diff --git a/.gitignore b/.gitignore
index e07275e..b77c66a 100644
--- a/.gitignore
+++ b/.gitignore
@@ -173,7 +173,17 @@ cython_debug/
 # PyPI configuration file
 .pypirc
 
+checkpoints/
+AtlaSetCombined.txt
+# Prevent accidental commits of cleaned or redacted DM exports
+data/*_cleaned.txt
+data/*_redacted.txt
+# Ignore derived and private data exports
+data/*.jsonl
+data/dm_bpe.txt
 data/private/
+# Ignore generated notebook outputs and local workspace files
+notebooks/*_executed.ipynb
 output/
-checkpoints/
-AtlaSetCombined.txt
\ No newline at end of file
+.vscode/
+*.code-workspace
\ No newline at end of file
diff --git a/.patch_read_plain_chat.py b/.patch_read_plain_chat.py
new file mode 100644
index 0000000..2221c82
--- /dev/null
+++ b/.patch_read_plain_chat.py
@@ -0,0 +1,83 @@
+import json
+from pathlib import Path
+path = Path(r'c:\Users\dglan\Documents\hackclub\llm\Train_Your_Language_Model_Course-clean\notebooks\1_DataCleaning.ipynb')
+nb = json.loads(path.read_text(encoding='utf-8'))
+changed = False
+for cell in nb['cells']:
+    if cell.get('cell_type') == 'code' and any('def read_plain_chat(file_path: str) -> pd.DataFrame:' in line for line in cell.get('source', [])):
+        cell['source'] = [
+            'def read_plain_chat(file_path: str) -> pd.DataFrame:\n',
+            '    import re\n',
+            '    # Generic cleaner for plain text chats (Discord or other exports)\n',
+            '    email_pattern = r"[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\\.[A-Z|a-z]{2,}"\n',
+            '    url_pattern = r"http[s]?://(?:[a-zA-Z]|[0-9]|[$-_@.&+]|[!*\\(\\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+"\n',
+            '    media_pattern = "<Media omitted>"\n',
+            '    edited_message = "<This message was edited>"\n',
+            '    deleted_message = "You deleted this message"\n',
+            '    null_message = "null"\n',
+            '    created_group_message = "created group"\n',
+            '    added_you_to_group_message = "added you"\n',
+            '    tagging_pattern = r"@[\\w]+"\n',
+            '    export_marker = r"Exported .*message\\(s\\)"\n',
+            '    with open(file_path, "r", encoding="utf-8") as f:\n',
+            '        lines = f.readlines()\n',
+            '    filtered = []\n',
+            '    for line in lines:\n',
+            '        line = line.strip()\n',
+            '        if not line:\n',
+            '            continue\n',
+            '        if re.match(r"^[=-]{4,}$", line):\n',
+            '            continue\n',
+            '        if (\n',
+            '            deleted_message in line or\n',
+            '            null_message in line or\n',
+            '            media_pattern in line or\n',
+            '            created_group_message in line or\n',
+            '            added_you_to_group_message in line or\n',
+            '            re.search(email_pattern, line) or\n',
+            '            re.search(url_pattern, line) or\n',
+            '            re.search(export_marker, line)\n',
+            '        ):\n',
+            '            continue\n',
+            '        line = line.replace(edited_message, "").strip()\n',
+            '        line = re.sub(tagging_pattern, "", line).strip()\n',
+            '        if line:\n',
+            '            filtered.append(line)\n',
+            '    # Normalize unicode and join\n',
+            '    content = "\\n".join(filtered)\n',
+            '    content = content.replace("\\u202f", " ")\n',
+            '    content = content.replace("\\u200E", "").replace("\\u200F", "")\n',
+            '    # Parse lines: prefer dd/mm/yyyy style timestamps first, then ISO-like timestamps, then fall back to \'sender: message\'\n',
+            '    messages = []\n',
+            '    dm_pattern = r"^(\\d{1,2}/\\d{1,2}/\\d{2,4}, \\d{1,2}:\\d{2}(?::\\d{2})?)\\s*-\\s*(.*?):\\s*(.*)$"\n',
+            '    iso_pattern = r"^\\[?(\\d{4}-\\d{2}-\\d{2}[ T]\\d{2}:\\d{2}:\\d{2})\\]?\\s*(.*?):\\s*(.*)$"\n',
+            '    for line in content.split("\\n"):\n',
+            '        if not line.strip():\n',
+            '            continue\n',
+            '        m = re.match(dm_pattern, line)\n',
+            '        if m:\n',
+            '            ts, sender, msg = m.groups()\n',
+            '        else:\n',
+            '            m = re.match(iso_pattern, line)\n',
+            '            if m:\n',
+            '                ts, sender, msg = m.groups()\n',
+            '            else:\n',
+            '                m2 = re.match(r"^(.*?):\\s*(.*)$", line)\n',
+            '                if m2:\n',
+            '                    sender, msg = m2.groups()\n',
+            '                    ts = ""\n',
+            '                else:\n',
+            '                    sender = ""\n',
+            '                    msg = line\n',
+            '                    ts = ""\n',
+            '        messages.append((ts, sender.strip(), msg.strip()))\n',
+            '    df = pd.DataFrame(messages, columns=["timestamp","sender","message"])\n',
+            '    df["timestamp"] = pd.to_datetime(df["timestamp"], dayfirst=True, errors="coerce")\n',
+            '    return df\n',
+        ]
+        changed = True
+        break
+if not changed:
+    raise SystemExit('No matching read_plain_chat cell found')
+path.write_text(json.dumps(nb, indent=1, ensure_ascii=False), encoding='utf-8')
+print('patched')
diff --git a/data/anonymize_dm.py b/data/anonymize_dm.py
new file mode 100644
index 0000000..c2f2138
--- /dev/null
+++ b/data/anonymize_dm.py
@@ -0,0 +1,108 @@
+#!/usr/bin/env python3
+"""Anonymize DM export into a privacy-preserving JSONL file.
+
+Usage: python data/anonymize_dm.py path/to/raw.txt
+
+Output: creates path/to/raw_anonymized.jsonl with records:
+  {"date":"YYYY-MM-DD","user_id":"u1","message_hash":"abcd...","attachment_count":N}
+
+Notes: removes full timestamps (keeps date), replaces usernames with stable ids, hashes message text,
+and counts attachments instead of storing URLs.
+"""
+import re
+import sys
+import json
+import hashlib
+from pathlib import Path
+
+TS_USER_RE = re.compile(r"^\[(?P<ts>[^\]]+)\]\s*(?P<user>\S.*)$")
+URL_RE = re.compile(r"https?://\S+")
+
+
+def date_only(ts: str) -> str:
+    # Try to extract YYYY-MM-DD-like or MM/DD/YYYY and convert; fallback to raw date substring
+    # Example input: '8/31/2025 3:56 PM'
+    m = re.search(r"(\d{1,2})/(\d{1,2})/(\d{4})", ts)
+    if m:
+        mm, dd, yyyy = m.group(1), m.group(2), m.group(3)
+        return f"{int(yyyy):04d}-{int(mm):02d}-{int(dd):02d}"
+    # fallback: try YYYY-MM-DD
+    m2 = re.search(r"(\d{4})-(\d{2})-(\d{2})", ts)
+    if m2:
+        return f"{m2.group(1)}-{m2.group(2)}-{m2.group(3)}"
+    return ts.split()[0]
+
+
+def hash_text(s: str) -> str:
+    h = hashlib.sha256(s.encode('utf-8', errors='ignore')).hexdigest()
+    return h[:16]
+
+
+def anonymize(path: Path):
+    text = path.read_text(encoding='utf-8', errors='replace')
+    lines = text.splitlines()
+    user_map = {}
+    next_id = 1
+    out_lines = []
+    i = 0
+    while i < len(lines):
+        line = lines[i].rstrip()
+        m = TS_USER_RE.match(line)
+        if m:
+            ts = m.group('ts')
+            user = m.group('user').strip()
+            i += 1
+            msg_lines = []
+            att_count = 0
+            while i < len(lines):
+                nxt = lines[i]
+                if TS_USER_RE.match(nxt):
+                    break
+                if nxt.strip() == '{Attachments}':
+                    i += 1
+                    # count following URL lines
+                    while i < len(lines) and URL_RE.search(lines[i]):
+                        att_count += 1
+                        i += 1
+                    continue
+                # count inline urls as attachments and remove them from message
+                if URL_RE.search(nxt):
+                    att_count += len(URL_RE.findall(nxt))
+                    # strip urls
+                    nxt = URL_RE.sub('', nxt)
+                msg_lines.append(nxt.strip())
+                i += 1
+
+            msg = ' '.join(l for l in msg_lines if l)
+            # map user to stable id
+            if user not in user_map:
+                user_map[user] = f'u{next_id}'
+                next_id += 1
+            rec = {
+                'date': date_only(ts),
+                'user_id': user_map[user],
+                'message_hash': hash_text(msg) if msg else None,
+                'attachment_count': att_count,
+            }
+            out_lines.append(json.dumps(rec, ensure_ascii=False))
+        else:
+            i += 1
+
+    out = path.with_name(path.stem + '_anonymized.jsonl')
+    out.write_text('\n'.join(out_lines) + '\n', encoding='utf-8')
+    print(f'Wrote anonymized file: {out} (records: {len(out_lines)})')
+
+
+def main():
+    if len(sys.argv) < 2:
+        print('Usage: python data/anonymize_dm.py path/to/raw.txt')
+        sys.exit(1)
+    p = Path(sys.argv[1])
+    if not p.exists():
+        print('File not found:', p)
+        sys.exit(2)
+    anonymize(p)
+
+
+if __name__ == '__main__':
+    main()
diff --git a/data/clean_dm.py b/data/clean_dm.py
new file mode 100644
index 0000000..b47f252
--- /dev/null
+++ b/data/clean_dm.py
@@ -0,0 +1,114 @@
+#!/usr/bin/env python3
+"""Simple DM cleaning script.
+
+Usage: python data/clean_dm.py "path/to/file.txt"
+
+Outputs a cleaned file next to the input with suffix _cleaned.txt and prints a small summary.
+"""
+import re
+import sys
+from pathlib import Path
+
+
+OFFENSIVE = {"retards", "retard"}
+
+
+def redact(text: str) -> str:
+    def repl(m):
+        return "[redacted]"
+
+    # word-boundary replace for offensive words (case-insensitive)
+    pattern = re.compile(r"\\b(" + "|".join(re.escape(w) for w in OFFENSIVE) + r")\\b", flags=re.IGNORECASE)
+    text = pattern.sub(repl, text)
+    # normalize whitespace
+    text = re.sub(r"[\t\u00A0]+", " ", text)
+    text = re.sub(r" +", " ", text)
+    text = text.strip()
+    return text
+
+
+TS_USER_RE = re.compile(r"^\[(?P<ts>[^\]]+)\]\s*(?P<user>\S.*)$")
+URL_RE = re.compile(r"https?://\S+")
+
+
+def simplify_url(url: str) -> str:
+    # keep only base path (no query params)
+    return url.split("?")[0]
+
+
+def clean_messages(text: str):
+    lines = text.splitlines()
+    out_lines = []
+    i = 0
+    total_msgs = 0
+    removed = 0
+    while i < len(lines):
+        line = lines[i].rstrip()
+        m = TS_USER_RE.match(line)
+        if m:
+            ts = m.group("ts")
+            user = m.group("user").strip()
+            # gather following message lines until next timestamp or blank separator
+            i += 1
+            msg_lines = []
+            while i < len(lines):
+                nxt = lines[i]
+                if TS_USER_RE.match(nxt):
+                    break
+                # treat attachment marker and raw urls
+                if nxt.strip() == "{Attachments}":
+                    i += 1
+                    # collect URLs following
+                    att_urls = []
+                    while i < len(lines) and URL_RE.search(lines[i]):
+                        u = URL_RE.search(lines[i]).group(0)
+                        att_urls.append(simplify_url(u))
+                        i += 1
+                    if att_urls:
+                        msg_lines.append("[Attachment] " + ", ".join(att_urls))
+                    continue
+                msg_lines.append(nxt)
+                i += 1
+
+            msg = " ".join(l.strip() for l in msg_lines if l.strip())
+            msg = redact(msg)
+            if not msg:
+                removed += 1
+                continue
+            # turn into single-line record: timestamp \t user: message
+            out_lines.append(f"[{ts}]\t{user}: {msg}")
+            total_msgs += 1
+        else:
+            # non-timestamp lines: keep if they contain URLs or text
+            if URL_RE.search(line):
+                out_lines.append("[orphaned_url] " + simplify_url(URL_RE.search(line).group(0)))
+                total_msgs += 1
+            elif line.strip():
+                s = redact(line)
+                if s:
+                    out_lines.append(s)
+                    total_msgs += 1
+                else:
+                    removed += 1
+            i += 1
+
+    return "\n".join(out_lines) + "\n", total_msgs, removed
+
+
+def main():
+    if len(sys.argv) < 2:
+        print("Usage: python data/clean_dm.py path/to/file.txt")
+        sys.exit(1)
+    p = Path(sys.argv[1])
+    if not p.exists():
+        print("File not found:", p)
+        sys.exit(2)
+    raw = p.read_text(encoding="utf-8", errors="replace")
+    cleaned, kept, removed = clean_messages(raw)
+    out = p.with_name(p.stem + "_cleaned" + p.suffix)
+    out.write_text(cleaned, encoding="utf-8")
+    print(f"Cleaned: wrote {out}\nMessages kept: {kept}, removed/empty: {removed}")
+
+
+if __name__ == "__main__":
+    main()
diff --git a/data/make_combined.py b/data/make_combined.py
new file mode 100644
index 0000000..e7509fc
--- /dev/null
+++ b/data/make_combined.py
@@ -0,0 +1,12 @@
+from pathlib import Path
+
+infile = Path('data') / 'dm_bpe.txt'
+outdir = Path('output')
+outdir.mkdir(parents=True, exist_ok=True)
+outfile = outdir / 'combined_text.txt'
+
+text = infile.read_text(encoding='utf-8', errors='replace')
+# join lines with a space to create a single long sequence
+text_sequence = ' '.join(line.strip() for line in text.splitlines() if line.strip())
+outfile.write_text(text_sequence, encoding='utf-8')
+print(f'Wrote {outfile} (chars: {len(text_sequence)})')
diff --git a/data/make_private_dm.py b/data/make_private_dm.py
new file mode 100644
index 0000000..dbb46c5
--- /dev/null
+++ b/data/make_private_dm.py
@@ -0,0 +1,94 @@
+#!/usr/bin/env python3
+"""Convert a raw DM export into a privacy-safe chat log in the same plain-text format as DummyData.txt."""
