[word] > 56_000:
      print(f"{word}:{sorted_word_counts[word]}")

len(sorted_n_gram_tok_counts)

len(sorted_sp_tok_n_grams)

52307513 + 696425

sorted_n_gram_tok_counts = dict(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1], reverse=True))

sorted_n_gram_tok_counts = sorted_n_gram_tok_counts | sorted_sp_tok_n_grams
len(sorted_n_gram_tok_counts)

for i in list(sorted_sp_tok_n_grams.keys())[:131_072]:
  if len(set(list(i))) == 1 and len(i) > 100:
    print(i[:5], len(i), sorted_sp_tok_n_grams[i], len(i)*sorted_sp_tok_n_grams[i])

top_tokens = {}

count = 0
c = 0
s = [0,0,0]
for i in list(sorted_n_gram_tok_counts.keys()):
  if count >= 131_072:
      break
  if i in sorted_sp_tok_n_grams:
    c += 1
  if len(set(list(i))) * 3 < len(i):
    if len(set(list(i))) > 1:
      print(i, sorted_n_gram_tok_counts[i])
    s[0] += 1
    if len(i) > 15:
      s[1] += 1
      if len(i) > 50:
        s[2] += 1
  else:
    count += 1
c, count, s

sorted_n_gram_tok_counts_anti_length = dict(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(len(x[0])), reverse=True))

max_length = 0
char_count = 0
for i in list(sorted_n_gram_tok_counts_anti_length.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    # print(i)
    char_count += 1
char_count, max_length

sorted_n_gram_tok_counts_anti_length_sq = dict(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(len(x[0])**2), reverse=True))

max_length = 0
s = [0,0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sq.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length

s = [0,0,0]
s2 = [0,0]
max_length = 0
for i in list(sorted_n_gram_tok_counts_anti_length_sq.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

sorted_n_gram_tok_counts_anti_length_cube = dict(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(len(x[0])**3), reverse=True))

max_length = 0
s = [0,0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_cube.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length

s = [0,0,0]
s2 = [0,0]
max_length = 0
for i in list(sorted_n_gram_tok_counts_anti_length_cube.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

sorted_n_gram_tok_counts_anti_length_quad = dict(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(len(x[0])**4), reverse=True))

max_length = 0
s = [0,0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_quad.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length

s = [0,0,0]
s2 = [0,0]
max_length = 0
for i in list(sorted_n_gram_tok_counts_anti_length_quad.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

# shows one minute but takes three minutes for some reason
sorted_n_gram_tok_counts_anti_length_cube_smart = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)* list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0])))))), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_cube_smart.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
max_length = 0
for i in list(sorted_n_gram_tok_counts_anti_length_cube_smart.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

# the three factors but in sq terms, taken 2/3 power
sorted_n_gram_tok_counts_anti_length_sq_smart = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)* list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0]))))))**(2/3), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sq_smart.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
max_length = 0
for i in list(sorted_n_gram_tok_counts_anti_length_sq_smart.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

def longest_same_char_sequence(s: str) -> int:
    if not s:
        return 0

    max_length = 1
    current_length = 1

    for i in range(1, len(s)):
        if s[i] == s[i - 1]:
            current_length += 1
        else:
            max_length = max(max_length, current_length)
            current_length = 1

    return max(max_length, current_length)

longest_same_char_sequence('\\\\\\\\\\\\\\\\\\\\\\')

sorted_n_gram_tok_counts_anti_length_cube_smart_2 = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)*longest_same_char_sequence(x[0])*list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0]))))))**(3/4), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_cube_smart_2.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
max_length = 0
for i in list(sorted_n_gram_tok_counts_anti_length_cube_smart_2.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_sq_smart_2 = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)*longest_same_char_sequence(x[0])*list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0]))))))**(2/4), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_sq_smart_2.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
max_length = 0
for i in list(sorted_n_gram_tok_counts_anti_length_sq_smart_2.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

class SuffixAutomaton:
    def __init__(self, s):
        self.states = [{'len': 0, 'link': -1, 'next': {}, 'occ': 0}]
        self.last = 0  # index of the last (active) state
        for ch in s:
            self.add_char(ch)
        self.propagate_occurrences()

    def add_char(self, ch):
        p = self.last
        curr = len(self.states)
        # Create a new state with occ = 1 for the new ending character.
        self.states.append({'len': self.states[p]['len'] + 1, 'link': 0, 'next': {}, 'occ': 1})
        while p != -1 and ch not in self.states[p]['next']:
            self.states[p]['next'][ch] = curr
            p = self.states[p]['link']
        if p == -1:
            self.states[curr]['link'] = 0
        else:
            q = self.states[p]['next'][ch]
            if self.states[p]['len'] + 1 == self.states[q]['len']:
                self.states[curr]['link'] = q
            else:
                clone = len(self.states)
                self.states.append({
                    'len': self.states[p]['len'] + 1,
                    'next': self.states[q]['next'].copy(),
                    'link': self.states[q]['link'],
                    'occ': 0  # clone does not add occurrences by itself
                })
                while p != -1 and self.states[p]['next'].get(ch) == q:
                    self.states[p]['next'][ch] = clone
                    p = self.states[p]['link']
                self.states[q]['link'] = self.states[curr]['link'] = clone
        self.last = curr

    def propagate_occurrences(self):
        # Process states in order of decreasing length
        order = sorted(range(len(self.states)), key=lambda i: self.states[i]['len'], reverse=True)
        for i in order:
            link = self.states[i]['link']
            if link != -1:
                self.states[link]['occ'] += self.states[i]['occ']

def max_substring_value(s):
    n = len(s)
    if n < 2:
        return 1  # ensure minimum return value is 1, even if s is too short for valid substrings
    sa = SuffixAutomaton(s)
    max_val = 0
    # For each state with at least length 2, use the effective length which is capped at n-1.
    for state in sa.states:
        if state['len'] >= 2:
            effective_length = min(state['len'], n - 1)
            product = state['occ'] * effective_length
            if product > max_val:
                max_val = product
    return max_val if max_val > 0 else 1

sorted_n_gram_tok_counts_anti_length_net_smart_3 = {}
for i in tqdm(sorted_n_gram_tok_counts):
  sorted_n_gram_tok_counts_anti_length_net_smart_3[i] = [sorted_n_gram_tok_counts[i], (((((((len(i)**2)*longest_same_char_sequence(i)*max_substring_value(i)*list(Counter(list(i)).items())[0][1]))))/(len(set(list(i))))))]

sorted_n_gram_tok_counts_anti_length_cube_smart_3 = dict(tqdm(sorted(sorted_n_gram_tok_counts_anti_length_net_smart_3.items(), key=lambda x: x[1][0]/(x[1][1]**(3/5)), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_cube_smart_3.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_cube_smart_3.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_cube_smart_2.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_sq_smart_3 = dict(tqdm(sorted(sorted_n_gram_tok_counts_anti_length_net_smart_3.items(), key=lambda x: x[1][0]/(x[1][1]**(2/5)), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_sq_smart_3.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sq_smart_3.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sq_smart_2.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_ac_cube_smart_3 = dict(tqdm(sorted(sorted_n_gram_tok_counts_anti_length_net_smart_3.items(), key=lambda x: x[1][0]/(x[1][1]**(3/6)), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_ac_cube_smart_3.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_ac_cube_smart_3.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_ac_sq_smart_3 = dict(tqdm(sorted(sorted_n_gram_tok_counts_anti_length_net_smart_3.items(), key=lambda x: x[1][0]/(x[1][1]**(2/6)), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_ac_sq_smart_3.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_ac_sq_smart_3.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_smart_3 = dict(tqdm(sorted(sorted_n_gram_tok_counts_anti_length_net_smart_3.items(), key=lambda x: x[1][0]/(x[1][1]**(1/5)), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_smart_3.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_smart_3.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_ac_smart_3 = dict(tqdm(sorted(sorted_n_gram_tok_counts_anti_length_net_smart_3.items(), key=lambda x: x[1][0]/(x[1][1]**(1/6)), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_ac_smart_3.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_ac_smart_3.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_smart_2 = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)*longest_same_char_sequence(x[0])*list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0]))))))**(1/4), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_smart_2.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_smart_2.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_sup_linear_smart_2 = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)*longest_same_char_sequence(x[0])*list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0]))))))**(1.5/4), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_sup_linear_smart_2.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sup_linear_smart_2.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2 = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)*longest_same_char_sequence(x[0])*list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0]))))))**(2.5/4), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

import numpy as np

counts = []
for i in tqdm(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2):
  counts.append(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2[i])
counts = np.array(counts)

counts2 = []
for i in tqdm(sorted_n_gram_tok_counts_anti_length_ac_cube_smart_3):
  counts2.append(sorted_n_gram_tok_counts_anti_length_ac_cube_smart_3[i])
counts2 = np.array(counts2)

counts2

counts2_ = counts2[:,0] / (counts2[:,1] ** (3/6))
counts2_

counts = counts[:131_072]
counts2_ = counts2_[:131_072]