+import re
+import sys
+from pathlib import Path
+from datetime import datetime
+
+TS_USER_RE = re.compile(r"^\[(?P<ts>[^\]]+)\]\s*(?P<user>\S.*)$")
+URL_RE = re.compile(r"https?://\S+")
+OFFENSIVE = {"retards", "retard"}
+
+
+def redact(text: str) -> str:
+    pattern = re.compile(r"\b(" + "|".join(re.escape(w) for w in OFFENSIVE) + r")\b", flags=re.IGNORECASE)
+    text = pattern.sub("[redacted]", text)
+    text = re.sub(r"[\t\u00A0]+", " ", text)
+    text = re.sub(r" +", " ", text)
+    return text.strip()
+
+
+def fmt_ts(raw: str) -> str:
+    # convert '8/31/2025 3:56 PM' -> '31/08/2025, 15:56'
+    try:
+        dt = datetime.strptime(raw, "%m/%d/%Y %I:%M %p")
+    except ValueError:
+        try:
+            dt = datetime.strptime(raw, "%m/%d/%Y %I:%M%p")
+        except ValueError:
+            return raw
+    return dt.strftime("%d/%m/%Y, %H:%M")
+
+
+def parse_messages(text: str):
+    lines = text.splitlines()
+    msgs = []
+    i = 0
+    while i < len(lines):
+        line = lines[i].rstrip()
+        m = TS_USER_RE.match(line)
+        if m:
+            ts = m.group("ts")
+            user = m.group("user").strip()
+            i += 1
+            msg_lines = []
+            while i < len(lines):
+                nxt = lines[i]
+                if TS_USER_RE.match(nxt):
+                    break
+                if nxt.strip() == "{Attachments}":
+                    i += 1
+                    while i < len(lines) and URL_RE.search(lines[i]):
+                        i += 1
+                    continue
+                msg_lines.append(nxt)
+                i += 1
+            message = " ".join(l.strip() for l in msg_lines if l.strip())
+            message = URL_RE.sub("", message)
+            message = redact(message)
+            if message:
+                msgs.append((fmt_ts(ts), user, message))
+            continue
+        i += 1
+    return msgs
+
+
+def write_private_file(input_path: Path, output_path: Path):
+    msgs = parse_messages(input_path.read_text(encoding="utf-8", errors="replace"))
+    name_map = {}
+    name_counter = 1
+    lines = []
+    for ts, user, msg in msgs:
+        if user not in name_map:
+            name_map[user] = f"Person {name_counter}"
+            name_counter += 1
+        lines.append(f"{ts} - {name_map[user]}: {msg}")
+    output_path.parent.mkdir(parents=True, exist_ok=True)
+    output_path.write_text("\n".join(lines) + "\n", encoding="utf-8")
+    print(f"Wrote {output_path} ({len(lines)} messages)")
+
+
+def main():
+    if len(sys.argv) < 2:
+        print("Usage: python data/make_private_dm.py path/to/raw_dm.txt")
+        sys.exit(1)
+    input_path = Path(sys.argv[1])
+    if not input_path.exists():
+        print("File not found:", input_path)
+        sys.exit(2)
+    output_path = Path("data/private") / input_path.name
+    write_private_file(input_path, output_path)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/data/prepare_bpe.py b/data/prepare_bpe.py
new file mode 100644
index 0000000..c6c0cd6
--- /dev/null
+++ b/data/prepare_bpe.py
@@ -0,0 +1,84 @@
+#!/usr/bin/env python3
+"""Prepare a BPE-ready corpus from a raw DM export.
+
+Output: data/dm_bpe.txt with one message per line.
+
+Behavior:
+- Collapse multi-line messages into single lines
+- Strip timestamps and usernames
+- Remove URLs and attachment markers
+- Lowercase and normalize whitespace
+- Filter out very short messages (<3 chars)
+"""
+import re
+import sys
+from pathlib import Path
+
+TS_USER_RE = re.compile(r"^\[(?P<ts>[^\]]+)\]\s*(?P<user>\S.*)$")
+URL_RE = re.compile(r"https?://\S+")
+
+
+def clean_msg_lines(msg_lines):
+    # join, remove urls, lowercase, normalize spaces
+    s = ' '.join(l.strip() for l in msg_lines if l and l.strip())
+    s = URL_RE.sub('', s)
+    s = re.sub(r"[\t\u00A0]+", ' ', s)
+    s = re.sub(r" +", ' ', s)
+    s = s.strip().lower()
+    return s
+
+
+def prepare(path: Path, outpath: Path):
+    text = path.read_text(encoding='utf-8', errors='replace')
+    lines = text.splitlines()
+    i = 0
+    count_in = 0
+    count_out = 0
+    with outpath.open('w', encoding='utf-8') as f:
+        while i < len(lines):
+            line = lines[i].rstrip()
+            m = TS_USER_RE.match(line)
+            if m:
+                count_in += 1
+                i += 1
+                msg_lines = []
+                while i < len(lines):
+                    nxt = lines[i]
+                    if TS_USER_RE.match(nxt):
+                        break
+                    if nxt.strip() == '{Attachments}':
+                        # skip attachments and following url lines
+                        i += 1
+                        while i < len(lines) and URL_RE.search(lines[i]):
+                            i += 1
+                        continue
+                    # treat inline url lines by removing urls later
+                    msg_lines.append(nxt)
+                    i += 1
+
+                cleaned = clean_msg_lines(msg_lines)
+                if len(cleaned) >= 3:
+                    f.write(cleaned + '\n')
+                    count_out += 1
+                continue
+            else:
+                # ignore non-timestamp top-level lines
+                i += 1
+
+    print(f'Prepared BPE corpus: wrote {outpath} (messages in: {count_in}, kept: {count_out})')
+
+
+def main():
+    if len(sys.argv) < 2:
+        print('Usage: python data/prepare_bpe.py path/to/raw.txt')
+        sys.exit(1)
+    p = Path(sys.argv[1])
+    if not p.exists():
+        print('File not found:', p)
+        sys.exit(2)
+    out = Path('data') / 'dm_bpe.txt'
+    prepare(p, out)
+
+
+if __name__ == '__main__':
+    main()
diff --git a/data/redact_all.py b/data/redact_all.py
new file mode 100644
index 0000000..a3bb9ea
--- /dev/null
+++ b/data/redact_all.py
@@ -0,0 +1,43 @@
+#!/usr/bin/env python3
+"""Create a fully redacted version of a cleaned DM file.
+
+Usage: python data/redact_all.py path/to/xxx_cleaned.txt [--delete]
+
+This writes a file with suffix _redacted.txt where every non-empty line is replaced
+with the single token [REDACTED].
+"""
+import sys
+from pathlib import Path
+
+
+def redact_file(p: Path, delete_original: bool = False):
+    text = p.read_text(encoding='utf-8', errors='replace')
+    lines = text.splitlines()
+    redacted_lines = []
+    for ln in lines:
+        if ln.strip():
+            redacted_lines.append('[REDACTED]')
+        else:
+            redacted_lines.append('')
+    out = p.with_name(p.stem.replace('_cleaned','') + '_redacted' + p.suffix)
+    out.write_text('\n'.join(redacted_lines) + '\n', encoding='utf-8')
+    print(f'Wrote redacted file: {out}')
+    if delete_original:
+        p.unlink()
+        print(f'Deleted original cleaned file: {p}')
+
+
+def main():
+    if len(sys.argv) < 2:
+        print('Usage: python data/redact_all.py path/to/xxx_cleaned.txt [--delete]')
+        sys.exit(1)
+    p = Path(sys.argv[1])
+    if not p.exists():
+        print('File not found:', p)
+        sys.exit(2)
+    delete = '--delete' in sys.argv[2:]
+    redact_file(p, delete_original=delete)
+
+
+if __name__ == '__main__':
+    main()
diff --git a/minbpe/regex.py b/minbpe/regex.py
index 3d9fb28..92c4d08 100644
--- a/minbpe/regex.py
+++ b/minbpe/regex.py
@@ -53,8 +53,10 @@ def train(self, text, vocab_size, verbose=False):
             for chunk_ids in ids:
                 # passing in stats will update it in place, adding up counts
                 get_stats(chunk_ids, stats)
+            if not stats:
+                break
             # find the pair with the highest count
-            pair = max(stats, key=stats.get)
+            pair = max(stats.items(), key=lambda item: item[1])[0]
             # mint a new token: assign it the next available id
             idx = 256 + i
             # replace all occurrences of pair in ids with idx
diff --git a/notebooks/1_DataCleaning.ipynb b/notebooks/1_DataCleaning.ipynb
index 25bb1bf..59d95f8 100644
--- a/notebooks/1_DataCleaning.ipynb
+++ b/notebooks/1_DataCleaning.ipynb
@@ -65,7 +65,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 4,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -139,10 +139,79 @@
    ]
   },
   {
-   "cell_type": "markdown",
+   "cell_type": "code",
+   "execution_count": 5,
    "metadata": {},
+   "outputs": [],
    "source": [
-    "The `all_chats` dictionary holds the content of each file as a dataframe with three columns: `timestamp`, `sender`, and `message`.  "
+    "def read_plain_chat(file_path: str) -> pd.DataFrame:\n",
+    "    import re\n",
+    "    # Generic cleaner for plain text chats (Discord or other exports)\n",
+    "    email_pattern = r\"[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\\.[A-Z|a-z]{2,}\"\n",
+    "    url_pattern = r\"http[s]?://(?:[a-zA-Z]|[0-9]|[$-_@.&+]|[!*\\(\\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+\"\n",
+    "    media_pattern = \"<Media omitted>\"\n",
+    "    edited_message = \"<This message was edited>\"\n",
+    "    deleted_message = \"You deleted this message\"\n",
+    "    null_message = \"null\"\n",
+    "    created_group_message = \"created group\"\n",
+    "    added_you_to_group_message = \"added you\"\n",
+    "    tagging_pattern = r\"@[\\w]+\"\n",
+    "    export_marker = r\"Exported .*message\\(s\\)\"\n",
+    "    with open(file_path, \"r\", encoding=\"utf-8\") as f:\n",
+    "        lines = f.readlines()\n",
+    "    filtered = []\n",
+    "    for line in lines:\n",
+    "        line = line.strip()\n",
+    "        if not line:\n",
+    "            continue\n",
+    "        if re.match(r\"^[=-]{4,}$\", line):\n",
+    "            continue\n",
+    "        if (\n",
+    "            deleted_message in line or\n",
+    "            null_message in line or\n",
+    "            media_pattern in line or\n",
+    "            created_group_message in line or\n",
+    "            added_you_to_group_message in line or\n",
+    "            re.search(email_pattern, line) or\n",
+    "            re.search(url_pattern, line) or\n",
+    "            re.search(export_marker, line)\n",
+    "        ):\n",
+    "            continue\n",
+    "        line = line.replace(edited_message, \"\").strip()\n",
+    "        line = re.sub(tagging_pattern, \"\", line).strip()\n",
+    "        if line:\n",
+    "            filtered.append(line)\n",
+    "    # Normalize unicode and join\n",
+    "    content = \"\\n\".join(filtered)\n",
+    "    content = content.replace(\"\\u202f\", \" \")\n",
+    "    content = content.replace(\"\\u200E\", \"\").replace(\"\\u200F\", \"\")\n",
+    "    # Parse lines: prefer dd/mm/yyyy style timestamps first, then ISO-like timestamps, then fall back to 'sender: message'\n",
+    "    messages = []\n",
+    "    dm_pattern = r\"^(\\d{1,2}/\\d{1,2}/\\d{2,4}, \\d{1,2}:\\d{2}(?::\\d{2})?)\\s*-\\s*(.*?):\\s*(.*)$\"\n",
+    "    iso_pattern = r\"^\\[?(\\d{4}-\\d{2}-\\d{2}[ T]\\d{2}:\\d{2}:\\d{2})\\]?\\s*(.*?):\\s*(.*)$\"\n",
+    "    for line in content.split(\"\\n\"):\n",
+    "        if not line.strip():\n",
+    "            continue\n",
+    "        m = re.match(dm_pattern, line)\n",
+    "        if m:\n",
+    "            ts, sender, msg = m.groups()\n",
+    "        else:\n",
+    "            m = re.match(iso_pattern, line)\n",
+    "            if m:\n",
+    "                ts, sender, msg = m.groups()\n",
+    "            else:\n",
+    "                m2 = re.match(r\"^(.*?):\\s*(.*)$\", line)\n",
+    "                if m2:\n",
+    "                    sender, msg = m2.groups()\n",
+    "                    ts = \"\"\n",
+    "                else:\n",
+    "                    sender = \"\"\n",
+    "                    msg = line\n",
+    "                    ts = \"\"\n",
+    "        messages.append((ts, sender.strip(), msg.strip()))\n",
+    "    df = pd.DataFrame(messages, columns=[\"timestamp\",\"sender\",\"message\"])\n",
+    "    df[\"timestamp\"] = pd.to_datetime(df[\"timestamp\"], dayfirst=True, errors=\"coerce\")\n",
+    "    return df\n"
    ]
   },
   {
@@ -150,14 +219,86 @@
    "execution_count": null,
    "metadata": {},
    "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "The `all_chats` dictionary holds the content of each file as a dataframe with three columns: `timestamp`, `sender`, and `message`.  "
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{'Direct Messages - Buncha retards [1411801796439183370]':                 timestamp                                   sender  \\\n",
+       " 0     2025-08-31 15:56:00                                 Person 1   \n",
+       " 1     2025-08-31 15:56:00                                 Person 1   \n",
+       " 2     2025-08-31 15:56:00                                 Person 2   \n",
+       " 3     2025-08-31 15:56:00                                 Person 3   \n",
+       " 4     2025-08-31 15:56:00  Person 1:\\n31/08/2025, 15:56 - Person 3   \n",