# counts = (counts - counts.mean())/counts.std()
# counts2_ = (counts2_ - counts2_.mean())/counts2_.std()

counts = counts - counts.mean()

counts2_ = counts2_ - counts2_.mean()

counts.std()

counts.mean()

counts2_.mean()

counts2_.std()

scale = counts.std() / counts2_.std()
scale

scale = 1

sorted_n_gram_tok_counts_anti_length_hybrid_smart = {}
ind = 0
for i in tqdm(list(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2.keys())[:131_072]):
  sorted_n_gram_tok_counts_anti_length_hybrid_smart[i] = counts[ind]

ind = 0
for i in tqdm(list(sorted_n_gram_tok_counts_anti_length_ac_cube_smart_3.keys())[:131_072]):
  if ind in sorted_n_gram_tok_counts_anti_length_hybrid_smart:
    sorted_n_gram_tok_counts_anti_length_hybrid_smart[i] += scale* 0.99994575 * counts2_[ind]
  else:
    sorted_n_gram_tok_counts_anti_length_hybrid_smart[i] = scale* 0.99994575 * counts2_[ind]

sorted_n_gram_tok_counts_anti_length_hybrid_smart_sorted = dict(tqdm(sorted(sorted_n_gram_tok_counts_anti_length_hybrid_smart.items(), key=lambda x: x[1], reverse=True)))

sorted_n_gram_tok_counts_anti_length_hybrid_smart_sorted

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_hybrid_smart_sorted.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_hybrid_smart_sorted.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

!wget https://huggingface.co/Qwen/QwQ-32B/resolve/main/tokenizer.json

import json
with open('tokenizer.json', 'r') as file:
    data = json.load(file)

data.keys()

len(data['model']['vocab'])

sim = 0
for i in tqdm(list(sorted_n_gram_tok_counts_anti_length_cube_smart_2.keys())):
  if i in data['model']['vocab'].keys():
    sim += 1
sim

max_length = 0
s = [0,0,0]
s2 = [0,0]
for i in list(data['model']['vocab'].keys()):
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(data['model']['vocab'].keys()):
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

from collections import Counter
import re

def count_overlapping_space_ngrams(text):
    # Find all sequences of consecutive spaces
    space_sequences = re.findall(r' +', text)
    ngram_counter = Counter()

    # For each sequence, count all possible contiguous sub-sequences (n-grams)
    for seq in tqdm(space_sequences):
        k = len(seq)  # length of the space run
        # For each possible n-gram length from 1 to k
        for n in range(1, k + 1):
            # There are (k - n + 1) contiguous sub-sequences of length n in this run
            ngram_counter[' ' * n] += (k - n + 1)

    # Sort results by the length of the space sequence for clarity
    sorted_ngrams = sorted(ngram_counter.items(), key=lambda x: len(x[0]))
    return sorted_ngrams

space_counts = count_overlapping_space_ngrams(net_corpus)
space_counts

space_counts_dict = {}
for i in space_counts:
  if i != ' ':
    space_counts_dict[i[0]] = i[1]
space_counts_dict

sorted_space_counts_dict = dict(tqdm(sorted(space_counts_dict.items(), key=lambda x: x[1], reverse=True)))

sorted_space_counts_dict

len(space_counts_dict)

len(sorted_n_gram_tok_counts)

for i in space_counts_dict:
  if i in sorted_n_gram_tok_counts:
    print(i)

for i in space_counts_dict:
  sorted_n_gram_tok_counts[i] = space_counts_dict[i]

len(sorted_n_gram_tok_counts)

sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)*longest_same_char_sequence(x[0])*list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0]))))))**(2.5/4), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_char_counts

len(sorted_char_counts)

top_131k_vals_final_app_with_sp_no_ind = {}
for i in list(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp.keys())[:131_072]:
  if i not in sorted_char_counts:
    top_131k_vals_final_app_with_sp_no_ind[i] = sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp[i]

len(top_131k_vals_final_app_with_sp_no_ind)

top_131k_vals_final_app_with_sp_and_ind = sorted_char_counts.copy()
j = len(top_131k_vals_final_app_with_sp_and_ind)
for i in top_131k_vals_final_app_with_sp_no_ind:
  if j>=131_072:
    break
  top_131k_vals_final_app_with_sp_and_ind[i] = top_131k_vals_final_app_with_sp_no_ind[i]
  j += 1

len(top_131k_vals_final_app_with_sp_and_ind)

sorted_top_131k_vals_final_app_with_sp_and_ind = dict(tqdm(sorted(top_131k_vals_final_app_with_sp_and_ind.items(), key=lambda x: x[1], reverse=True)))
sorted_top_131k_vals_final_app_with_sp_and_ind

def save_dict_to_json(data, filename):
    """Saves a dictionary to a JSON file."""
    with open(filename, 'w', encoding='utf-8') as f:
        json.dump(data, f, ensure_ascii=False, indent=4)

!pwd

# Commented out IPython magic to ensure Python compatibility.
# %cd /content

keys = list(sorted_top_131k_vals_final_app_with_sp_and_ind.keys())

ordered_tokeniser = {}
for id, i in enumerate(keys):
  ordered_tokeniser[i] = id

import random
random.shuffle(keys)

unordered_tokeniser = {}
for id, i in enumerate(keys):
  unordered_tokeniser[i] = id

unordered_tokeniser

ordered_tokeniser

save_dict_to_json(sorted_top_131k_vals_final_app_with_sp_and_ind, "count_tokenizer_0.5b_val_data_raw.json")

save_dict_to_json(ordered_tokeniser, "ordered_tokenizer_0.5b_val_data_raw.json")

save_dict_to_json(unordered_tokeniser, "unordered_tokenizer_0.5b_val_data_raw.json")

sorted_n_gram_tok_counts['#']

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_top_131k_vals_final_app_with_sp_and_ind.keys())[:131_072]:
  if i in sorted_n_gram_tok_counts:
    if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
      last_i_val = sorted_n_gram_tok_counts[i]
      last_i = i
  else:
    if sorted_char_counts[i] < last_i_val and len(i) > 1:
      last_i_val = sorted_char_counts[i]
      last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

try:
  print(list(sorted_n_gram_tok_counts.keys()).index(last_i))
except:
  pass

top_all_vals_final_app_with_sp_no_ind = {}
for i in tqdm(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp.keys()):
  if i not in sorted_char_counts:
    top_all_vals_final_app_with_sp_no_ind[i] = sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp[i]

len(top_all_vals_final_app_with_sp_no_ind)

for i in tqdm(top_all_vals_final_app_with_sp_no_ind):
  if i in sorted_char_counts:
    print(i)

print(list(sorted_char_counts.keys())[-1])

len(sorted_char_counts)

net_corpus_keys_order = list(sorted_char_counts.keys())
net_corpus_keys_order.extend(top_all_vals_final_app_with_sp_no_ind)

net_corpus_keys_order

net_corpus = sorted_char_counts | top_all_vals_final_app_with_sp_no_ind
net_corpus

len(net_corpus)

save_dict_to_json(net_corpus, "net_corpus_0.5b_val.json")

save_dict_to_json(sorted_char_counts, "sorted_char_counts_0.5b_val.json")

save_dict_to_json(top_all_vals_final_app_with_sp_no_ind, "top_all_vals_final_app_with_sp_no_ind_0.5b_val.json")

with open("net_corpus_keys_order_0.5b_val.txt", 'w') as file:
        for item in tqdm(net_corpus_keys_order):
            file.write(str(item) + '\n')

!git clone https://github.com/Tasmay-Tibrewal/tokenizer.git

import json

if not continue_seamless:
  with open("tokenizer/net_corpus_0.5b_val.json", "r", encoding="utf-8") as file:
      net_corpus = json.load(file)

len(net_corpus)

net_corpus

from tqdm.auto import tqdm

net_corpus_keys_pos = {}
for id,i in tqdm(enumerate(net_corpus)):
  net_corpus_keys_pos[i] = id
net_corpus_keys_pos

len(net_corpus_keys_pos)

mean_len_term = 0
for i in net_corpus:
  mean_len_term += len(i) * (len(i) + 1) / 2
mean_len_term /= len(net_corpus)
mean_len_term

total_its = mean_len_term*len(net_corpus)
round(total_its)

parent_strings_dict = {}
for id, i in tqdm(enumerate(net_corpus)):
  if len(i) == 1:
    if i not in parent_strings_dict:
      parent_strings_dict[i] = [id]
    else:
      parent_strings_dict[i].append(id)
    continue
  elif len(i)>50 or net_corpus[i] < 10:
    continue
  for j in range(len(i)):
    for k in range(j + 1, len(i) + 1):
      sub_word = i[j:k]
      if sub_word not in parent_strings_dict:
        parent_strings_dict[sub_word] = [id]
      else:
        parent_strings_dict[sub_word].append(id)

parent_strings_dict

len(parent_strings_dict)

keys = list(net_corpus.keys())
keys

def test_max_num(max_parent_id):
  rem_counts = 0
  single_max_occ = 0
  total_added = 0
  last_occ = 0
  max_parent_id_occ = 0
  min_parent_id_occ_outside = -1
  net_corpus_small_adj = {}
  for id,i in tqdm(enumerate(net_corpus)):
    if total_added >= 131_072:
      break
    elif total_added % 10_000 == 0:
      print(f"{total_added/1000:.0f}K ", end="")
    if i not in parent_strings_dict:
      print(f"'{i[:50]}'", len(i), net_corpus[i])
      continue
    single_occ = 0
    add = 1
    if len(i) > 1:
      for parent_id in parent_strings_dict[i]:
        parent = keys[parent_id]
        if len(i) > len(parent):
          print("Some issue, less or same length",i, parent)
          continue
        elif len(i) == len(parent) and i == parent:
          single_occ += 1
          continue
        elif len(i) == len(parent):
          print("Child parent length same but child != parent", i, parent)
          continue
        if net_corpus[parent] > net_corpus[i] * 0.7 :
          if parent_id < max_parent_id:
            if parent_id > max_parent_id_occ:
              max_parent_id_occ = parent_id
            single_occ += 1
            rem_counts += 1
            add = 0
            break
          else:
            if min_parent_id_occ_outside == -1:
              min_parent_id_occ_outside = parent_id
            if parent_id < min_parent_id_occ_outside:
              min_parent_id_occ_outside = parent_id
    if add == 1:
      last_occ = id
      total_added += 1
      net_corpus_small_adj[i] = net_corpus[i]
    if single_occ > single_max_occ:
      single_max_occ = single_occ
  return last_occ, max_parent_id_occ, min_parent_id_occ_outside

max_parent_id = len(net_corpus)
min = 131_072
max = max_parent_id
it = 0
while True:
  last_occ, max_parent_id_occ, min_parent_id_occ_outside = test_max_num(max_parent_id)
  print(f"\nIt: {it}, Threshold: {max_parent_id}, Looped till: {last_occ}")
  print(f"Inner max id: {max_parent_id_occ}, Outer min id: {min_parent_id_occ_outside}")
  print(f"Id b/w inner max and outer min: {(last_occ < min_parent_id_occ_outside and last_occ > max_parent_id_occ)}")
  if last_occ < min_parent_id_occ_outside and last_occ > max_parent_id_occ:
    break
  if last_occ < max_parent_id:
    max = last_occ
  elif last_occ > max_parent_id:
    min = last_occ
  max_parent_id = (min + max)/2
  print(f"New overall min: {min}, max: {max}\nNew Max parent id: {max_parent_id}\n{'='*40}\n\n")

max_parent_id

import math
max_parent_id = math.ceil(max_parent_id)
max_parent_id

rem_counts = 0
single_max_occ = 0
total_added = 0
last_occ = 0
net_corpus_small_adj = {}
for id,i in tqdm(enumerate(net_corpus)):
  if total_added >= 131_072:
    break
  elif total_added % 10_000 == 0:
    print(f"{total_added/1000:.0f}K ", end="")
  if i not in parent_strings_dict:
    print(f"'{i[:50]}'", len(i), net_corpus[i])
    continue
  single_occ = 0
  add = 1
  if len(i) > 1:
    for parent_id in parent_strings_dict[i]:
      parent = keys[parent_id]
      if len(i) > len(parent):
        print("Some issue, less or same length",i, parent)
        continue
      elif len(i) == len(parent) and i == parent:
        single_occ += 1
        continue
      elif len(i) == len(parent):
        print("Child parent length same but child != parent", i, parent)
        continue
      if net_corpus[parent] > net_corpus[i] * 0.7 :
        if parent_id < max_parent_id:
          # if parent_id > max_parent_id_occ:
          #   max_parent_id_occ = parent_id
          single_occ += 1
          rem_counts += 1
          add = 0
          break
        # else:
        #   if min_parent_id_occ_outside == -1:
        #     min_parent_id_occ_outside = parent_id
        #   if parent_id < min_parent_id_occ_outside:
        #     min_parent_id_occ_outside = parent_id
  if add == 1:
    last_occ = id
    total_added += 1
    net_corpus_small_adj[i] = net_corpus[i]
  if single_occ > single_max_occ:
    single_max_occ = single_occ

single_max_occ

last_occ

total_added

rem_counts

rem_counts + total_added, rem_counts + total_added == last_occ + 1

len(net_corpus_small_adj)

net_corpus_small_adj

first_index = -1
for id, i in tqdm(enumerate(net_corpus)):
  if id > last_occ:
    break
  if len(i) == 1:
    if i not in parent_strings_dict:
      pass
    else:
      pass
    continue
  elif len(i)>50 or net_corpus[i] < 10:
    if first_index == -1:
      first_index = id
    print(id, len(i), net_corpus[i], i)
    continue
  else:
    pass

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = net_corpus[' ']
last_i = None
for i in list(net_corpus_small_adj.keys())[:131_072]:
  if net_corpus[i] < last_i_val and len(i) > 1:
    last_i_val = net_corpus[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(net_corpus_small_adj.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

net_corpus['ља']

net_corpus[last_i]

sorted_net_corpus_small_adj = dict(tqdm(sorted(net_corpus_small_adj.items(), key=lambda x: x[1], reverse=True)))
sorted_net_corpus_small_adj

keys = list(net_corpus_small_adj.keys())

ordered_tokeniser = {}
for id, i in enumerate(keys):
  ordered_tokeniser[i] = id

import random
random.shuffle(keys)

unordered_tokeniser = {}
for id, i in enumerate(keys):
  unordered_tokeniser[i] = id

ordered_tokeniser

unordered_tokeniser

def save_dict_to_json(data, filename):
    """Saves a dictionary to a JSON file."""
    with open(filename, 'w', encoding='utf-8') as f:
        json.dump(data, f, ensure_ascii=False, indent=4)

!pwd

# Commented out IPython magic to ensure Python compatibility.
# %cd /content

save_dict_to_json(sorted_net_corpus_small_adj, "count_tokenizer_0.5b_val_data.json")

save_dict_to_json(ordered_tokeniser, "ordered_tokenizer_0.5b_val_data.json")

save_dict_to_json(unordered_tokeniser, "unordered_tokenizer_0.5b_val_data.json")

"""# Tokenisation Whole (Experimenation) - 1b"""

!pip install zstandard

!pip install datasets

from huggingface_hub import hf_hub_download
from huggingface_hub import list_repo_files
from datasets import load_dataset

# Commented out IPython magic to ensure Python compatibility.
!mkdir SlimPajama-627B
# %cd SlimPajama-627B
!mkdir test

# Commented out IPython magic to ensure Python compatibility.
# %cd test

!git init
!git remote add -f origin https://huggingface.co/datasets/cerebras/SlimPajama-627B
!git config core.sparseCheckout true
!echo "test/*" >> .git/info/sparse-checkout
!git pull origin main

# Commented out IPython magic to ensure Python compatibility.
# %cd /content/SlimPajama-627B
!mkdir validation
# %cd validation
!git init
!git remote add -f origin https://huggingface.co/datasets/cerebras/SlimPajama-627B
!git config core.sparseCheckout true
!echo "validation/*" >> .git/info/sparse-checkout
!git pull origin main

# Commented out IPython magic to ensure Python compatibility.
# %cd /content/SlimPajama-627B/test/test

!pwd

!pip install chardet

import os
import json
import chardet
from tqdm.auto import tqdm
from collections import Counter