+       " ...                   ...                                      ...   \n",
+       " 27247 2026-06-30 12:35:00                                 Person 3   \n",
+       " 27248 2026-06-30 12:35:00                                 Person 3   \n",
+       " 27249 2026-06-30 12:36:00                                 Person 9   \n",
+       " 27250 2026-06-30 12:36:00                                 Person 1   \n",
+       " 27251 2026-06-30 12:36:00                                 Person 9   \n",
+       " \n",
+       "                                                  message  \n",
+       " 0                                                     Yo  \n",
+       " 1                         This confusing me texting both  \n",
+       " 2                                                   Lose  \n",
+       " 3                                                    wtf  \n",
+       " 4                                                   what  \n",
+       " ...                                                  ...  \n",
+       " 27247                                      youtube maybe  \n",
+       " 27248                                    post rednote fr  \n",
+       " 27249                          wait y did bro get kicked  \n",
+       " 27250                             i aint gonna post that  \n",
+       " 27251  what did bro do? =============================...  \n",
+       " \n",
+       " [27252 rows x 3 columns],\n",
+       " 'discord':                 timestamp    sender                         message\n",
+       " 0     2025-08-31 15:56:00  Person 1                              Yo\n",
+       " 1     2025-08-31 15:56:00  Person 1  This confusing me texting both\n",
+       " 2     2025-08-31 15:56:00  Person 2                            Lose\n",
+       " 3     2025-08-31 15:56:00  Person 3                             wtf\n",
+       " 4     2025-08-31 15:56:00  Person 1                                \n",
+       " ...                   ...       ...                             ...\n",
+       " 28005 2026-06-30 12:35:00  Person 3                       tik might\n",
+       " 28006 2026-06-30 12:35:00  Person 3                   youtube maybe\n",
+       " 28007 2026-06-30 12:35:00  Person 3                 post rednote fr\n",
+       " 28008 2026-06-30 12:36:00  Person 9       wait y did bro get kicked\n",
+       " 28009 2026-06-30 12:36:00  Person 1          i aint gonna post that\n",
+       " \n",
+       " [28010 rows x 3 columns]}"
+      ]
+     },
+     "execution_count": 6,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "from pathlib import Path\n",
     "\n",
     "all_chats = {}\n",
     "data_directory = Path(\"../data/private\")\n",
-    "for file in data_directory.glob('*.txt'):\n",
-    "    file_name = file.stem\n",
-    "    all_chats[file_name] = read_whatsapp_chat(file)"
+    "if data_directory.exists():\n",
+    "    for file in data_directory.glob(\"*.txt\"):\n",
+    "        file_name = file.stem\n",
+    "        all_chats[file_name] = read_whatsapp_chat(str(file))\n",
+    "\n",
+    "discord_file = Path(\"../data/private/Direct Messages - Buncha retards [1411801796439183370].txt\")\n",
+    "if discord_file.exists():\n",
+    "    all_chats['discord'] = read_plain_chat(str(discord_file))\n",
+    "\n",
+    "all_chats\n"
    ]
   },
   {
@@ -176,9 +317,20 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 7,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "1036745"
+      ]
+     },
+     "execution_count": 7,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "text_sequence = \"\"\n",
     "for file_name in all_chats.keys():\n",
@@ -189,18 +341,21 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 8,
    "metadata": {},
    "outputs": [],
    "source": [
-    "with open(\"../output/combined_text.txt\", \"w\", encoding=\"utf-8\") as f:\n",
+    "output_file = Path(\"../output/combined_text.txt\")\n",
+    "output_file.parent.mkdir(parents=True, exist_ok=True)\n",
+    "\n",
+    "with open(output_file, \"w\", encoding=\"utf-8\") as f:\n",
     "    f.write(text_sequence)"
    ]
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "vincent",
+   "display_name": "Python 3",
    "language": "python",
    "name": "python3"
   },
@@ -214,7 +369,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.11.10"
+   "version": "3.12.10"
   }
  },
  "nbformat": 4,
diff --git a/notebooks/2_BytePairEncoding.ipynb b/notebooks/2_BytePairEncoding.ipynb
index 90ba6c9..0a85402 100644
--- a/notebooks/2_BytePairEncoding.ipynb
+++ b/notebooks/2_BytePairEncoding.ipynb
@@ -9,9 +9,20 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 5,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "1036745"
+      ]
+     },
+     "execution_count": 5,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "with open(\"../output/combined_text.txt\", \"r\", encoding=\"utf-8\") as f:\n",
     "    text_sequence = f.read()\n",
@@ -35,7 +46,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 5,
+   "execution_count": 6,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -52,9 +63,17 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 6,
+   "execution_count": 7,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "100%|██████████| 768/768 [02:44<00:00,  4.68it/s]\n"
+     ]
+    }
+   ],
    "source": [
     "from minbpe import BasicTokenizer\n",
     "\n",
@@ -71,9 +90,1020 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 8,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "{0: b'\\x00',\n",
+       " 1: b'\\x01',\n",
+       " 2: b'\\x02',\n",
+       " 3: b'\\x03',\n",
+       " 4: b'\\x04',\n",
+       " 5: b'\\x05',\n",
+       " 6: b'\\x06',\n",
+       " 7: b'\\x07',\n",
+       " 8: b'\\x08',\n",
+       " 9: b'\\t',\n",
+       " 10: b'\\n',\n",
+       " 11: b'\\x0b',\n",
+       " 12: b'\\x0c',\n",
+       " 13: b'\\r',\n",
+       " 14: b'\\x0e',\n",
+       " 15: b'\\x0f',\n",
+       " 16: b'\\x10',\n",
+       " 17: b'\\x11',\n",
+       " 18: b'\\x12',\n",
+       " 19: b'\\x13',\n",
+       " 20: b'\\x14',\n",
+       " 21: b'\\x15',\n",
+       " 22: b'\\x16',\n",
+       " 23: b'\\x17',\n",
+       " 24: b'\\x18',\n",
+       " 25: b'\\x19',\n",
+       " 26: b'\\x1a',\n",
+       " 27: b'\\x1b',\n",
+       " 28: b'\\x1c',\n",
+       " 29: b'\\x1d',\n",
+       " 30: b'\\x1e',\n",
+       " 31: b'\\x1f',\n",
+       " 32: b' ',\n",
+       " 33: b'!',\n",
+       " 34: b'\"',\n",
+       " 35: b'#',\n",
+       " 36: b'$',\n",
+       " 37: b'%',\n",
+       " 38: b'&',\n",
+       " 39: b\"'\",\n",
+       " 40: b'(',\n",
+       " 41: b')',\n",
+       " 42: b'*',\n",
+       " 43: b'+',\n",
+       " 44: b',',\n",
+       " 45: b'-',\n",
+       " 46: b'.',\n",
+       " 47: b'/',\n",
+       " 48: b'0',\n",
+       " 49: b'1',\n",
+       " 50: b'2',\n",
+       " 51: b'3',\n",
+       " 52: b'4',\n",
+       " 53: b'5',\n",
+       " 54: b'6',\n",
+       " 55: b'7',\n",
+       " 56: b'8',\n",
+       " 57: b'9',\n",
+       " 58: b':',\n",
+       " 59: b';',\n",
+       " 60: b'<',\n",
+       " 61: b'=',\n",
+       " 62: b'>',\n",
+       " 63: b'?',\n",
+       " 64: b'@',\n",
+       " 65: b'A',\n",
+       " 66: b'B',\n",
+       " 67: b'C',\n",
+       " 68: b'D',\n",
+       " 69: b'E',\n",
+       " 70: b'F',\n",
+       " 71: b'G',\n",
+       " 72: b'H',\n",
+       " 73: b'I',\n",
+       " 74: b'J',\n",
+       " 75: b'K',\n",
+       " 76: b'L',\n",
+       " 77: b'M',\n",
+       " 78: b'N',\n",
+       " 79: b'O',\n",
+       " 80: b'P',\n",
+       " 81: b'Q',\n",
+       " 82: b'R',\n",
+       " 83: b'S',\n",
+       " 84: b'T',\n",
+       " 85: b'U',\n",
+       " 86: b'V',\n",
+       " 87: b'W',\n",
+       " 88: b'X',\n",
+       " 89: b'Y',\n",
+       " 90: b'Z',\n",
+       " 91: b'[',\n",
+       " 92: b'\\\\',\n",
+       " 93: b']',\n",
+       " 94: b'^',\n",
+       " 95: b'_',\n",
+       " 96: b'`',\n",
+       " 97: b'a',\n",
+       " 98: b'b',\n",
+       " 99: b'c',\n",
+       " 100: b'd',\n",
+       " 101: b'e',\n",
+       " 102: b'f',\n",
+       " 103: b'g',\n",
+       " 104: b'h',\n",
+       " 105: b'i',\n",
+       " 106: b'j',\n",
+       " 107: b'k',\n",
+       " 108: b'l',\n",
+       " 109: b'm',\n",
+       " 110: b'n',\n",
+       " 111: b'o',\n",
+       " 112: b'p',\n",
+       " 113: b'q',\n",
+       " 114: b'r',\n",
+       " 115: b's',\n",
+       " 116: b't',\n",
+       " 117: b'u',\n",
+       " 118: b'v',\n",
+       " 119: b'w',\n",
+       " 120: b'x',\n",
+       " 121: b'y',\n",
+       " 122: b'z',\n",
+       " 123: b'{',\n",
+       " 124: b'|',\n",
+       " 125: b'}',\n",
+       " 126: b'~',\n",
+       " 127: b'\\x7f',\n",
+       " 128: b'\\x80',\n",
+       " 129: b'\\x81',\n",
+       " 130: b'\\x82',\n",
+       " 131: b'\\x83',\n",
+       " 132: b'\\x84',\n",
+       " 133: b'\\x85',\n",
+       " 134: b'\\x86',\n",
+       " 135: b'\\x87',\n",
+       " 136: b'\\x88',\n",
+       " 137: b'\\x89',\n",
+       " 138: b'\\x8a',\n",
+       " 139: b'\\x8b',\n",
+       " 140: b'\\x8c',\n",
+       " 141: b'\\x8d',\n",
+       " 142: b'\\x8e',\n",
+       " 143: b'\\x8f',\n",
+       " 144: b'\\x90',\n",
+       " 145: b'\\x91',\n",
+       " 146: b'\\x92',\n",
+       " 147: b'\\x93',\n",
+       " 148: b'\\x94',\n",
+       " 149: b'\\x95',\n",
+       " 150: b'\\x96',\n",
+       " 151: b'\\x97',\n",
+       " 152: b'\\x98',\n",
+       " 153: b'\\x99',\n",
+       " 154: b'\\x9a',\n",
+       " 155: b'\\x9b',\n",
+       " 156: b'\\x9c',\n",
+       " 157: b'\\x9d',\n",
+       " 158: b'\\x9e',\n",
+       " 159: b'\\x9f',\n",
+       " 160: b'\\xa0',\n",
+       " 161: b'\\xa1',\n",
+       " 162: b'\\xa2',\n",
+       " 163: b'\\xa3',\n",
+       " 164: b'\\xa4',\n",
+       " 165: b'\\xa5',\n",
+       " 166: b'\\xa6',\n",
+       " 167: b'\\xa7',\n",
+       " 168: b'\\xa8',\n",
+       " 169: b'\\xa9',\n",
+       " 170: b'\\xaa',\n",
+       " 171: b'\\xab',\n",
+       " 172: b'\\xac',\n",
+       " 173: b'\\xad',\n",
+       " 174: b'\\xae',\n",
+       " 175: b'\\xaf',\n",
+       " 176: b'\\xb0',\n",
+       " 177: b'\\xb1',\n",
+       " 178: b'\\xb2',\n",
+       " 179: b'\\xb3',\n",
+       " 180: b'\\xb4',\n",
+       " 181: b'\\xb5',\n",
+       " 182: b'\\xb6',\n",
+       " 183: b'\\xb7',\n",
+       " 184: b'\\xb8',\n",
+       " 185: b'\\xb9',\n",
+       " 186: b'\\xba',\n",
+       " 187: b'\\xbb',\n",
+       " 188: b'\\xbc',\n",
+       " 189: b'\\xbd',\n",
+       " 190: b'\\xbe',\n",
+       " 191: b'\\xbf',\n",
+       " 192: b'\\xc0',\n",
+       " 193: b'\\xc1',\n",
+       " 194: b'\\xc2',\n",
+       " 195: b'\\xc3',\n",
+       " 196: b'\\xc4',\n",
+       " 197: b'\\xc5',\n",
+       " 198: b'\\xc6',\n",
+       " 199: b'\\xc7',\n",
+       " 200: b'\\xc8',\n",
+       " 201: b'\\xc9',\n",
+       " 202: b'\\xca',\n",
+       " 203: b'\\xcb',\n",
+       " 204: b'\\xcc',\n",
+       " 205: b'\\xcd',\n",
+       " 206: b'\\xce',\n",
+       " 207: b'\\xcf',\n",
+       " 208: b'\\xd0',\n",
+       " 209: b'\\xd1',\n",
+       " 210: b'\\xd2',\n",
+       " 211: b'\\xd3',\n",
+       " 212: b'\\xd4',\n",
+       " 213: b'\\xd5',\n",
+       " 214: b'\\xd6',\n",
+       " 215: b'\\xd7',\n",
+       " 216: b'\\xd8',\n",
+       " 217: b'\\xd9',\n",
+       " 218: b'\\xda',\n",
+       " 219: b'\\xdb',\n",
+       " 220: b'\\xdc',\n",
+       " 221: b'\\xdd',\n",
+       " 222: b'\\xde',\n",
+       " 223: b'\\xdf',\n",
+       " 224: b'\\xe0',\n",
+       " 225: b'\\xe1',\n",
+       " 226: b'\\xe2',\n",
+       " 227: b'\\xe3',\n",
+       " 228: b'\\xe4',\n",
+       " 229: b'\\xe5',\n",
+       " 230: b'\\xe6',\n",
+       " 231: b'\\xe7',\n",
+       " 232: b'\\xe8',\n",
+       " 233: b'\\xe9',\n",
+       " 234: b'\\xea',\n",
+       " 235: b'\\xeb',\n",
+       " 236: b'\\xec',\n",
+       " 237: b'\\xed',\n",
+       " 238: b'\\xee',\n",
+       " 239: b'\\xef',\n",
+       " 240: b'\\xf0',\n",
+       " 241: b'\\xf1',\n",
+       " 242: b'\\xf2',\n",
+       " 243: b'\\xf3',\n",
+       " 244: b'\\xf4',\n",
+       " 245: b'\\xf5',\n",
+       " 246: b'\\xf6',\n",
+       " 247: b'\\xf7',\n",
+       " 248: b'\\xf8',\n",
+       " 249: b'\\xf9',\n",
+       " 250: b'\\xfa',\n",
+       " 251: b'\\xfb',\n",
+       " 252: b'\\xfc',\n",
+       " 253: b'\\xfd',\n",
+       " 254: b'\\xfe',\n",
+       " 255: b'\\xff',\n",
+       " 256: b'e ',\n",
+       " 257: b't ',\n",
+       " 258: b's ',\n",
+       " 259: b'o ',\n",
+       " 260: b'n ',\n",
+       " 261: b'y ',\n",
+       " 262: b'th',\n",
+       " 263: b'd ',\n",
+       " 264: b'in',\n",
+       " 265: b'r ',\n",
+       " 266: b'ou',\n",
+       " 267: b'g ',\n",
+       " 268: b'an',\n",
+       " 269: b'm ',\n",
+       " 270: b'on',\n",
+       " 271: b'ing ',\n",
+       " 272: b'l ',\n",
+       " 273: b'h ',\n",