!ls -lh /content/SlimPajama-627B/test/test

!ls -lh /content/SlimPajama-627B/test/test/chunk1

!ls -l /content/SlimPajama-627B/test/test/chunk1 | grep -v '^d' | wc -l

import os
import json
import zstandard as zstd
from tqdm import tqdm

# Define the path to the test dataset
dataset_dir = "/content/SlimPajama-627B/test/test"

# Initialize an empty list to store the extracted text
corpus = []

# Function to read .jsonl.zst files
def read_jsonl_zst(file_path):
    with open(file_path, "rb") as f:
        dctx = zstd.ZstdDecompressor()
        with dctx.stream_reader(f) as reader:
            decompressed_data = reader.read()  # Read all decompressed data
            for line in decompressed_data.splitlines():  # Split into lines
                try:
                    data = json.loads(line.decode("utf-8"))  # Decode and parse JSON
                    if isinstance(data, dict) and "text" in data:
                        yield data["text"]
                except json.JSONDecodeError:
                    print(f"Skipping malformed JSON in {file_path}")

# Recursively iterate over all files in the dataset directory
for root, _, files in tqdm(os.walk(dataset_dir)):
    for file in tqdm(files):
        file_path = os.path.join(root, file)

        # Process only .jsonl.zst files
        if file.endswith(".jsonl.zst"):
            corpus.extend(read_jsonl_zst(file_path))

# The corpus now contains all extracted text
print(f"Extracted {len(corpus)} text entries.")

len(corpus)

# Commented out IPython magic to ensure Python compatibility.
# %cd /content/SlimPajama-627B/validation/validation

!pwd

!ls -lh /content/SlimPajama-627B/validation/validation

!ls -lh /content/SlimPajama-627B/validation/validation/chunk1

!ls -l /content/SlimPajama-627B/validation/validation/chunk1 | grep -v '^d' | wc -l

# Define the path to the test dataset
dataset_dir = "/content/SlimPajama-627B/validation/validation"

# Function to read .jsonl.zst files
def read_jsonl_zst(file_path):
    with open(file_path, "rb") as f:
        dctx = zstd.ZstdDecompressor()
        with dctx.stream_reader(f) as reader:
            decompressed_data = reader.read()  # Read all decompressed data
            for line in decompressed_data.splitlines():  # Split into lines
                try:
                    data = json.loads(line.decode("utf-8"))  # Decode and parse JSON
                    if isinstance(data, dict) and "text" in data:
                        yield data["text"]
                except json.JSONDecodeError:
                    print(f"Skipping malformed JSON in {file_path}")

# Recursively iterate over all files in the dataset directory
for root, _, files in tqdm(os.walk(dataset_dir)):
    for file in tqdm(files):
        file_path = os.path.join(root, file)

        # Process only .jsonl.zst files
        if file.endswith(".jsonl.zst"):
            corpus.extend(read_jsonl_zst(file_path))

# The corpus now contains all extracted text
print(f"Extracted {len(corpus)} text entries.")

len(corpus)

corpus[0]

corpus[1][-50:]

corpus[2]

corpus[len(corpus)//2][-50:]

corpus[len(corpus)//2+1]

corpus[len(corpus)//2+2]

len(corpus)

first_quarter_corpus = " ".join(corpus[:len(corpus)//4])
second_quarter_corpus = " ".join(corpus[len(corpus)//4:2*(len(corpus)//4)])
third_quarter_corpus = " ".join(corpus[2*(len(corpus)//4):3*(len(corpus)//4)])
fourth_quarter_corpus = " ".join(corpus[3*(len(corpus)//4):])

net_corpus = " ".join(corpus)
# net_corpus = first_half_corpus + " " + second_half_corpus

net_corpus[:500]

len(net_corpus)

!pwd

# Commented out IPython magic to ensure Python compatibility.
# %cd /content

with open("whole_corpus_1b_val_test.txt", "w") as file:
    file.write(net_corpus)

with open("first_quarter_corpus_1b_val_test.txt", "w") as file:
    file.write(first_quarter_corpus)

with open("second_quarter_corpus_1b_val_test.txt", "w") as file:
    file.write(second_quarter_corpus)

with open("third_quarter_corpus_1b_val_test.txt", "w") as file:
    file.write(third_quarter_corpus)

with open("fourth_quarter_corpus_1b_val_test.txt", "w") as file:
    file.write(fourth_quarter_corpus)

with open("whole_corpus_ind_texts_1b_val_test.json", "w") as file:
    json.dump(corpus, file, indent=4)

with open("first_quarter_corpus_ind_texts_1b_val_test.json", "w") as file:
    json.dump(corpus[:len(corpus)//4], file, indent=4)

with open("second_quarter_corpus_ind_texts_1b_val_test.json", "w") as file:
    json.dump(corpus[len(corpus)//4:2*(len(corpus)//4)], file, indent=4)

with open("third_quarter_corpus_ind_texts_1b_val_test.json", "w") as file:
    json.dump(corpus[2*(len(corpus)//4):3*(len(corpus)//4)], file, indent=4)

with open("fourth_quarter_corpus_ind_texts_1b_val_test.json", "w") as file:
    json.dump(corpus[3*(len(corpus)//4):], file, indent=4)

# Commented out IPython magic to ensure Python compatibility.
# %cd /content/SlimPajama-627B/validation/validation

char_counts = Counter(net_corpus)

len(char_counts)

sorted_char_counts = dict(sorted(char_counts.items(), key=lambda item: item[1], reverse=True))

total_char_occ = 0
for i in sorted_char_counts:
  total_char_occ += sorted_char_counts[i]
total_char_occ

sorted_char_counts[' ']

sorted_char_counts

len(net_corpus)

import regex
import re

def split_text_into_words(text):
    # Unicode-aware splitting: removes punctuation, spaces, and special characters
    words = regex.split(r'[^\p{L}\p{N}_]+', text)

    # Remove empty strings
    return [word for word in words if word]

sp_ch_sep_toks = split_text_into_words(net_corpus)
len(sp_ch_sep_toks)

word_counts = Counter(sp_ch_sep_toks)

len(word_counts)

word_counts = word_counts | sorted_char_counts

len(word_counts)

word_counts

sorted_word_counts = dict(sorted(word_counts.items(), key=lambda x: x[1], reverse=True))
sorted_word_counts

max_len = 0
for word in word_counts:
  if len(word) > max_len:
    max_len = len(word)
print(max_len)

len(sorted_word_counts)

word_len_count_map = {}
for word in word_counts:
  if len(word) not in word_len_count_map:
    word_len_count_map[len(word)] = word_counts[word]
  else:
    word_len_count_map[len(word)] += word_counts[word]

# Sort in descending order based on the second element
sorted_word_len_count_map = dict(sorted(word_len_count_map.items(), key=lambda x: x[1], reverse=True))
sorted_word_len_count_map

len(sorted_word_len_count_map)

sum = 0
for i in sorted_word_len_count_map:
  sum += sorted_word_len_count_map[i]
sum

lengths = {}
for word in list(sorted_word_counts.keys())[:131_072]:
  if len(word) not in lengths:
    lengths[len(word)] = [sorted_word_counts[word],sorted_word_counts[word],1]
  else:
    lengths[len(word)][0] += sorted_word_counts[word]
    lengths[len(word)][2] += 1
    if sorted_word_counts[word] > lengths[len(word)][1]:
      lengths[len(word)][1] = sorted_word_counts[word]
lengths

len(lengths)

sorted_lengths_max_ind_tok_counts = dict(sorted(lengths.items(), key=lambda item: item[1][1], reverse=True))
sorted_lengths_max_ind_tok_counts

from collections import defaultdict

def count_char_ngrams_in_words(word_list, max_n=14):
    """
    Count character-level n-grams for each word in a list.

    Args:
        word_list (list): List of words to analyze
        max_n (int): Maximum size of n-grams to count (default: 14)

    Returns:
        dict: Nested dictionary with n as outer key and
              inner dictionaries containing n-grams and their counts
    """
    # Initialize nested defaultdict to store counts
    ngram_counts = defaultdict(lambda: defaultdict(int))

    # Process each word separately
    for word in tqdm(word_list):
        word_length = len(word)

        # For each possible n-gram size (from 1 to max_n or word length)
        for n in range(2, min(max_n + 1, word_length)):
            # Use sliding window within each word
            for i in range(word_length - n + 1):
                # Extract the character n-gram
                ngram = word[i:i+n]

                # Increment the count for this n-gram
                ngram_counts[n][ngram] += 1

    # Convert defaultdict to regular dict for return
    return {n: dict(counts) for n, counts in ngram_counts.items()}

n_gram = count_char_ngrams_in_words(sp_ch_sep_toks,max_n=15)

threshold = 56_515
count_above_threshold = 0
for n in n_gram:
  for i in n_gram[n]:
    if n_gram[n][i] > threshold:
      count_above_threshold += 1
count_above_threshold

threshold = 141_993
count_above_threshold = 0
for n in n_gram:
  for i in n_gram[n]:
    if n_gram[n][i] > threshold:
      count_above_threshold += 1
count_above_threshold

t_sum_below = 0
for i in sorted_lengths_max_ind_tok_counts:
  if sorted_lengths_max_ind_tok_counts[i][1] < threshold:
    t_sum_below += sorted_lengths_max_ind_tok_counts[i][2]
t_sum_below

t_sum_below

n_gram_tok = {}
for n in tqdm(n_gram):
  for i in n_gram[n]:
    n_gram_tok[i] = n_gram[n][i]
n_gram_tok

sorted_n_gram_tok_counts = dict(sorted(n_gram_tok.items(), key=lambda item: item[1], reverse=True))
sorted_n_gram_tok_counts

sorted_word_counts

n_gram_lengths = {}
for i in list(sorted_n_gram_tok_counts.keys())[:131_072]:
    if len(i) not in n_gram_lengths:
      n_gram_lengths[len(i)] = [sorted_n_gram_tok_counts[i],sorted_n_gram_tok_counts[i],1]
    else:
      n_gram_lengths[len(i)][0] += sorted_n_gram_tok_counts[i]
      n_gram_lengths[len(i)][2] += 1
      if sorted_n_gram_tok_counts[i] > n_gram_lengths[len(i)][1]:
        n_gram_lengths[len(i)][1] = sorted_n_gram_tok_counts[i]
n_gram_lengths

sorted_n_gram_lengths_tok_counts = dict(sorted(n_gram_lengths.items(), key=lambda item: item[1][1], reverse=True))
sorted_n_gram_lengths_tok_counts

sorted_lengths_max_ind_tok_counts

import regex
from collections import Counter

def count_special_char_ngrams(text, max_n):
    """
    Count occurrences of character-level ngrams (only special characters, excluding whitespace) with n > 1.

    Parameters:
        text (str): The input text.
        max_n (int): The highest n-gram length to consider.

    Returns:
        dict: A mapping of special character ngrams to their counts.
    """
    # Pattern to match sequences of special characters (not letters, digits, underscore, or whitespace)
    special_pattern = r'[^\p{L}\p{N}_\s]+'

    counts = Counter()

    # Iterate over contiguous blocks of non-whitespace special characters
    for match in tqdm(regex.finditer(special_pattern, text)):
        block = match.group()
        L = len(block)

        # Generate ngrams for lengths from 2 to max_n (or block length, whichever is smaller)
        for n in range(2, min(L, max_n) + 1):
            for i in range(L - n + 1):
                ngram = block[i:i+n]
                counts[ngram] += 1

    return dict(counts)

sp_tok_n_grams = count_special_char_ngrams(net_corpus, 1000)
sp_tok_n_grams

sorted_sp_tok_n_grams = dict(sorted(sp_tok_n_grams.items(), key=lambda x: x[1], reverse=True))
sorted_sp_tok_n_grams

len(sorted_sp_tok_n_grams)

sorted_sp_tok_n_grams_len_count_map = {}
for sp_tok in sorted_sp_tok_n_grams:
  if len(sp_tok) not in sorted_sp_tok_n_grams_len_count_map:
    sorted_sp_tok_n_grams_len_count_map[len(sp_tok)] = sorted_sp_tok_n_grams[sp_tok]
  else:
    sorted_sp_tok_n_grams_len_count_map[len(sp_tok)] += sorted_sp_tok_n_grams[sp_tok]
# sorted_sp_tok_n_grams_len_count_map

# Sort in descending order based on the second element
sorted_sorted_sp_tok_n_grams_len_count_map = dict(sorted(sorted_sp_tok_n_grams_len_count_map.items(), key=lambda x: x[1], reverse=True))
sorted_sorted_sp_tok_n_grams_len_count_map

len(sorted_sp_tok_n_grams_len_count_map)

sum = 0
for i in sorted_sp_tok_n_grams_len_count_map:
  sum += sorted_sp_tok_n_grams_len_count_map[i]
sum

sp_token_ngram_lengths = {}
for sp_tok in list(sorted_sp_tok_n_grams.keys())[:131_072]:
  if len(sp_tok) not in sp_token_ngram_lengths:
    sp_token_ngram_lengths[len(sp_tok)] = [sorted_sp_tok_n_grams[sp_tok],sorted_sp_tok_n_grams[sp_tok],1]
  else:
    sp_token_ngram_lengths[len(sp_tok)][0] += sorted_sp_tok_n_grams[sp_tok]
    sp_token_ngram_lengths[len(sp_tok)][2] += 1
    if sorted_sp_tok_n_grams[sp_tok] > sp_token_ngram_lengths[len(sp_tok)][1]:
      sp_token_ngram_lengths[len(sp_tok)][1] = sorted_sp_tok_n_grams[sp_tok]
sp_token_ngram_lengths

def max_contiguous_occurrences(corpus):
    # Initialize a dictionary to store the maximum run length for each character
    max_runs = {}

    # Start tracking the current character and run length
    current_char = corpus[0]
    current_run = 1

    # Iterate over the corpus starting from the second character
    for ch in tqdm(corpus[1:]):
        if ch == current_char:
            # Increase the run if the same character continues
            current_run += 1
        else:
            if current_char not in max_runs:
                max_runs[current_char] = current_run
            else:
                max_runs[current_char] = max(max_runs[current_char], current_run)
            # Reset current_char and current_run for the new character
            current_char = ch
            current_run = 1

    if current_char not in max_runs:
        max_runs[current_char] = current_run
    else:
        max_runs[current_char] = max(max_runs[current_char], current_run)

    return max_runs

unique_char_list = sorted_char_counts.keys()

max_occ_map = max_contiguous_occurrences(net_corpus)

# Sort in descending order based on the second element
sorted_max_occ_map = dict(sorted(max_occ_map.items(), key=lambda x: x[1], reverse=True))
sorted_max_occ_map

sorted_word_counts

sorted_n_gram_tok_counts

sorted_sp_tok_n_grams

len(sorted_word_counts)

len(sorted_n_gram_tok_counts)

3809463 + 49495136

len(sorted_sp_tok_n_grams)

for i in tqdm(sorted_word_counts):
  if i in sorted_sp_tok_n_grams:
    print(i)

for i in tqdm(sorted_n_gram_tok_counts):
  if i in sorted_sp_tok_n_grams:
    print(i)

for i in tqdm(sorted_sp_tok_n_grams):
  if i in sorted_word_counts:
    print(i)

  if i in sorted_n_gram_tok_counts:
    print(i)

for i in tqdm(sorted_word_counts):
  if i not in sorted_n_gram_tok_counts:
    sorted_n_gram_tok_counts[i] = sorted_word_counts[i]
    continue

  sorted_n_gram_tok_counts[i] += sorted_word_counts[i]

sorted_n_gram_tok_counts = dict(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1], reverse=True))

sorted_n_gram_tok_counts

len(sorted_n_gram_tok_counts)

for i in tqdm(sorted_n_gram_tok_counts):
  if i in sorted_sp_tok_n_grams:
    print(i)

for i in tqdm(sorted_sp_tok_n_grams):
  if i in sorted_n_gram_tok_counts:
    print(i)

for i in list(sorted_n_gram_tok_counts.keys())[:131_072]:
  if len(set(list(i))) == 1 and len(i) > 3:
    print(i, len(i))

n_gram_lengths = {}
for i in list(sorted_n_gram_tok_counts.keys())[:131_072]:
    if len(i) not in n_gram_lengths:
      n_gram_lengths[len(i)] = [sorted_n_gram_tok_counts[i],sorted_n_gram_tok_counts[i],1]
    else:
      n_gram_lengths[len(i)][0] += sorted_n_gram_tok_counts[i]
      n_gram_lengths[len(i)][2] += 1
      if sorted_n_gram_tok_counts[i] > n_gram_lengths[len(i)][1]:
        n_gram_lengths[len(i)][1] = sorted_n_gram_tok_counts[i]
n_gram_lengths

for word in tqdm(sorted_n_gram_tok_counts):
  if word.strip(' \n').isnumeric() and sorted_n_gram_tok_counts[word] > 56_000:
      print(f"{word}:{sorted_word_counts[word]}")

len(sorted_n_gram_tok_counts)

len(sorted_sp_tok_n_grams)