+       " 274: b'ha',\n",
+       " 275: b'ea',\n",
+       " 276: b'i ',\n",
+       " 277: b'er',\n",
+       " 278: b'is ',\n",
+       " 279: b'k ',\n",
+       " 280: b'  ',\n",
+       " 281: b'a ',\n",
+       " 282: b'ar',\n",
+       " 283: b'st',\n",
+       " 284: b'al',\n",
+       " 285: b'f ',\n",
+       " 286: b'\\xf0\\x9f',\n",
+       " 287: b'or',\n",
+       " 288: b'you',\n",
+       " 289: b'er ',\n",
+       " 290: b'li',\n",
+       " 291: b'ow',\n",
+       " 292: b'en',\n",
+       " 293: b'on ',\n",
+       " 294: b'ed ',\n",
+       " 295: b'it ',\n",
+       " 296: b'ly ',\n",
+       " 297: b'be',\n",
+       " 298: b'you ',\n",
+       " 299: b'st ',\n",
+       " 300: b'p ',\n",
+       " 301: b'the ',\n",
+       " 302: b'? ',\n",
+       " 303: b'an ',\n",
+       " 304: b'ke ',\n",
+       " 305: b'to ',\n",
+       " 306: b'me ',\n",
+       " 307: b'ot ',\n",
+       " 308: b'ti',\n",
+       " 309: b'la',\n",
+       " 310: b'wa',\n",
+       " 311: b'ts ',\n",
+       " 312: b'es ',\n",
+       " 313: b'or ',\n",
+       " 314: b'ow ',\n",
+       " 315: b'at ',\n",
+       " 316: b'en ',\n",
+       " 317: b'I ',\n",
+       " 318: b'in ',\n",
+       " 319: b'} ',\n",
+       " 320: b'oo',\n",
+       " 321: b'ra',\n",
+       " 322: b're',\n",
+       " 323: b'hat ',\n",
+       " 324: b'    ',\n",
+       " 325: b'wh',\n",
+       " 326: b'so ',\n",
+       " 327: b'ro',\n",
+       " 328: b'ic',\n",
+       " 329: b'and ',\n",
+       " 330: b've ',\n",
+       " 331: b'\\xe2\\xa0',\n",
+       " 332: b'bed',\n",
+       " 333: b'Em',\n",
+       " 334: b'ma',\n",
+       " 335: b'{Em',\n",
+       " 336: b'{Embed',\n",
+       " 337: b'et ',\n",
+       " 338: b'{Embed} ',\n",
+       " 339: b'like ',\n",
+       " 340: b'sh',\n",
+       " 341: b'se ',\n",
+       " 342: b'll ',\n",
+       " 343: b'br',\n",
+       " 344: b'not ',\n",
+       " 345: b'\\xf0\\x9f\\x98',\n",
+       " 346: b'id',\n",
+       " 347: b'bu',\n",
+       " 348: b'sa',\n",
+       " 349: b'le ',\n",
+       " 350: b'ch',\n",
+       " 351: b'da',\n",
+       " 352: b'ba',\n",
+       " 353: b'ig',\n",
+       " 354: b'tu',\n",
+       " 355: b'ust ',\n",
+       " 356: b'th ',\n",
+       " 357: b'that ',\n",
+       " 358: b'no ',\n",
+       " 359: b'im ',\n",
+       " 360: b'\\xe2\\xa0\\x80',\n",
+       " 361: b'ca',\n",
+       " 362: b'ne',\n",
+       " 363: b'what ',\n",
+       " 364: b'le',\n",
+       " 365: b'ld ',\n",
+       " 366: b'ch ',\n",
+       " 367: b'one ',\n",
+       " 368: b'\\xf0\\x9f\\x98\\xad',\n",
+       " 369: b'tt',\n",
+       " 370: b'wi',\n",
+       " 371: b'lo',\n",
+       " 372: b'be ',\n",
+       " 373: b'om',\n",
+       " 374: b'he ',\n",
+       " 375: b'0 ',\n",
+       " 376: b'my ',\n",
+       " 377: b'do ',\n",
+       " 378: b'for ',\n",
+       " 379: b'ac',\n",
+       " 380: b'di',\n",
+       " 381: b'\\xe2\\xa0\\x80\\xe2\\xa0\\x80',\n",
+       " 382: b'bro ',\n",
+       " 383: b'ho',\n",
+       " 384: b'of ',\n",
+       " 385: b'su',\n",
+       " 386: b'se',\n",
+       " 387: b'just ',\n",
+       " 388: b'are ',\n",
+       " 389: b'ga',\n",
+       " 390: b'we ',\n",
+       " 391: b'ir',\n",
+       " 392: b'ht ',\n",
+       " 393: b'pla',\n",
+       " 394: b'he',\n",
+       " 395: b'uh ',\n",
+       " 396: b'ta',\n",
+       " 397: b'ere ',\n",
+       " 398: b'\\xf0\\x9f\\x98\\xad ',\n",
+       " 399: b'        ',\n",
+       " 400: b'ev',\n",
+       " 401: b'un',\n",
+       " 402: b'rea',\n",
+       " 403: b'all ',\n",
+       " 404: b'can ',\n",
+       " 405: b'te ',\n",
+       " 406: b'to',\n",
+       " 407: b'u ',\n",
+       " 408: b'ss ',\n",
+       " 409: b'tr',\n",
+       " 410: b'ould ',\n",
+       " 411: b'get ',\n",
+       " 412: b'ck ',\n",
+       " 413: b'have ',\n",
+       " 414: b'was ',\n",
+       " 415: b'now ',\n",
+       " 416: b'this ',\n",
+       " 417: b'. ',\n",
+       " 418: b'w ',\n",
+       " 419: b'do',\n",
+       " 420: b'te',\n",
+       " 421: b'mo',\n",
+       " 422: b'ally ',\n",
+       " 423: b'gu',\n",
+       " 424: b'ck',\n",
+       " 425: b'ff ',\n",
+       " 426: b'ent ',\n",
+       " 427: b'its ',\n",
+       " 428: b'thin',\n",
+       " 429: b'el ',\n",
+       " 430: b'out ',\n",
+       " 431: b'ey ',\n",
+       " 432: b'if ',\n",
+       " 433: b', ',\n",
+       " 434: b'op',\n",
+       " 435: b'ant ',\n",
+       " 436: b'si',\n",
+       " 437: b'ys ',\n",
+       " 438: b'fu',\n",
+       " 439: b'pi',\n",
+       " 440: b'me',\n",
+       " 441: b'min',\n",
+       " 442: b'go',\n",
+       " 443: b'ore ',\n",
+       " 444: b'ri',\n",
+       " 445: b' \\xf0\\x9f',\n",
+       " 446: b'ce ',\n",
+       " 447: b'ight ',\n",
+       " 448: b'go ',\n",
+       " 449: b'2 ',\n",
+       " 450: b'than ',\n",
+       " 451: b'don',\n",
+       " 452: b'wor',\n",
+       " 453: b'You',\n",
+       " 454: b'bo',\n",
+       " 455: b'ter',\n",
+       " 456: b'but ',\n",
+       " 457: b'ood ',\n",
+       " 458: b'with ',\n",
+       " 459: b'cu',\n",
+       " 460: b'sp',\n",
+       " 461: b'vi',\n",
+       " 462: b'mu',\n",
+       " 463: b'cti',\n",
+       " 464: b'\\xe2\\x80',\n",
+       " 465: b'du',\n",
+       " 466: b'ge ',\n",
+       " 467: b'how ',\n",
+       " 468: b'll',\n",
+       " 469: b'day ',\n",
+       " 470: b'yea',\n",
+       " 471: b'b ',\n",
+       " 472: b'pp',\n",
+       " 473: b'T ',\n",
+       " 474: b'1 ',\n",
+       " 475: b'E ',\n",
+       " 476: b're ',\n",
+       " 477: b'use ',\n",
+       " 478: b'think ',\n",
+       " 479: b'oun',\n",
+       " 480: b'lea',\n",
+       " 481: b'c ',\n",
+       " 482: b'we',\n",
+       " 483: b'ex',\n",
+       " 484: b\"'s \",\n",
+       " 485: b'Wh',\n",
+       " 486: b'al ',\n",
+       " 487: b'\\xe2\\xa3',\n",
+       " 488: b'dont ',\n",
+       " 489: b'ill ',\n",
+       " 490: b'ying ',\n",
+       " 491: b'ge',\n",
+       " 492: b'fa',\n",
+       " 493: b'hh ',\n",
+       " 494: b'oug',\n",
+       " 495: b'5 ',\n",
+       " 496: b'Er',\n",
+       " 497: b'why ',\n",
+       " 498: b'id ',\n",
+       " 499: b'You ',\n",
+       " 500: b'What ',\n",
+       " 501: b'na',\n",
+       " 502: b'con',\n",
+       " 503: b'po',\n",
+       " 504: b'thing ',\n",
+       " 505: b'pro',\n",
+       " 506: b'na ',\n",
+       " 507: b'got ',\n",
+       " 508: b'know ',\n",
+       " 509: b'ds ',\n",
+       " 510: b'good ',\n",
+       " 511: b'ter ',\n",
+       " 512: b'co',\n",
+       " 513: b'ab',\n",
+       " 514: b'\\xe2\\x80\\x99',\n",
+       " 515: b'gr',\n",
+       " 516: b'Erm ',\n",
+       " 517: b'ani',\n",
+       " 518: b'est',\n",
+       " 519: b'so',\n",
+       " 520: b'No ',\n",
+       " 521: b'ons',\n",
+       " 522: b'S ',\n",
+       " 523: b'king ',\n",
+       " 524: b'                ',\n",
+       " 525: b'com',\n",
+       " 526: b'your ',\n",
+       " 527: b'it',\n",
+       " 528: b'So ',\n",
+       " 529: b'mm ',\n",
+       " 530: b'ut ',\n",
+       " 531: b'de ',\n",
+       " 532: b'\\xe2\\xa0\\x80\\xe2\\xa0\\x80\\xe2\\xa0\\x80\\xe2\\xa0\\x80',\n",
+       " 533: b'holy ',\n",
+       " 534: b'Rea',\n",
+       " 535: b'tf ',\n",
+       " 536: b'ver',\n",
+       " 537: b'sc',\n",
+       " 538: b'Ca',\n",
+       " 539: b'3 ',\n",
+       " 540: b'is',\n",
+       " 541: b'ec',\n",
+       " 542: b'ybe ',\n",
+       " 543: b'need ',\n",
+       " 544: b'ctions',\n",
+       " 545: b'es',\n",
+       " 546: b'wait ',\n",
+       " 547: b'{Rea',\n",
+       " 548: b'{Reactions',\n",
+       " 549: b'{Reactions} ',\n",
+       " 550: b'rn ',\n",
+       " 551: b'no',\n",
+       " 552: b'{Embed} {Embed} ',\n",
+       " 553: b'Th',\n",
+       " 554: b'then ',\n",
+       " 555: b'em ',\n",
+       " 556: b'when ',\n",
+       " 557: b'lu',\n",
+       " 558: b'ong ',\n",
+       " 559: b'qu',\n",
+       " 560: b\"I'\",\n",
+       " 561: b'ic ',\n",
+       " 562: b'son ',\n",
+       " 563: b'tuff ',\n",
+       " 564: b'nt ',\n",
+       " 565: b'ver ',\n",
+       " 566: b'mar',\n",
+       " 567: b'yes ',\n",
+       " 568: b'play ',\n",
+       " 569: b'ro ',\n",
+       " 570: b'end ',\n",
+       " 571: b'ther ',\n",
+       " 572: b'ough ',\n",
+       " 573: b'up ',\n",
+       " 574: b'stu',\n",
+       " 575: b'ru',\n",
+       " 576: b'ted ',\n",
+       " 577: b'x ',\n",
+       " 578: b'par',\n",
+       " 579: b'am ',\n",
+       " 580: b'Li',\n",
+       " 581: b'el',\n",
+       " 582: b': ',\n",
+       " 583: b'more ',\n",
+       " 584: b\"I'm \",\n",
+       " 585: b'even ',\n",
+       " 586: b'did ',\n",
+       " 587: b'dy ',\n",
+       " 588: b'ff',\n",
+       " 589: b'cause ',\n",
+       " 590: b'bi',\n",
+       " 591: b'tha',\n",
+       " 592: b'ep ',\n",
+       " 593: b'Y ',\n",
+       " 594: b'actu',\n",
+       " 595: b'low',\n",
+       " 596: b'gra',\n",
+       " 597: b'also ',\n",
+       " 598: b'Wa',\n",
+       " 599: b'oh ',\n",
+       " 600: b'tion ',\n",
+       " 601: b'should ',\n",
+       " 602: b'An',\n",
+       " 603: b'they ',\n",
+       " 604: b'say ',\n",
+       " 605: b'tch ',\n",
+       " 606: b'bad ',\n",
+       " 607: b'ya',\n",
+       " 608: b'any ',\n",
+       " 609: b'ano ',\n",
+       " 610: b'But ',\n",
+       " 611: b'ing',\n",
+       " 612: b'lly ',\n",
+       " 613: b' H',\n",
+       " 614: b'yeah ',\n",
+       " 615: b'har',\n",
+       " 616: b'for',\n",
+       " 617: b'by ',\n",
+       " 618: b'idk ',\n",
+       " 619: b'6 ',\n",
+       " 620: b'pa',\n",
+       " 621: b'ine ',\n",
+       " 622: b'rom ',\n",
+       " 623: b'game ',\n",
+       " 624: b'key ',\n",
+       " 625: b'ethan ',\n",
+       " 626: b'Cap ',\n",
+       " 627: b'time ',\n",
+       " 628: b'too ',\n",
+       " 629: b'4 ',\n",
+       " 630: b'about ',\n",
+       " 631: b'sk',\n",
+       " 632: b'de',\n",
+       " 633: b'ding ',\n",
+       " 634: b'gi',\n",
+       " 635: b'tter ',\n",
+       " 636: b'us',\n",
+       " 637: b'mb',\n",
+       " 638: b'D ',\n",
+       " 639: b'right ',\n",
+       " 640: b'ft ',\n",
+       " 641: b'who ',\n",
+       " 642: b'IN',\n",
+       " 643: b'aniel ',\n",
+       " 644: b'way ',\n",
+       " 645: b'mi',\n",
+       " 646: b'hou',\n",
+       " 647: b'TH',\n",
+       " 648: b'Ho',\n",
+       " 649: b'cap ',\n",
+       " 650: b'We ',\n",
+       " 651: b'shit ',\n",
+       " 652: b'Why ',\n",
+       " 653: b'ting ',\n",
+       " 654: b'there ',\n",
+       " 655: b'wan',\n",
+       " 656: b'from ',\n",
+       " 657: b'said ',\n",
+       " 658: b'ean ',\n",
+       " 659: b'Yes ',\n",
+       " 660: b'pre',\n",
+       " 661: b'7 ',\n",
+       " 662: b'pe',\n",
+       " 663: b'much ',\n",
+       " 664: b'cant ',\n",
+       " 665: b'da ',\n",
+       " 666: b'O ',\n",
+       " 667: b'How ',\n",
+       " 668: b'\\xf0\\x9f\\x9f',\n",
+       " 669: b'per',\n",
+       " 670: b'only ',\n",
+       " 671: b'\\xe2\\xac',\n",
+       " 672: b'maybe ',\n",
+       " 673: b'hat',\n",
+       " 674: b'ome',\n",
+       " 675: b'gon',\n",
+       " 676: b'actually ',\n",
+       " 677: b'ahh ',\n",
+       " 678: b'??',\n",
+       " 679: b'Ma',\n",
+       " 680: b'ks ',\n",
+       " 681: b'ear',\n",
+       " 682: b'sta',\n",
+       " 683: b'still ',\n",
+       " 684: b'\\xe2\\x96',\n",
+       " 685: b'va',\n",
+       " 686: b'see ',\n",
+       " 687: b'cus ',\n",
+       " 688: b'ke',\n",
+       " 689: b'as ',\n",
+       " 690: b'ound ',\n",
+       " 691: b'ran',\n",
+       " 692: b'And ',\n",
+       " 693: b'i think ',\n",
+       " 694: b'mon',\n",
+       " 695: b'mean ',\n",
+       " 696: b'10 ',\n",
+       " 697: b'{Reactions} \\xf0\\x9f',\n",
+       " 698: b'cra',\n",
+       " 699: b'\" ',\n",
+       " 700: b'own ',\n",
+       " 701: b'..',\n",
+       " 702: b'\\x92\\x80',\n",
+       " 703: b'gonna ',\n",
+       " 704: b\"'t \",\n",
+       " 705: b'ent',\n",
+       " 706: b'z ',\n",
+       " 707: b'fir',\n",
+       " 708: b'ss',\n",
+       " 709: b'zy ',\n",
+       " 710: b'ur',\n",
+       " 711: b'yo ',\n",
+       " 712: b'im',\n",
+       " 713: b'would ',\n",
+       " 714: b'guys ',\n",
+       " 715: b'movi',\n",
+       " 716: b'Huh ',\n",
+       " 717: b'Holy ',\n",
+       " 718: b'ope ',\n",
+       " 719: b'af',\n",
+       " 720: b'all',\n",
+       " 721: b'ded ',\n",
+       " 722: b'though ',\n",
+       " 723: b'- ',\n",
+       " 724: b'Ok ',\n",
+       " 725: b'jo',\n",
+       " 726: b'R ',\n",
+       " 727: b'8 ',\n",
+       " 728: b'\\xe2\\xac\\x9b',\n",
+       " 729: b'going ',\n",
+       " 730: b'K ',\n",
+       " 731: b'star',\n",
+       " 732: b'us ',\n",
+       " 733: b'guy ',\n",
+       " 734: b'ever',\n",
+       " 735: b'ble ',\n",
+       " 736: b'In',\n",
+       " 737: b'ur ',\n",
+       " 738: b'yaan ',\n",
+       " 739: b'make ',\n",
+       " 740: b'ool ',\n",
+       " 741: b'cking ',\n",
+       " 742: b'huh ',\n",
+       " 743: b'better ',\n",
+       " 744: b'loo',\n",
+       " 745: b'Id',\n",
+       " 746: b'il',\n",
+       " 747: b'Wait ',\n",
+       " 748: b'AN',\n",
+       " 749: b'any',\n",
+       " 750: b'! ',\n",
+       " 751: b'00 ',\n",
+       " 752: b'will ',\n",
+       " 753: b'the',\n",
+       " 754: b'\\xef\\xb8',\n",
+       " 755: b'\\xef\\xb8\\x8f',\n",
+       " 756: b'ard ',\n",
+       " 757: b'                                ',\n",
+       " 758: b'did',\n",
+       " 759: b'ers ',\n",
+       " 760: b'lowkey ',\n",
+       " 761: b'ok ',\n",
+       " 762: b'p on ',\n",
+       " 763: b'fr',\n",
+       " 764: b'could ',\n",
+       " 765: b'The ',\n",
+       " 766: b'him ',\n",
+       " 767: b'now',\n",
+       " 768: b'\\xf0\\x9f\\x9f\\xa9',\n",
+       " 769: b'movie ',\n",
+       " 770: b'want ',\n",
+       " 771: b'blo',\n",
+       " 772: b'pl',\n",
+       " 773: b'pe ',\n",
+       " 774: b'ide',\n",
+       " 775: b'Idk ',\n",
+       " 776: b'thats ',\n",
+       " 777: b'Al',\n",
+       " 778: b'Like ',\n",
+       " 779: b'Not ',\n",
+       " 780: b'after ',\n",
+       " 781: b'ever ',\n",
+       " 782: b'tting ',\n",
+       " 783: b'9 ',\n",
+       " 784: b'grou',\n",
+       " 785: b'ass ',\n",
+       " 786: b'pu',\n",
+       " 787: b'fe',\n",
+       " 788: b'Its ',\n",
+       " 789: b'sm',\n",
+       " 790: b'work ',\n",
+       " 791: b'tea',\n",
+       " 792: b'hmm ',\n",
+       " 793: b'piano ',\n",
+       " 794: b'really ',\n",
+       " 795: b'xt ',\n",
+       " 796: b'30 ',\n",
+       " 797: b'\\xe2\\x95',\n",
+       " 798: b'Is ',\n",
+       " 799: b'other ',\n",
+       " 800: b'first ',\n",
+       " 801: b'bro',\n",
+       " 802: b'ep',\n",
+       " 803: b'eat ',\n",
+       " 804: b'in the ',\n",
+       " 805: b'\\xe2\\x80\\x99m ',\n",
+       " 806: b'lf ',\n",
+       " 807: b'fin',\n",
+       " 808: b'ear ',\n",
+       " 809: b'man ',\n",
+       " 810: b'tal',\n",
+       " 811: b'G ',\n",
+       " 812: b'ha ',\n",
+       " 813: b'ed to ',\n",
+       " 814: b'ay ',\n",
+       " 815: b'has ',\n",
+       " 816: b'hop on ',\n",
+       " 817: b'tor',\n",
+       " 818: b'doing ',\n",
+       " 819: b'int ',\n",
+       " 820: b'are you ',\n",
+       " 821: b'Oh ',\n",
+       " 822: b'sch',\n",
+       " 823: b'as',\n",
+       " 824: b'... ',\n",
+       " 825: b'play',\n",
+       " 826: b'i can ',\n",
+       " 827: b'kin',\n",
+       " 828: b'some',\n",
+       " 829: b'minu',\n",
+       " 830: b'ms ',\n",
+       " 831: b'terally ',\n",
+       " 832: b'ment ',\n",
+       " 833: b'ed',\n",
+       " 834: b'chan',\n",
+       " 835: b'**',\n",
+       " 836: b'\\x92\\x80 ',\n",
+       " 837: b'where ',\n",
+       " 838: b'ig ',\n",
+       " 839: b'mp',\n",
+       " 840: b'hh',\n",
+       " 841: b'playing ',\n",
+       " 842: b'der ',\n",
+       " 843: b'He',\n",
+       " 844: b'can',\n",
+       " 845: b'ons ',\n",
+       " 846: b'sure ',\n",
+       " 847: b'ph',\n",
+       " 848: b'i have ',\n",
+       " 849: b'ad',\n",
+       " 850: b'whi',\n",
+       " 851: b'sy ',\n",
+       " 852: b'heem ',\n",
+       " 853: b'his ',\n",
+       " 854: b') ',\n",
+       " 855: b'sh ',\n",
+       " 856: b'rs ',\n",
+       " 857: b'terest',\n",
+       " 858: b'ople ',\n",
+       " 859: b'fucking ',\n",
+       " 860: b'ton ',\n",
+       " 861: b'W ',\n",
+       " 862: b'igh',\n",
+       " 863: b'20',\n",
+       " 864: b'M ',\n",
+       " 865: b'dam',\n",
+       " 866: b'Or ',\n",
+       " 867: b'ence ',\n",
+       " 868: b'ead ',\n",
+       " 869: b'same ',\n",
+       " 870: b'It ',\n",
+       " 871: b'wanna ',\n",
+       " 872: b'vro ',\n",
+       " 873: b'ON',\n",
+       " 874: b'wee',\n",
+       " 875: b'lun',\n",
+       " 876: b'fi',\n",
+       " 877: b'\\xe2\\xa2',\n",
+       " 878: b'tra',\n",
+       " 879: b'Lo',\n",
+       " 880: b'buy ',\n",
+       " 881: b'ice ',\n",
+       " 882: b'might ',\n",
+       " 883: b'Hmm ',\n",
+       " 884: b'ans ',\n",
+       " 885: b'erm ',\n",
+       " 886: b'eak ',\n",
+       " 887: b'\\xf0\\x9f\\x98\\xad\\xf0\\x9f\\x98\\xad',\n",
+       " 888: b'prob ',\n",
+       " 889: b'hehe ',\n",
+       " 890: b'people ',\n",
+       " 891: b'vide',\n",
+       " 892: b'cla',\n",
+       " 893: b'kinda ',\n",
+       " 894: b'Daniel ',\n",
+       " 895: b'ning ',\n",
+       " 896: b'fun',\n",
+       " 897: b'i dont ',\n",
+       " 898: b'ry ',\n",
+       " 899: b'mes ',\n",
+       " 900: b'rr ',\n",
+       " 901: b'uh',\n",
+       " 902: b'lar',\n",
+       " 903: b'fore ',\n",
+       " 904: b'ty ',\n",
+       " 905: b'ju',\n",
+       " 906: b'tes ',\n",
+       " 907: b'back ',\n",
+       " 908: b'mus',\n",
+       " 909: b'\\xe2\\xa0\\x80\\xe2\\xa0\\x80\\xe2\\xa0\\x80\\xe2\\xa0\\x80\\xe2\\xa0\\x80\\xe2\\xa0\\x80\\xe2\\xa0\\x80\\xe2\\xa0\\x80',\n",
+       " 910: b'wtf ',\n",
+       " 911: b'some ',\n",
+       " 912: b'crazy ',\n",
+       " 913: b'call ',\n",
+       " 914: b'happ',\n",
+       " 915: b'H ',\n",
+       " 916: b'ir ',\n",
+       " 917: b'ird ',\n",
+       " 918: b'does ',\n",
+       " 919: b'Im ',\n",
+       " 920: b'marcus ',\n",
+       " 921: b'Maybe ',\n",
+       " 922: b'literally ',\n",
+       " 923: b'didnt ',\n",
+       " 924: b'tch',\n",
+       " 925: b'LL',\n",
+       " 926: b'tm',\n",
+       " 927: b'Also ',\n",
+       " 928: b'coo',\n",
+       " 929: b'down ',\n",
+       " 930: b'teresting ',\n",
+       " 931: b'dum ',\n",
+       " 932: b'fo',\n",
+       " 933: b'ick ',\n",
+       " 934: b'ked ',\n",
+       " 935: b'before ',\n",
+       " 936: b'ways ',\n",
+       " 937: b'hard ',\n",
+       " 938: b'year ',\n",
+       " 939: b'ty',\n",
+       " 940: b'were ',\n",
+       " 941: b'hats ',\n",
+       " 942: b'ine',\n",
+       " 943: b'ball ',\n",
+       " 944: b'bad',\n",
+       " 945: b'ents ',\n",
+       " 946: b'very ',\n",
+       " 947: b'tar',\n",
+       " 948: b'dle ',\n",
+       " 949: b'The',\n",
+       " 950: b'My ',\n",
+       " 951: b'10',\n",
+       " 952: b'ort ',\n",
+       " 953: b'?? ',\n",
+       " 954: b'cor',\n",
+       " 955: b'v ',\n",
+       " 956: b'watch ',\n",
+       " 957: b'He ',\n",
+       " 958: b'der',\n",
+       " 959: b'Ethan ',\n",
+       " 960: b'ad ',\n",
+       " 961: b'ster ',\n",
+       " 962: b'inst',\n",
+       " 963: b'This ',\n",
+       " 964: b'come ',\n",
+       " 965: b'our ',\n",
+       " 966: b'vo',\n",
+       " 967: b'fuck ',\n",
+       " 968: b'car',\n",
+       " 969: b'cou',\n",
+       " 970: b'Pi',\n",
+       " 971: b'\\xe2\\xa3\\xbf',\n",
+       " 972: b'them ',\n",
+       " 973: b'had ',\n",
+       " 974: b'ps ',\n",
+       " 975: b'OU',\n",
+       " 976: b'han',\n",
+       " 977: b'been ',\n",
+       " 978: b'ki',\n",
+       " 979: b'lig',\n",
+       " 980: b'jj',\n",
+       " 981: b' S',\n",
+       " 982: b'sw',\n",
+       " 983: b'ayaan ',\n",
+       " 984: b'bab',\n",
+       " 985: b'tw ',\n",
+       " 986: b'Bro ',\n",
+       " 987: b'fri',\n",
+       " 988: b'ich ',\n",
+       " 989: b'ze ',\n",
+       " 990: b'\\xe2\\x80\\x99s ',\n",
+       " 991: b'every',\n",
+       " 992: b'ye ',\n",
+       " 993: b'gay ',\n",
+       " 994: b'never ',\n",
+       " 995: b'gs ',\n",
+       " 996: b'i was ',\n",
+       " 997: b'of the ',\n",
+       " 998: b'If ',\n",
+       " 999: b'ler ',\n",
+       " ...}"
+      ]
+     },
+     "execution_count": 8,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "vocab = tokenizer.vocab\n",
     "vocab"
@@ -88,31 +1118,42 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 9,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "[497, 820, 389, 121]"
+      ]
+     },
+     "execution_count": 9,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
-    "tokenizer.encode(\"Salam labas\")"
+    "tokenizer.encode(\"why are you gay\")"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 9,
+   "execution_count": 10,
    "metadata": {},
    "outputs": [
     {
      "data": {
       "text/plain": [
-       "'Salam labas'"
+       "'why are you gay'"
       ]
      },
-     "execution_count": 9,
+     "execution_count": 10,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
-    "tokenizer.decode([702, 310, 346, 115])"
+    "tokenizer.decode([497, 820, 389, 121])"
    ]
   },
   {
@@ -124,7 +1165,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 11,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -147,9 +1188,20 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 12,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "411925"
+      ]
+     },
+     "execution_count": 12,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "len(tokenizer.encode(text_sequence))"
    ]
@@ -163,7 +1215,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 12,
+   "execution_count": 13,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -173,7 +1225,7 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "vincent",
+   "display_name": "torch-rocm",
    "language": "python",
    "name": "python3"
   },
@@ -187,7 +1239,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.11.10"
+   "version": "3.10.20"
   }
  },
  "nbformat": 4,
diff --git a/notebooks/3_TransformerModel.ipynb b/notebooks/3_TransformerModel.ipynb
index b4131ae..7163df4 100644
--- a/notebooks/3_TransformerModel.ipynb
+++ b/notebooks/3_TransformerModel.ipynb
@@ -9,7 +9,7 @@
 		},
 		{
 			"cell_type": "code",
-			"execution_count": 1,
+			"execution_count": 12,
 			"metadata": {},
 			"outputs": [],
 			"source": [
@@ -19,7 +19,7 @@
 		},
 		{
 			"cell_type": "code",
-			"execution_count": null,
+			"execution_count": 13,
 			"metadata": {},
 			"outputs": [],
 			"source": [
@@ -31,7 +31,7 @@
 		},
 		{
 			"cell_type": "code",
-			"execution_count": null,
+			"execution_count": 14,
 			"metadata": {},
 			"outputs": [],
 			"source": [
@@ -138,7 +138,7 @@
 		},
 		{
 			"cell_type": "code",
-			"execution_count": null,
+			"execution_count": 15,
 			"metadata": {},
 			"outputs": [],
 			"source": [
@@ -163,7 +163,7 @@
 		},
 		{
 			"cell_type": "code",
-			"execution_count": null,
+			"execution_count": 16,
 			"metadata": {},
 			"outputs": [],
 			"source": [
@@ -214,7 +214,7 @@
 		},
 		{
 			"cell_type": "code",
-			"execution_count": null,
+			"execution_count": 17,
 			"metadata": {},
 			"outputs": [],
 			"source": [
@@ -233,6 +233,13 @@
 				"        return out"
 			]
 		},
+		{
+			"cell_type": "code",
+			"execution_count": null,
+			"metadata": {},
+			"outputs": [],
+			"source": []
+		},
 		{
 			"cell_type": "markdown",
 			"metadata": {},
@@ -242,7 +249,7 @@
 		},
 		{
 			"cell_type": "code",
-			"execution_count": null,
+			"execution_count": 18,
 			"metadata": {},
 			"outputs": [],
 			"source": [
@@ -289,7 +296,7 @@
 		},
 		{
 			"cell_type": "code",
-			"execution_count": null,
+			"execution_count": 19,
 			"metadata": {},
 			"outputs": [],
 			"source": [
@@ -388,18 +395,20 @@
 		},
 		{
 			"cell_type": "code",
-			"execution_count": null,
+			"execution_count": 20,
 			"metadata": {},
 			"outputs": [
 				{
 					"name": "stdout",
 					"output_type": "stream",
 					"text": [
-						"11.532552 M parameters\n"
+						"11.53409 M parameters\n"
 					]
 				}
 			],
 			"source": [
+				"\n",
+				"\n",
 				"model = GPTLanguageModel()\n",
 				"model = model.to(device)\n",
 				"# print the number of parameters in the model\n",
@@ -408,14 +417,14 @@
 		},
 		{
 			"cell_type": "code",
-			"execution_count": null,
+			"execution_count": 21,
 			"metadata": {},
 			"outputs": [
 				{
 					"name": "stdout",
 					"output_type": "stream",
 					"text": [
-						"torch.Size([1, 6, 1032]) None\n"
+						"torch.Size([1, 6, 1034]) None\n"
 					]
 				}
 			],
@@ -438,14 +447,14 @@
 		},
 		{
 			"cell_type": "code",
-			"execution_count": null,
+			"execution_count": 22,
 			"metadata": {},
 			"outputs": [
 				{
 					"name": "stdout",
 					"output_type": "stream",
 					"text": [
-						"├─ token_embedding_table: Embedding (396,288 parameters)\n",
+						"├─ token_embedding_table: Embedding (397,056 parameters)\n",
 						"├─ position_embedding_table: Embedding (98,304 parameters)\n",
 						"├─ blocks: Sequential (10,639,872 parameters)\n",
 						"│  ├─ 0: Block (1,773,312 parameters)\n",
@@ -707,7 +716,7 @@
 						"│  │  ├─ layer_norm_1: LayerNorm (768 parameters)\n",
 						"│  │  ├─ layer_norm_2: LayerNorm (768 parameters)\n",
 						"├─ final_layer_norm: LayerNorm (768 parameters)\n",
-						"├─ final_linear_layer: Linear (397,320 parameters)\n"
+						"├─ final_linear_layer: Linear (398,090 parameters)\n"
 					]
 				}
 			],
@@ -727,398 +736,11 @@
 		},
 		{
 			"cell_type": "code",
-			"execution_count": null,
+			"execution_count": 23,