52307513 + 696425

sorted_n_gram_tok_counts = dict(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1], reverse=True))

sorted_n_gram_tok_counts = sorted_n_gram_tok_counts | sorted_sp_tok_n_grams
len(sorted_n_gram_tok_counts)

for i in list(sorted_sp_tok_n_grams.keys())[:131_072]:
  if len(set(list(i))) == 1 and len(i) > 100:
    print(i[:5], len(i), sorted_sp_tok_n_grams[i], len(i)*sorted_sp_tok_n_grams[i])

top_tokens = {}

count = 0
c = 0
s = [0,0,0]
for i in list(sorted_n_gram_tok_counts.keys()):
  if count >= 131_072:
      break
  if i in sorted_sp_tok_n_grams:
    c += 1
  if len(set(list(i))) * 3 < len(i):
    if len(set(list(i))) > 1:
      print(i, sorted_n_gram_tok_counts[i])
    s[0] += 1
    if len(i) > 15:
      s[1] += 1
      if len(i) > 50:
        s[2] += 1
  else:
    count += 1
c, count, s

sorted_n_gram_tok_counts_anti_length = dict(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(len(x[0])), reverse=True))

max_length = 0
char_count = 0
for i in list(sorted_n_gram_tok_counts_anti_length.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    # print(i)
    char_count += 1
char_count, max_length

sorted_n_gram_tok_counts_anti_length_sq = dict(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(len(x[0])**2), reverse=True))

max_length = 0
s = [0,0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sq.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length

s = [0,0,0]
s2 = [0,0]
max_length = 0
for i in list(sorted_n_gram_tok_counts_anti_length_sq.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

sorted_n_gram_tok_counts_anti_length_cube = dict(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(len(x[0])**3), reverse=True))

max_length = 0
s = [0,0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_cube.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length

s = [0,0,0]
s2 = [0,0]
max_length = 0
for i in list(sorted_n_gram_tok_counts_anti_length_cube.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

sorted_n_gram_tok_counts_anti_length_quad = dict(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(len(x[0])**4), reverse=True))

max_length = 0
s = [0,0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_quad.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length

s = [0,0,0]
s2 = [0,0]
max_length = 0
for i in list(sorted_n_gram_tok_counts_anti_length_quad.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

# shows one minute but takes three minutes for some reason
sorted_n_gram_tok_counts_anti_length_cube_smart = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)* list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0])))))), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_cube_smart.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
max_length = 0
for i in list(sorted_n_gram_tok_counts_anti_length_cube_smart.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

# the three factors but in sq terms, taken 2/3 power
sorted_n_gram_tok_counts_anti_length_sq_smart = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)* list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0]))))))**(2/3), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sq_smart.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
max_length = 0
for i in list(sorted_n_gram_tok_counts_anti_length_sq_smart.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

def longest_same_char_sequence(s: str) -> int:
    if not s:
        return 0

    max_length = 1
    current_length = 1

    for i in range(1, len(s)):
        if s[i] == s[i - 1]:
            current_length += 1
        else:
            max_length = max(max_length, current_length)
            current_length = 1

    return max(max_length, current_length)

longest_same_char_sequence('\\\\\\\\\\\\\\\\\\\\\\')

sorted_n_gram_tok_counts_anti_length_cube_smart_2 = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)*longest_same_char_sequence(x[0])*list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0]))))))**(3/4), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_cube_smart_2.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
max_length = 0
for i in list(sorted_n_gram_tok_counts_anti_length_cube_smart_2.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_sq_smart_2 = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)*longest_same_char_sequence(x[0])*list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0]))))))**(2/4), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_sq_smart_2.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
max_length = 0
for i in list(sorted_n_gram_tok_counts_anti_length_sq_smart_2.keys())[:131_072]:
  if len(i) > max_length:
    max_length = len(i)
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

class SuffixAutomaton:
    def __init__(self, s):
        self.states = [{'len': 0, 'link': -1, 'next': {}, 'occ': 0}]
        self.last = 0  # index of the last (active) state
        for ch in s:
            self.add_char(ch)
        self.propagate_occurrences()

    def add_char(self, ch):
        p = self.last
        curr = len(self.states)
        # Create a new state with occ = 1 for the new ending character.
        self.states.append({'len': self.states[p]['len'] + 1, 'link': 0, 'next': {}, 'occ': 1})
        while p != -1 and ch not in self.states[p]['next']:
            self.states[p]['next'][ch] = curr
            p = self.states[p]['link']
        if p == -1:
            self.states[curr]['link'] = 0
        else:
            q = self.states[p]['next'][ch]
            if self.states[p]['len'] + 1 == self.states[q]['len']:
                self.states[curr]['link'] = q
            else:
                clone = len(self.states)
                self.states.append({
                    'len': self.states[p]['len'] + 1,
                    'next': self.states[q]['next'].copy(),
                    'link': self.states[q]['link'],
                    'occ': 0  # clone does not add occurrences by itself
                })
                while p != -1 and self.states[p]['next'].get(ch) == q:
                    self.states[p]['next'][ch] = clone
                    p = self.states[p]['link']
                self.states[q]['link'] = self.states[curr]['link'] = clone
        self.last = curr

    def propagate_occurrences(self):
        # Process states in order of decreasing length
        order = sorted(range(len(self.states)), key=lambda i: self.states[i]['len'], reverse=True)
        for i in order:
            link = self.states[i]['link']
            if link != -1:
                self.states[link]['occ'] += self.states[i]['occ']

def max_substring_value(s):
    n = len(s)
    if n < 2:
        return 1  # ensure minimum return value is 1, even if s is too short for valid substrings
    sa = SuffixAutomaton(s)
    max_val = 0
    # For each state with at least length 2, use the effective length which is capped at n-1.
    for state in sa.states:
        if state['len'] >= 2:
            effective_length = min(state['len'], n - 1)
            product = state['occ'] * effective_length
            if product > max_val:
                max_val = product
    return max_val if max_val > 0 else 1

sorted_n_gram_tok_counts_anti_length_net_smart_3 = {}
for i in tqdm(sorted_n_gram_tok_counts):
  sorted_n_gram_tok_counts_anti_length_net_smart_3[i] = [sorted_n_gram_tok_counts[i], (((((((len(i)**2)*longest_same_char_sequence(i)*max_substring_value(i)*list(Counter(list(i)).items())[0][1]))))/(len(set(list(i))))))]

sorted_n_gram_tok_counts_anti_length_cube_smart_3 = dict(tqdm(sorted(sorted_n_gram_tok_counts_anti_length_net_smart_3.items(), key=lambda x: x[1][0]/(x[1][1]**(3/5)), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_cube_smart_3.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_cube_smart_3.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_cube_smart_2.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_sq_smart_3 = dict(tqdm(sorted(sorted_n_gram_tok_counts_anti_length_net_smart_3.items(), key=lambda x: x[1][0]/(x[1][1]**(2/5)), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_sq_smart_3.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sq_smart_3.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sq_smart_2.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_ac_cube_smart_3 = dict(tqdm(sorted(sorted_n_gram_tok_counts_anti_length_net_smart_3.items(), key=lambda x: x[1][0]/(x[1][1]**(3/6)), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_ac_cube_smart_3.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_ac_cube_smart_3.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_ac_sq_smart_3 = dict(tqdm(sorted(sorted_n_gram_tok_counts_anti_length_net_smart_3.items(), key=lambda x: x[1][0]/(x[1][1]**(2/6)), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_ac_sq_smart_3.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_ac_sq_smart_3.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_smart_3 = dict(tqdm(sorted(sorted_n_gram_tok_counts_anti_length_net_smart_3.items(), key=lambda x: x[1][0]/(x[1][1]**(1/5)), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_smart_3.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_smart_3.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_ac_smart_3 = dict(tqdm(sorted(sorted_n_gram_tok_counts_anti_length_net_smart_3.items(), key=lambda x: x[1][0]/(x[1][1]**(1/6)), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_ac_smart_3.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_ac_smart_3.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_smart_2 = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)*longest_same_char_sequence(x[0])*list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0]))))))**(1/4), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_smart_2.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_smart_2.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_sup_linear_smart_2 = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)*longest_same_char_sequence(x[0])*list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0]))))))**(1.5/4), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_sup_linear_smart_2.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sup_linear_smart_2.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2 = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)*longest_same_char_sequence(x[0])*list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0]))))))**(2.5/4), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

import numpy as np

counts = []
for i in tqdm(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2):
  counts.append(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2[i])
counts = np.array(counts)

counts2 = []
for i in tqdm(sorted_n_gram_tok_counts_anti_length_ac_cube_smart_3):
  counts2.append(sorted_n_gram_tok_counts_anti_length_ac_cube_smart_3[i])
counts2 = np.array(counts2)

counts2

counts2_ = counts2[:,0] / (counts2[:,1] ** (3/6))
counts2_

counts = counts[:131_072]
counts2_ = counts2_[:131_072]