 			"metadata": {},
 			"outputs": [
 				{
 					"data": {
-						"application/vnd.microsoft.datawrangler.viewer.v0+json": {
-							"columns": [
-								{
-									"name": "index",
-									"rawType": "int64",
-									"type": "integer"
-								},
-								{
-									"name": "Layer Name",
-									"rawType": "object",
-									"type": "string"
-								},
-								{
-									"name": "Type",
-									"rawType": "object",
-									"type": "string"
-								},
-								{
-									"name": "Parameters",
-									"rawType": "int64",
-									"type": "integer"
-								},
-								{
-									"name": "Trainable",
-									"rawType": "int64",
-									"type": "integer"
-								}
-							],
-							"conversionMethod": "pd.DataFrame",
-							"ref": "41f6a192-649f-4276-8acd-5e81c92f3aa9",
-							"rows": [
-								[
-									"0",
-									"token_embedding_table",
-									"Embedding",
-									"396288",
-									"396288"
-								],
-								[
-									"1",
-									"position_embedding_table",
-									"Embedding",
-									"98304",
-									"98304"
-								],
-								[
-									"2",
-									"blocks.0.self_attention.heads.0.key",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"3",
-									"blocks.0.self_attention.heads.0.query",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"4",
-									"blocks.0.self_attention.heads.0.value",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"5",
-									"blocks.0.self_attention.heads.0.dropout",
-									"Dropout",
-									"0",
-									"0"
-								],
-								[
-									"6",
-									"blocks.0.self_attention.heads.1.key",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"7",
-									"blocks.0.self_attention.heads.1.query",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"8",
-									"blocks.0.self_attention.heads.1.value",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"9",
-									"blocks.0.self_attention.heads.1.dropout",
-									"Dropout",
-									"0",
-									"0"
-								],
-								[
-									"10",
-									"blocks.0.self_attention.heads.2.key",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"11",
-									"blocks.0.self_attention.heads.2.query",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"12",
-									"blocks.0.self_attention.heads.2.value",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"13",
-									"blocks.0.self_attention.heads.2.dropout",
-									"Dropout",
-									"0",
-									"0"
-								],
-								[
-									"14",
-									"blocks.0.self_attention.heads.3.key",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"15",
-									"blocks.0.self_attention.heads.3.query",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"16",
-									"blocks.0.self_attention.heads.3.value",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"17",
-									"blocks.0.self_attention.heads.3.dropout",
-									"Dropout",
-									"0",
-									"0"
-								],
-								[
-									"18",
-									"blocks.0.self_attention.heads.4.key",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"19",
-									"blocks.0.self_attention.heads.4.query",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"20",
-									"blocks.0.self_attention.heads.4.value",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"21",
-									"blocks.0.self_attention.heads.4.dropout",
-									"Dropout",
-									"0",
-									"0"
-								],
-								[
-									"22",
-									"blocks.0.self_attention.heads.5.key",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"23",
-									"blocks.0.self_attention.heads.5.query",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"24",
-									"blocks.0.self_attention.heads.5.value",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"25",
-									"blocks.0.self_attention.heads.5.dropout",
-									"Dropout",
-									"0",
-									"0"
-								],
-								[
-									"26",
-									"blocks.0.self_attention.projection",
-									"Linear",
-									"147840",
-									"147840"
-								],
-								[
-									"27",
-									"blocks.0.self_attention.dropout",
-									"Dropout",
-									"0",
-									"0"
-								],
-								[
-									"28",
-									"blocks.0.feed_forward.net.0",
-									"Linear",
-									"591360",
-									"591360"
-								],
-								[
-									"29",
-									"blocks.0.feed_forward.net.1",
-									"ReLU",
-									"0",
-									"0"
-								],
-								[
-									"30",
-									"blocks.0.feed_forward.net.2",
-									"Linear",
-									"590208",
-									"590208"
-								],
-								[
-									"31",
-									"blocks.0.feed_forward.net.3",
-									"Dropout",
-									"0",
-									"0"
-								],
-								[
-									"32",
-									"blocks.0.layer_norm_1",
-									"LayerNorm",
-									"768",
-									"768"
-								],
-								[
-									"33",
-									"blocks.0.layer_norm_2",
-									"LayerNorm",
-									"768",
-									"768"
-								],
-								[
-									"34",
-									"blocks.1.self_attention.heads.0.key",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"35",
-									"blocks.1.self_attention.heads.0.query",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"36",
-									"blocks.1.self_attention.heads.0.value",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"37",
-									"blocks.1.self_attention.heads.0.dropout",
-									"Dropout",
-									"0",
-									"0"
-								],
-								[
-									"38",
-									"blocks.1.self_attention.heads.1.key",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"39",
-									"blocks.1.self_attention.heads.1.query",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"40",
-									"blocks.1.self_attention.heads.1.value",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"41",
-									"blocks.1.self_attention.heads.1.dropout",
-									"Dropout",
-									"0",
-									"0"
-								],
-								[
-									"42",
-									"blocks.1.self_attention.heads.2.key",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"43",
-									"blocks.1.self_attention.heads.2.query",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"44",
-									"blocks.1.self_attention.heads.2.value",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"45",
-									"blocks.1.self_attention.heads.2.dropout",
-									"Dropout",
-									"0",
-									"0"
-								],
-								[
-									"46",
-									"blocks.1.self_attention.heads.3.key",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"47",
-									"blocks.1.self_attention.heads.3.query",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"48",
-									"blocks.1.self_attention.heads.3.value",
-									"Linear",
-									"24576",
-									"24576"
-								],
-								[
-									"49",
-									"blocks.1.self_attention.heads.3.dropout",
-									"Dropout",
-									"0",
-									"0"
-								]
-							],
-							"shape": {
-								"columns": 4,
-								"rows": 196
-							}
-						},
 						"text/html": [
 							"<div>\n",
 							"<style scoped>\n",
@@ -1149,8 +771,8 @@
 							"      <th>0</th>\n",
 							"      <td>token_embedding_table</td>\n",
 							"      <td>Embedding</td>\n",
-							"      <td>396288</td>\n",
-							"      <td>396288</td>\n",
+							"      <td>397056</td>\n",
+							"      <td>397056</td>\n",
 							"    </tr>\n",
 							"    <tr>\n",
 							"      <th>1</th>\n",
@@ -1219,8 +841,8 @@
 							"      <th>195</th>\n",
 							"      <td>final_linear_layer</td>\n",
 							"      <td>Linear</td>\n",
-							"      <td>397320</td>\n",
-							"      <td>397320</td>\n",
+							"      <td>398090</td>\n",
+							"      <td>398090</td>\n",
 							"    </tr>\n",
 							"  </tbody>\n",
 							"</table>\n",
@@ -1229,7 +851,7 @@
 						],
 						"text/plain": [
 							"                                Layer Name       Type  Parameters  Trainable\n",
-							"0                    token_embedding_table  Embedding      396288     396288\n",
+							"0                    token_embedding_table  Embedding      397056     397056\n",
 							"1                 position_embedding_table  Embedding       98304      98304\n",
 							"2      blocks.0.self_attention.heads.0.key     Linear       24576      24576\n",
 							"3    blocks.0.self_attention.heads.0.query     Linear       24576      24576\n",
@@ -1239,12 +861,12 @@
 							"192                  blocks.5.layer_norm_1  LayerNorm         768        768\n",
 							"193                  blocks.5.layer_norm_2  LayerNorm         768        768\n",
 							"194                       final_layer_norm  LayerNorm         768        768\n",
-							"195                     final_linear_layer     Linear      397320     397320\n",
+							"195                     final_linear_layer     Linear      398090     398090\n",
 							"\n",
 							"[196 rows x 4 columns]"
 						]
 					},
-					"execution_count": 32,
+					"execution_count": 23,
 					"metadata": {},
 					"output_type": "execute_result"
 				}
@@ -1277,7 +899,7 @@
 	],
 	"metadata": {
 		"kernelspec": {
-			"display_name": "vincent",
+			"display_name": "Python 3",
 			"language": "python",
 			"name": "python3"
 		},
@@ -1291,7 +913,7 @@
 			"name": "python",
 			"nbconvert_exporter": "python",
 			"pygments_lexer": "ipython3",
-			"version": "3.11.10"
+			"version": "3.12.10"
 		}
 	},
 	"nbformat": 4,
diff --git a/notebooks/4_1_ModelTrainingAllBatches.ipynb b/notebooks/4_1_ModelTrainingAllBatches.ipynb
index ba7aa8e..6a3481e 100644
--- a/notebooks/4_1_ModelTrainingAllBatches.ipynb
+++ b/notebooks/4_1_ModelTrainingAllBatches.ipynb
@@ -9,7 +9,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 5,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -19,7 +19,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 6,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -45,9 +45,20 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 7,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "<torch._C.Generator at 0x193abd89fd0>"
+      ]
+     },
+     "execution_count": 7,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "import torch\n",
     "torch.manual_seed(3647)"
@@ -55,9 +66,17 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 8,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "13.795338 M parameters\n"
+     ]
+    }
+   ],
    "source": [
     "from transformer.model import GPTLanguageModel\n",
     "\n",
@@ -79,7 +98,8 @@
     "    dropout=dropout,\n",
     "    device=device\n",
     ").to(device)\n",
-    "model = torch.compile(model)\n",
+    "model = getattr(model, \"__wrapped__\", model)  # Ensure the raw model object is used\n",
+    "# model = torch.compile(model)  # Disabled on Windows unless MSVC compiler is installed\n",
     "\n",
     "print(sum(p.numel() for p in model.parameters())/1e6, 'M parameters')"
    ]
@@ -100,9 +120,20 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 9,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "411925"
+      ]
+     },
+     "execution_count": 9,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "with open(\"../output/combined_text.txt\", \"r\", encoding='utf-8') as f:\n",
     "    text_sequence = f.read()\n",
@@ -120,7 +151,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 10,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -139,7 +170,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 11,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -192,9 +223,20 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 12,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "(torch.Size([64, 256]), torch.Size([64, 256]))"
+      ]
+     },
+     "execution_count": 12,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "train_loader, val_loader = get_dataloaders(\n",
     "    train_data=train_data,\n",
@@ -216,7 +258,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 13,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -249,7 +291,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 14,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -271,13 +313,70 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 26,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "iteration 0 / step 0: train loss 4.8303, val loss 4.7976\n",
+      "iteration 0 / step 1: train loss 5.0402, val loss 5.0015\n",
+      "iteration 0 / step 2: train loss 4.9341, val loss 4.9175\n",
+      "iteration 0 / step 3: train loss 4.8912, val loss 4.8635\n",
+      "iteration 0 / step 4: train loss 4.8736, val loss 4.8194\n",
+      "iteration 0 / step 5: train loss 4.8610, val loss 4.8014\n",
+      "iteration 0 / step 6: train loss 4.8296, val loss 4.7966\n",
+      "iteration 0 / step 7: train loss 4.8389, val loss 4.7861\n",
+      "iteration 0 / step 8: train loss 4.8129, val loss 4.7714\n",
+      "iteration 0 / step 9: train loss 4.8052, val loss 4.7603\n",
+      "iteration 0 / step 10: train loss 4.7945, val loss 4.7531\n",
+      "iteration 0 / step 11: train loss 4.7880, val loss 4.7465\n",
+      "iteration 0 / step 12: train loss 4.7821, val loss 4.7406\n",
+      "iteration 0 / step 13: train loss 4.7689, val loss 4.7326\n",
+      "iteration 0 / step 14: train loss 4.7581, val loss 4.7270\n",
+      "iteration 0 / step 15: train loss 4.7592, val loss 4.7203\n",
+      "iteration 0 / step 16: train loss 4.7498, val loss 4.7146\n",
+      "iteration 0 / step 17: train loss 4.7439, val loss 4.7088\n",
+      "iteration 0 / step 18: train loss 4.7342, val loss 4.7029\n",
+      "iteration 0 / step 19: train loss 4.7275, val loss 4.6971\n"
+     ]
+    },
+    {
+     "ename": "KeyboardInterrupt",
+     "evalue": "",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[31m---------------------------------------------------------------------------\u001b[39m",
+      "\u001b[31mKeyboardInterrupt\u001b[39m                         Traceback (most recent call last)",
+      "\u001b[36mCell\u001b[39m\u001b[36m \u001b[39m\u001b[32mIn[26]\u001b[39m\u001b[32m, line 26\u001b[39m\n\u001b[32m     22\u001b[39m \u001b[38;5;28;01mfor\u001b[39;00m iteration \u001b[38;5;28;01min\u001b[39;00m range(max_iters):\n\u001b[32m     23\u001b[39m     \u001b[38;5;28;01mfor\u001b[39;00m batch_idx, (x_batch, y_batch) \u001b[38;5;28;01min\u001b[39;00m enumerate(train_loader):\n\u001b[32m     24\u001b[39m         \u001b[38;5;66;03m# Evaluation\u001b[39;00m\n\u001b[32m     25\u001b[39m         \u001b[38;5;28;01mif\u001b[39;00m batch_idx % eval_interval == \u001b[32m0\u001b[39m \u001b[38;5;28;01mor\u001b[39;00m batch_idx == len(train_loader) - \u001b[32m1\u001b[39m:\n\u001b[32m---> \u001b[39m\u001b[32m26\u001b[39m             losses = estimate_loss(\n\u001b[32m     27\u001b[39m                 model=model,\n\u001b[32m     28\u001b[39m                 train_loader=train_loader,\n\u001b[32m     29\u001b[39m                 val_loader=val_loader,\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\utils\\_contextlib.py:116\u001b[39m, in \u001b[36mcontext_decorator.<locals>.decorate_context\u001b[39m\u001b[34m(*args, **kwargs)\u001b[39m\n\u001b[32m    113\u001b[39m \u001b[38;5;129m@functools\u001b[39m.wraps(func)\n\u001b[32m    114\u001b[39m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34mdecorate_context\u001b[39m(*args, **kwargs):\n\u001b[32m    115\u001b[39m     \u001b[38;5;28;01mwith\u001b[39;00m ctx_factory():\n\u001b[32m--> \u001b[39m\u001b[32m116\u001b[39m         \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mfunc\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43margs\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43mkwargs\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n",
+      "\u001b[36mCell\u001b[39m\u001b[36m \u001b[39m\u001b[32mIn[13]\u001b[39m\u001b[32m, line 20\u001b[39m, in \u001b[36mestimate_loss\u001b[39m\u001b[34m(model, train_loader, val_loader, eval_iters)\u001b[39m\n\u001b[32m     16\u001b[39m         \u001b[38;5;28;01mfor\u001b[39;00m i, (x, y) \u001b[38;5;28;01min\u001b[39;00m enumerate(loader):\n\u001b[32m     17\u001b[39m             \u001b[38;5;28;01mif\u001b[39;00m i >= eval_iters:\n\u001b[32m     18\u001b[39m                 \u001b[38;5;28;01mbreak\u001b[39;00m\n\u001b[32m     19\u001b[39m             \u001b[38;5;28;01mwith\u001b[39;00m torch.no_grad():\n\u001b[32m---> \u001b[39m\u001b[32m20\u001b[39m                 _, loss = model(x, y)\n\u001b[32m     21\u001b[39m             losses[i] = loss.item()\n\u001b[32m     22\u001b[39m         output[split] = losses.mean().item()\n\u001b[32m     23\u001b[39m \n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\modules\\module.py:1736\u001b[39m, in \u001b[36mModule._wrapped_call_impl\u001b[39m\u001b[34m(self, *args, **kwargs)\u001b[39m\n\u001b[32m   1734\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m._compiled_call_impl(*args, **kwargs)  \u001b[38;5;66;03m# type: ignore[misc]\u001b[39;00m\n\u001b[32m   1735\u001b[39m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[32m-> \u001b[39m\u001b[32m1736\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43m_call_impl\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43margs\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43mkwargs\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\modules\\module.py:1747\u001b[39m, in \u001b[36mModule._call_impl\u001b[39m\u001b[34m(self, *args, **kwargs)\u001b[39m\n\u001b[32m   1742\u001b[39m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[32m   1743\u001b[39m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[32m   1744\u001b[39m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m._backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._forward_pre_hooks\n\u001b[32m   1745\u001b[39m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[32m   1746\u001b[39m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[32m-> \u001b[39m\u001b[32m1747\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mforward_call\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43margs\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43mkwargs\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n\u001b[32m   1749\u001b[39m result = \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[32m   1750\u001b[39m called_always_called_hooks = \u001b[38;5;28mset\u001b[39m()\n",
+      "\u001b[36mFile \u001b[39m\u001b[32md:\\Train_Your_Language_Model_Course-clean\\notebooks\\..\\transformer\\model.py:136\u001b[39m, in \u001b[36mGPTLanguageModel.forward\u001b[39m\u001b[34m(self, input_tokens, targets)\u001b[39m\n\u001b[32m    133\u001b[39m positional_embedding = \u001b[38;5;28mself\u001b[39m.position_embedding_table(\n\u001b[32m    134\u001b[39m     torch.arange(T, device=\u001b[38;5;28mself\u001b[39m.device))\n\u001b[32m    135\u001b[39m x = token_embedding + positional_embedding\n\u001b[32m--> \u001b[39m\u001b[32m136\u001b[39m x = \u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mblocks\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43mx\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n\u001b[32m    137\u001b[39m x = \u001b[38;5;28mself\u001b[39m.final_layer_norm(x)\n\u001b[32m    138\u001b[39m logits = \u001b[38;5;28mself\u001b[39m.final_linear_layer(x)\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\modules\\module.py:1736\u001b[39m, in \u001b[36mModule._wrapped_call_impl\u001b[39m\u001b[34m(self, *args, **kwargs)\u001b[39m\n\u001b[32m   1734\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m._compiled_call_impl(*args, **kwargs)  \u001b[38;5;66;03m# type: ignore[misc]\u001b[39;00m\n\u001b[32m   1735\u001b[39m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[32m-> \u001b[39m\u001b[32m1736\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43m_call_impl\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43margs\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43mkwargs\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\modules\\module.py:1747\u001b[39m, in \u001b[36mModule._call_impl\u001b[39m\u001b[34m(self, *args, **kwargs)\u001b[39m\n\u001b[32m   1742\u001b[39m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[32m   1743\u001b[39m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[32m   1744\u001b[39m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m._backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._forward_pre_hooks\n\u001b[32m   1745\u001b[39m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[32m   1746\u001b[39m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[32m-> \u001b[39m\u001b[32m1747\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mforward_call\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43margs\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43mkwargs\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n\u001b[32m   1749\u001b[39m result = \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[32m   1750\u001b[39m called_always_called_hooks = \u001b[38;5;28mset\u001b[39m()\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\modules\\container.py:250\u001b[39m, in \u001b[36mSequential.forward\u001b[39m\u001b[34m(self, input)\u001b[39m\n\u001b[32m    248\u001b[39m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34mforward\u001b[39m(\u001b[38;5;28mself\u001b[39m, \u001b[38;5;28minput\u001b[39m):\n\u001b[32m    249\u001b[39m     \u001b[38;5;28;01mfor\u001b[39;00m module \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mself\u001b[39m:\n\u001b[32m--> \u001b[39m\u001b[32m250\u001b[39m         \u001b[38;5;28minput\u001b[39m = \u001b[30;43mmodule\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43minput\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n\u001b[32m    251\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28minput\u001b[39m\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\modules\\module.py:1736\u001b[39m, in \u001b[36mModule._wrapped_call_impl\u001b[39m\u001b[34m(self, *args, **kwargs)\u001b[39m\n\u001b[32m   1734\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m._compiled_call_impl(*args, **kwargs)  \u001b[38;5;66;03m# type: ignore[misc]\u001b[39;00m\n\u001b[32m   1735\u001b[39m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[32m-> \u001b[39m\u001b[32m1736\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43m_call_impl\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43margs\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43mkwargs\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\modules\\module.py:1747\u001b[39m, in \u001b[36mModule._call_impl\u001b[39m\u001b[34m(self, *args, **kwargs)\u001b[39m\n\u001b[32m   1742\u001b[39m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[32m   1743\u001b[39m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[32m   1744\u001b[39m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m._backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._forward_pre_hooks\n\u001b[32m   1745\u001b[39m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[32m   1746\u001b[39m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[32m-> \u001b[39m\u001b[32m1747\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mforward_call\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43margs\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43mkwargs\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n\u001b[32m   1749\u001b[39m result = \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[32m   1750\u001b[39m called_always_called_hooks = \u001b[38;5;28mset\u001b[39m()\n",
+      "\u001b[36mFile \u001b[39m\u001b[32md:\\Train_Your_Language_Model_Course-clean\\notebooks\\..\\transformer\\model.py:87\u001b[39m, in \u001b[36mBlock.forward\u001b[39m\u001b[34m(self, x)\u001b[39m\n\u001b[32m     86\u001b[39m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34mforward\u001b[39m(\u001b[38;5;28mself\u001b[39m, x: torch.Tensor) -> torch.Tensor:\n\u001b[32m---> \u001b[39m\u001b[32m87\u001b[39m     x = x + \u001b[38;5;28mself\u001b[39m.self_attention(\u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mlayer_norm_1\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43mx\u001b[39;49m\u001b[30;43m)\u001b[39;49m)\n\u001b[32m     88\u001b[39m     x = x + \u001b[38;5;28mself\u001b[39m.feed_forward(\u001b[38;5;28mself\u001b[39m.layer_norm_2(x))\n\u001b[32m     89\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m x\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\modules\\module.py:1736\u001b[39m, in \u001b[36mModule._wrapped_call_impl\u001b[39m\u001b[34m(self, *args, **kwargs)\u001b[39m\n\u001b[32m   1734\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m._compiled_call_impl(*args, **kwargs)  \u001b[38;5;66;03m# type: ignore[misc]\u001b[39;00m\n\u001b[32m   1735\u001b[39m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[32m-> \u001b[39m\u001b[32m1736\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43m_call_impl\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43margs\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43mkwargs\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\modules\\module.py:1747\u001b[39m, in \u001b[36mModule._call_impl\u001b[39m\u001b[34m(self, *args, **kwargs)\u001b[39m\n\u001b[32m   1742\u001b[39m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[32m   1743\u001b[39m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[32m   1744\u001b[39m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m._backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._forward_pre_hooks\n\u001b[32m   1745\u001b[39m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[32m   1746\u001b[39m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[32m-> \u001b[39m\u001b[32m1747\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mforward_call\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43margs\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43mkwargs\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n\u001b[32m   1749\u001b[39m result = \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[32m   1750\u001b[39m called_always_called_hooks = \u001b[38;5;28mset\u001b[39m()\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\modules\\normalization.py:217\u001b[39m, in \u001b[36mLayerNorm.forward\u001b[39m\u001b[34m(self, input)\u001b[39m\n\u001b[32m    216\u001b[39m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34mforward\u001b[39m(\u001b[38;5;28mself\u001b[39m, \u001b[38;5;28minput\u001b[39m: Tensor) -> Tensor:\n\u001b[32m--> \u001b[39m\u001b[32m217\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mF\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mlayer_norm\u001b[39;49m\u001b[30;43m(\u001b[39;49m\n\u001b[32m    218\u001b[39m \u001b[30;43m        \u001b[39;49m\u001b[30;43minput\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mnormalized_shape\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mweight\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mbias\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43meps\u001b[39;49m\n\u001b[32m    219\u001b[39m \u001b[30;43m    \u001b[39;49m\u001b[30;43m)\u001b[39;49m\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\functional.py:2900\u001b[39m, in \u001b[36mlayer_norm\u001b[39m\u001b[34m(input, normalized_shape, weight, bias, eps)\u001b[39m\n\u001b[32m   2890\u001b[39m \u001b[38;5;28;01mif\u001b[39;00m has_torch_function_variadic(\u001b[38;5;28minput\u001b[39m, weight, bias):\n\u001b[32m   2891\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m handle_torch_function(\n\u001b[32m   2892\u001b[39m         layer_norm,\n\u001b[32m   2893\u001b[39m         (\u001b[38;5;28minput\u001b[39m, weight, bias),\n\u001b[32m   (...)