# counts = (counts - counts.mean())/counts.std()
# counts2_ = (counts2_ - counts2_.mean())/counts2_.std()

counts = counts - counts.mean()

counts2_ = counts2_ - counts2_.mean()

counts.std()

counts.mean()

counts2_.mean()

counts2_.std()

scale = counts.std() / counts2_.std()
scale

scale = 1

sorted_n_gram_tok_counts_anti_length_hybrid_smart = {}
ind = 0
for i in tqdm(list(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2.keys())[:131_072]):
  sorted_n_gram_tok_counts_anti_length_hybrid_smart[i] = counts[ind]

ind = 0
for i in tqdm(list(sorted_n_gram_tok_counts_anti_length_ac_cube_smart_3.keys())[:131_072]):
  if ind in sorted_n_gram_tok_counts_anti_length_hybrid_smart:
    sorted_n_gram_tok_counts_anti_length_hybrid_smart[i] += scale* 0.99994575 * counts2_[ind]
  else:
    sorted_n_gram_tok_counts_anti_length_hybrid_smart[i] = scale* 0.99994575 * counts2_[ind]

sorted_n_gram_tok_counts_anti_length_hybrid_smart_sorted = dict(tqdm(sorted(sorted_n_gram_tok_counts_anti_length_hybrid_smart.items(), key=lambda x: x[1], reverse=True)))

sorted_n_gram_tok_counts_anti_length_hybrid_smart_sorted

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_hybrid_smart_sorted.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_hybrid_smart_sorted.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

!wget https://huggingface.co/Qwen/QwQ-32B/resolve/main/tokenizer.json

import json
with open('tokenizer.json', 'r') as file:
    data = json.load(file)

data.keys()

len(data['model']['vocab'])

sim = 0
for i in tqdm(list(sorted_n_gram_tok_counts_anti_length_cube_smart_2.keys())):
  if i in data['model']['vocab'].keys():
    sim += 1
sim

max_length = 0
s = [0,0,0]
s2 = [0,0]
for i in list(data['model']['vocab'].keys()):
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(data['model']['vocab'].keys()):
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

from collections import Counter
import re

def count_overlapping_space_ngrams(text):
    # Find all sequences of consecutive spaces
    space_sequences = re.findall(r' +', text)
    ngram_counter = Counter()

    # For each sequence, count all possible contiguous sub-sequences (n-grams)
    for seq in tqdm(space_sequences):
        k = len(seq)  # length of the space run
        # For each possible n-gram length from 1 to k
        for n in range(1, k + 1):
            # There are (k - n + 1) contiguous sub-sequences of length n in this run
            ngram_counter[' ' * n] += (k - n + 1)

    # Sort results by the length of the space sequence for clarity
    sorted_ngrams = sorted(ngram_counter.items(), key=lambda x: len(x[0]))
    return sorted_ngrams

space_counts = count_overlapping_space_ngrams(net_corpus)
space_counts

space_counts_dict = {}
for i in space_counts:
  if i != ' ':
    space_counts_dict[i[0]] = i[1]
space_counts_dict

sorted_space_counts_dict = dict(tqdm(sorted(space_counts_dict.items(), key=lambda x: x[1], reverse=True)))

sorted_space_counts_dict

len(space_counts_dict)

len(sorted_n_gram_tok_counts)

for i in space_counts_dict:
  if i in sorted_n_gram_tok_counts:
    print(i)

for i in space_counts_dict:
  sorted_n_gram_tok_counts[i] = space_counts_dict[i]

len(sorted_n_gram_tok_counts)

sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp = dict(tqdm(sorted(sorted_n_gram_tok_counts.items(), key=lambda x: x[1]/(((((((len(x[0])**2)*longest_same_char_sequence(x[0])*list(Counter(list(x[0])).items())[0][1]))))/(len(set(list(x[0]))))))**(2.5/4), reverse=True)))

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp.keys())[:131_072]:
  if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
    last_i_val = sorted_n_gram_tok_counts[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

list(sorted_n_gram_tok_counts.keys()).index(last_i)

sorted_char_counts

len(sorted_char_counts)

top_131k_vals_final_app_with_sp_no_ind = {}
for i in list(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp.keys())[:131_072]:
  if i not in sorted_char_counts:
    top_131k_vals_final_app_with_sp_no_ind[i] = sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp[i]

len(top_131k_vals_final_app_with_sp_no_ind)

top_131k_vals_final_app_with_sp_and_ind = sorted_char_counts.copy()
j = len(top_131k_vals_final_app_with_sp_and_ind)
for i in top_131k_vals_final_app_with_sp_no_ind:
  if j>=131_072:
    break
  top_131k_vals_final_app_with_sp_and_ind[i] = top_131k_vals_final_app_with_sp_no_ind[i]
  j += 1

len(top_131k_vals_final_app_with_sp_and_ind)

sorted_top_131k_vals_final_app_with_sp_and_ind = dict(tqdm(sorted(top_131k_vals_final_app_with_sp_and_ind.items(), key=lambda x: x[1], reverse=True)))
sorted_top_131k_vals_final_app_with_sp_and_ind

def save_dict_to_json(data, filename):
    """Saves a dictionary to a JSON file."""
    with open(filename, 'w', encoding='utf-8') as f:
        json.dump(data, f, ensure_ascii=False, indent=4)

!pwd

# Commented out IPython magic to ensure Python compatibility.
# %cd /content

keys = list(sorted_top_131k_vals_final_app_with_sp_and_ind.keys())

ordered_tokeniser = {}
for id, i in enumerate(keys):
  ordered_tokeniser[i] = id

import random
random.shuffle(keys)

unordered_tokeniser = {}
for id, i in enumerate(keys):
  unordered_tokeniser[i] = id

unordered_tokeniser

ordered_tokeniser

save_dict_to_json(sorted_top_131k_vals_final_app_with_sp_and_ind, "count_tokenizer_1b_val_test_data_raw.json")

save_dict_to_json(ordered_tokeniser, "ordered_tokenizer_1b_val_test_data_raw.json")

save_dict_to_json(unordered_tokeniser, "unordered_tokenizer_1b_val_test_data_raw.json")

sorted_n_gram_tok_counts['#']

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = sorted_n_gram_tok_counts[' ']
last_i = None
for i in list(sorted_top_131k_vals_final_app_with_sp_and_ind.keys())[:131_072]:
  if i in sorted_n_gram_tok_counts:
    if sorted_n_gram_tok_counts[i] < last_i_val and len(i) > 1:
      last_i_val = sorted_n_gram_tok_counts[i]
      last_i = i
  else:
    if sorted_char_counts[i] < last_i_val and len(i) > 1:
      last_i_val = sorted_char_counts[i]
      last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

try:
  print(list(sorted_n_gram_tok_counts.keys()).index(last_i))
except:
  pass

top_all_vals_final_app_with_sp_no_ind = {}
for i in tqdm(sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp.keys()):
  if i not in sorted_char_counts:
    top_all_vals_final_app_with_sp_no_ind[i] = sorted_n_gram_tok_counts_anti_length_sup_sq_smart_2_with_sp[i]

len(top_all_vals_final_app_with_sp_no_ind)

for i in tqdm(top_all_vals_final_app_with_sp_no_ind):
  if i in sorted_char_counts:
    print(i)

print(list(sorted_char_counts.keys())[-1])

len(sorted_char_counts)

net_corpus_keys_order = list(sorted_char_counts.keys())
net_corpus_keys_order.extend(top_all_vals_final_app_with_sp_no_ind)

net_corpus_keys_order

net_corpus = sorted_char_counts | top_all_vals_final_app_with_sp_no_ind
net_corpus

len(net_corpus)

save_dict_to_json(net_corpus, "net_corpus_1b_val_test.json")

save_dict_to_json(sorted_char_counts, "sorted_char_counts_1b_val_test.json")

save_dict_to_json(top_all_vals_final_app_with_sp_no_ind, "top_all_vals_final_app_with_sp_no_ind_1b_val_test.json")

with open("net_corpus_keys_order_1b_val_test.txt", 'w') as file:
        for item in tqdm(net_corpus_keys_order):
            file.write(str(item) + '\n')

!git clone https://github.com/Tasmay-Tibrewal/tokenizer.git

import json

if not continue_seamless:
  with open("tokenizer/net_corpus_1b_val_test.json", "r", encoding="utf-8") as file:
      net_corpus = json.load(file)

len(net_corpus)

net_corpus

from tqdm.auto import tqdm

net_corpus_keys_pos = {}
for id,i in tqdm(enumerate(net_corpus)):
  net_corpus_keys_pos[i] = id
net_corpus_keys_pos

len(net_corpus_keys_pos)

mean_len_term = 0
for i in net_corpus:
  mean_len_term += len(i) * (len(i) + 1) / 2
mean_len_term /= len(net_corpus)
mean_len_term

total_its = mean_len_term*len(net_corpus)
round(total_its)

parent_strings_dict = {}
for id, i in tqdm(enumerate(net_corpus)):
  if len(i) == 1:
    if i not in parent_strings_dict:
      parent_strings_dict[i] = [id]
    else:
      parent_strings_dict[i].append(id)
    continue
  elif len(i)>50 or net_corpus[i] < 10:
    continue
  for j in range(len(i)):
    for k in range(j + 1, len(i) + 1):
      sub_word = i[j:k]
      if sub_word not in parent_strings_dict:
        parent_strings_dict[sub_word] = [id]
      else:
        parent_strings_dict[sub_word].append(id)

parent_strings_dict

len(parent_strings_dict)

keys = list(net_corpus.keys())
keys

def test_max_num(max_parent_id):
  rem_counts = 0
  single_max_occ = 0
  total_added = 0
  last_occ = 0
  max_parent_id_occ = 0
  min_parent_id_occ_outside = -1
  net_corpus_small_adj = {}
  for id,i in tqdm(enumerate(net_corpus)):
    if total_added >= 131_072:
      break
    elif total_added % 10_000 == 0:
      print(f"{total_added/1000:.0f}K ", end="")
    if i not in parent_strings_dict:
      print(f"'{i[:50]}'", len(i), net_corpus[i])
      continue
    single_occ = 0
    add = 1
    if len(i) > 1:
      for parent_id in parent_strings_dict[i]:
        parent = keys[parent_id]
        if len(i) > len(parent):
          print("Some issue, less or same length",i, parent)
          continue
        elif len(i) == len(parent) and i == parent:
          single_occ += 1
          continue
        elif len(i) == len(parent):
          print("Child parent length same but child != parent", i, parent)
          continue
        if net_corpus[parent] > net_corpus[i] * 0.7 :
          if parent_id < max_parent_id:
            if parent_id > max_parent_id_occ:
              max_parent_id_occ = parent_id
            single_occ += 1
            rem_counts += 1
            add = 0
            break
          else:
            if min_parent_id_occ_outside == -1:
              min_parent_id_occ_outside = parent_id
            if parent_id < min_parent_id_occ_outside:
              min_parent_id_occ_outside = parent_id
    if add == 1:
      last_occ = id
      total_added += 1
      net_corpus_small_adj[i] = net_corpus[i]
    if single_occ > single_max_occ:
      single_max_occ = single_occ
  return last_occ, max_parent_id_occ, min_parent_id_occ_outside

max_parent_id = len(net_corpus)
min = 131_072
max = max_parent_id
it = 0
while True:
  last_occ, max_parent_id_occ, min_parent_id_occ_outside = test_max_num(max_parent_id)
  print(f"\nIt: {it}, Threshold: {max_parent_id}, Looped till: {last_occ}")
  print(f"Inner max id: {max_parent_id_occ}, Outer min id: {min_parent_id_occ_outside}")
  print(f"Id b/w inner max and outer min: {(last_occ < min_parent_id_occ_outside and last_occ > max_parent_id_occ)}")
  if last_occ < min_parent_id_occ_outside and last_occ > max_parent_id_occ:
    break
  if last_occ < max_parent_id:
    max = last_occ
  elif last_occ > max_parent_id:
    min = last_occ
  max_parent_id = (min + max)/2
  print(f"New overall min: {min}, max: {max}\nNew Max parent id: {max_parent_id}\n{'='*40}\n\n")

max_parent_id

import math
max_parent_id = math.ceil(max_parent_id)
max_parent_id

rem_counts = 0
single_max_occ = 0
total_added = 0
last_occ = 0
net_corpus_small_adj = {}
for id,i in tqdm(enumerate(net_corpus)):
  if total_added >= 131_072:
    break
  elif total_added % 10_000 == 0:
    print(f"{total_added/1000:.0f}K ", end="")
  if i not in parent_strings_dict:
    print(f"'{i[:50]}'", len(i), net_corpus[i])
    continue
  single_occ = 0
  add = 1
  if len(i) > 1:
    for parent_id in parent_strings_dict[i]:
      parent = keys[parent_id]
      if len(i) > len(parent):
        print("Some issue, less or same length",i, parent)
        continue
      elif len(i) == len(parent) and i == parent:
        single_occ += 1
        continue
      elif len(i) == len(parent):
        print("Child parent length same but child != parent", i, parent)
        continue
      if net_corpus[parent] > net_corpus[i] * 0.7 :
        if parent_id < max_parent_id:
          # if parent_id > max_parent_id_occ:
          #   max_parent_id_occ = parent_id
          single_occ += 1
          rem_counts += 1
          add = 0
          break
        # else:
        #   if min_parent_id_occ_outside == -1:
        #     min_parent_id_occ_outside = parent_id
        #   if parent_id < min_parent_id_occ_outside:
        #     min_parent_id_occ_outside = parent_id
  if add == 1:
    last_occ = id
    total_added += 1
    net_corpus_small_adj[i] = net_corpus[i]
  if single_occ > single_max_occ:
    single_max_occ = single_occ

single_max_occ

last_occ

total_added

rem_counts

rem_counts + total_added, rem_counts + total_added == last_occ + 1

len(net_corpus_small_adj)

net_corpus_small_adj

first_index = -1
for id, i in tqdm(enumerate(net_corpus)):
  if id > last_occ:
    break
  if len(i) == 1:
    if i not in parent_strings_dict:
      pass
    else:
      pass
    continue
  elif len(i)>50 or net_corpus[i] < 10:
    if first_index == -1:
      first_index = id
    print(id, len(i), net_corpus[i], i)
    continue
  else:
    pass

max_length = 0
s = [0,0,0]
s2 = [0,0]
last_i_val = net_corpus[' ']
last_i = None
for i in list(net_corpus_small_adj.keys())[:131_072]:
  if net_corpus[i] < last_i_val and len(i) > 1:
    last_i_val = net_corpus[i]
    last_i = i
  if len(i) > max_length:
    max_length = len(i)
  if len(set(list(i))) * 3 < len(i):
    print(i)
    if len(i) > 5:
      if len(i) > 10:
        s2[1] += 1
      s2[0] += 1
    if len(i) > 15:
      if len(i) > 50:
        s[2] += 1
      s[1] += 1
    s[0] += 1
s, max_length, s2

s = [0,0,0]
s2 = [0,0]
for i in list(net_corpus_small_adj.keys())[:131_072]:
  if len(i) > 5:
    if len(i) > 10:
      s2[1] += 1
    s2[0] += 1
  if len(i) > 15:
    if len(i) > 50:
      s[2] += 1
    s[1] += 1
  s[0] += 1
s, max_length, s2

last_i

net_corpus['ља']

net_corpus[last_i]

sorted_net_corpus_small_adj = dict(tqdm(sorted(net_corpus_small_adj.items(), key=lambda x: x[1], reverse=True)))
sorted_net_corpus_small_adj

keys = list(net_corpus_small_adj.keys())

ordered_tokeniser = {}
for id, i in enumerate(keys):
  ordered_tokeniser[i] = id

import random
random.shuffle(keys)

unordered_tokeniser = {}
for id, i in enumerate(keys):
  unordered_tokeniser[i] = id

ordered_tokeniser

unordered_tokeniser

def save_dict_to_json(data, filename):
    """Saves a dictionary to a JSON file."""
    with open(filename, 'w', encoding='utf-8') as f:
        json.dump(data, f, ensure_ascii=False, indent=4)

!pwd

# Commented out IPython magic to ensure Python compatibility.
# %cd /content

save_dict_to_json(sorted_net_corpus_small_adj, "count_tokenizer_1b_val_test_data.json")

save_dict_to_json(ordered_tokeniser, "ordered_tokenizer_1b_val_test_data.json")

save_dict_to_json(unordered_tokeniser, "unordered_tokenizer_1b_val_test_data.json")