\u001b[39m\u001b[32m   2898\u001b[39m         eps=eps,\n\u001b[32m   2899\u001b[39m     )\n\u001b[32m-> \u001b[39m\u001b[32m2900\u001b[39m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mtorch\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mlayer_norm\u001b[39;49m\u001b[30;43m(\u001b[39;49m\n\u001b[32m   2901\u001b[39m \u001b[30;43m    \u001b[39;49m\u001b[30;43minput\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43mnormalized_shape\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43mweight\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43mbias\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43meps\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43mtorch\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mbackends\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mcudnn\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43menabled\u001b[39;49m\n\u001b[32m   2902\u001b[39m \u001b[30;43m\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n",
+      "\u001b[31mKeyboardInterrupt\u001b[39m: "
+     ]
+    }
+   ],
    "source": [
-    "max_iters = 1\n",
-    "eval_interval = 100\n",
-    "eval_iters = 200\n",
+    "import torch\n",
+    "import torch.nn as nn\n",
+    "import os\n",
+    "os.makedirs(\"../output/pre_training/run_4\", exist_ok=True)\n",
+    "max_iters = 50\n",
+    "eval_interval = 1\n",
+    "eval_iters = 50\n",
     "learning_rate = 3e-4\n",
     "\n",
     "optimizer = torch.optim.AdamW(model.parameters(), lr=learning_rate)\n",
@@ -288,7 +387,7 @@
     "    batch_size=batch_size,\n",
     "    device=device\n",
     ")\n",
-    "\n",
+    "model.train()\n",
     "train_losses = []\n",
     "val_losses = []\n",
     "\n",
@@ -329,9 +428,20 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 27,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "image/png": "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",
+      "text/plain": [
+       "<Figure size 1000x500 with 1 Axes>"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
    "source": [
     "import matplotlib.pyplot as plt\n",
     "\n",
@@ -348,11 +458,19 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 29,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "vroll in Fesoon from haictickfor an what is that We is and detic like Eo Thame Its chat if If you out Oh Is When ting shirl how not like if aheem were vieed to spon? I canss u short inder max have more pretty Im gay which pintorithing first should i guy To can just do that pretty too much min bro should row \n"
+     ]
+    }
+   ],
    "source": [
-    "input_tokens = tokenizer.encode(\"Salam labas \")\n",
+    "input_tokens = tokenizer.encode(\"vro\")\n",
     "input_tokens = torch.tensor(\n",
     "    input_tokens, dtype=torch.long).unsqueeze(0).to(device)\n",
     "\n",
@@ -380,7 +498,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.11.9"
+   "version": "3.12.10"
   }
  },
  "nbformat": 4,
diff --git a/notebooks/4_2_ModelTrainingRandomBatches.ipynb b/notebooks/4_2_ModelTrainingRandomBatches.ipynb
index c2894bd..e35b409 100644
--- a/notebooks/4_2_ModelTrainingRandomBatches.ipynb
+++ b/notebooks/4_2_ModelTrainingRandomBatches.ipynb
@@ -9,7 +9,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 2,
+   "execution_count": 1,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -19,7 +19,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 3,
+   "execution_count": 2,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -45,9 +45,20 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 3,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "<torch._C.Generator at 0x210b5674710>"
+      ]
+     },
+     "execution_count": 3,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "import torch\n",
     "torch.manual_seed(3647)"
@@ -57,7 +68,21 @@
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "ename": "AssertionError",
+     "evalue": "n_embd 512 must be divisible by n_head 12",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[31m---------------------------------------------------------------------------\u001b[39m",
+      "\u001b[31mAssertionError\u001b[39m                            Traceback (most recent call last)",
+      "\u001b[36mCell\u001b[39m\u001b[36m \u001b[39m\u001b[32mIn[4]\u001b[39m\u001b[32m, line 12\u001b[39m\n\u001b[32m      8\u001b[39m batch_size = \u001b[32m64\u001b[39m\n\u001b[32m      9\u001b[39m vocab_size = get_vocab_size(tokenizer)\n\u001b[32m     10\u001b[39m device = \u001b[33m'cuda'\u001b[39m \u001b[38;5;28;01mif\u001b[39;00m torch.cuda.is_available() \u001b[38;5;28;01melse\u001b[39;00m \u001b[33m'cpu'\u001b[39m\n\u001b[32m     11\u001b[39m \n\u001b[32m---> \u001b[39m\u001b[32m12\u001b[39m model = GPTLanguageModel(\n\u001b[32m     13\u001b[39m     vocab_size=vocab_size,\n\u001b[32m     14\u001b[39m     block_size=block_size,\n\u001b[32m     15\u001b[39m     n_embd=n_embd,\n",
+      "\u001b[36mFile \u001b[39m\u001b[32md:\\Train_Your_Language_Model_Course-clean\\notebooks\\..\\transformer\\model.py:112\u001b[39m, in \u001b[36mGPTLanguageModel.__init__\u001b[39m\u001b[34m(self, vocab_size, n_embd, n_head, block_size, n_layer, dropout, device, ignore_index)\u001b[39m\n\u001b[32m    109\u001b[39m \u001b[38;5;28mself\u001b[39m.token_embedding_table = nn.Embedding(vocab_size, n_embd)\n\u001b[32m    110\u001b[39m \u001b[38;5;28mself\u001b[39m.position_embedding_table = nn.Embedding(block_size, n_embd)\n\u001b[32m    111\u001b[39m \u001b[38;5;28mself\u001b[39m.blocks = nn.Sequential(*[\n\u001b[32m--> \u001b[39m\u001b[32m112\u001b[39m     \u001b[30;43mBlock\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43mn_embd\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43mn_head\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43mblock_size\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43mdropout\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n\u001b[32m    113\u001b[39m     \u001b[38;5;28;01mfor\u001b[39;00m _ \u001b[38;5;129;01min\u001b[39;00m \u001b[38;5;28mrange\u001b[39m(n_layer)\n\u001b[32m    114\u001b[39m ])\n\u001b[32m    115\u001b[39m \u001b[38;5;28mself\u001b[39m.final_layer_norm = nn.LayerNorm(n_embd)\n\u001b[32m    116\u001b[39m \u001b[38;5;28mself\u001b[39m.final_linear_layer = nn.Linear(n_embd, vocab_size)\n",
+      "\u001b[36mFile \u001b[39m\u001b[32md:\\Train_Your_Language_Model_Course-clean\\notebooks\\..\\transformer\\model.py:74\u001b[39m, in \u001b[36mBlock.__init__\u001b[39m\u001b[34m(self, n_embd, n_head, block_size, dropout)\u001b[39m\n\u001b[32m     72\u001b[39m head_size = n_embd // n_head\n\u001b[32m     73\u001b[39m error_message = \u001b[33mf\u001b[39m\u001b[33m\"\u001b[39m\u001b[33mn_embd \u001b[39m\u001b[38;5;132;01m{\u001b[39;00mn_embd\u001b[38;5;132;01m}\u001b[39;00m\u001b[33m must be divisible by n_head \u001b[39m\u001b[38;5;132;01m{\u001b[39;00mn_head\u001b[38;5;132;01m}\u001b[39;00m\u001b[33m\"\u001b[39m\n\u001b[32m---> \u001b[39m\u001b[32m74\u001b[39m \u001b[38;5;28;01massert\u001b[39;00m head_size * n_head == n_embd, error_message\n\u001b[32m     75\u001b[39m \u001b[38;5;28mself\u001b[39m.self_attention = MultiHeadAttention(\n\u001b[32m     76\u001b[39m     n_embd=n_embd,\n\u001b[32m     77\u001b[39m     num_heads=n_head,\n\u001b[32m   (...)\u001b[39m\u001b[32m     80\u001b[39m     dropout=dropout\n\u001b[32m     81\u001b[39m )\n\u001b[32m     82\u001b[39m \u001b[38;5;28mself\u001b[39m.feed_forward = FeedForward(n_embd, dropout)\n",
+      "\u001b[31mAssertionError\u001b[39m: n_embd 512 must be divisible by n_head 12"
+     ]
+    }
+   ],
    "source": [
     "from transformer.model import GPTLanguageModel\n",
     "\n",
@@ -79,7 +104,7 @@
     "    dropout=dropout,\n",
     "    device=device\n",
     ").to(device)\n",
-    "model = torch.compile(model)\n",
+    "# model = torch.compile(model)  # Disabled on Windows unless the MSVC C++ compiler is installed\n",
     "\n",
     "print(sum(p.numel() for p in model.parameters())/1e6, 'M parameters')"
    ]
@@ -100,12 +125,24 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 17,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "411925"
+      ]
+     },
+     "execution_count": 17,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "with open(\"../output/combined_text.txt\", \"r\") as f:\n",
-    "    text_sequence = f.read()\n",
+    "    with open(\"../output/combined_text.txt\", \"r\", encoding=\"utf-8\") as f:\n",
+    "        text_sequence = f.read()\n",
     "\n",
     "encoded_text_sequence = tokenizer.encode(text_sequence)\n",
     "len(encoded_text_sequence)"
@@ -120,7 +157,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 10,
+   "execution_count": 6,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -139,7 +176,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 11,
+   "execution_count": 7,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -158,9 +195,20 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 8,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "(torch.Size([128, 512]), torch.Size([128, 512]))"
+      ]
+     },
+     "execution_count": 8,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "x, y = get_batch('train')\n",
     "x.shape, y.shape"
@@ -175,7 +223,7 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 13,
+   "execution_count": 9,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -194,14 +242,14 @@
     "            x, y = get_batch(split)\n",
     "            _, loss = model(x, y)\n",
     "            losses[k] = loss.item()\n",
-    "        output[split] = losses.mean()\n",
+    "        output[split] = losses.mean().item()\n",
     "    model.train()\n",
     "    return output"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 14,
+   "execution_count": 10,
    "metadata": {},
    "outputs": [],
    "source": [
@@ -223,10 +271,54 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 15,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "ename": "KeyboardInterrupt",
+     "evalue": "",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[31m---------------------------------------------------------------------------\u001b[39m",
+      "\u001b[31mKeyboardInterrupt\u001b[39m                         Traceback (most recent call last)",
+      "\u001b[36mCell\u001b[39m\u001b[36m \u001b[39m\u001b[32mIn[15]\u001b[39m\u001b[32m, line 36\u001b[39m\n\u001b[32m     32\u001b[39m val_losses = []\n\u001b[32m     33\u001b[39m \n\u001b[32m     34\u001b[39m \u001b[38;5;28;01mfor\u001b[39;00m iteration \u001b[38;5;28;01min\u001b[39;00m range(max_iters):\n\u001b[32m     35\u001b[39m     \u001b[38;5;28;01mif\u001b[39;00m iteration % eval_interval == \u001b[32m0\u001b[39m \u001b[38;5;28;01mor\u001b[39;00m iteration == max_iters - \u001b[32m1\u001b[39m:\n\u001b[32m---> \u001b[39m\u001b[32m36\u001b[39m         losses = estimate_loss()\n\u001b[32m     37\u001b[39m         print(\n\u001b[32m     38\u001b[39m             f\"step {iteration}: \"\n\u001b[32m     39\u001b[39m             f\"train loss {losses[\u001b[33m'train'\u001b[39m]:.4f}, \"\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\utils\\_contextlib.py:116\u001b[39m, in \u001b[36mcontext_decorator.<locals>.decorate_context\u001b[39m\u001b[34m(*args, **kwargs)\u001b[39m\n\u001b[32m    113\u001b[39m \u001b[38;5;129m@functools\u001b[39m.wraps(func)\n\u001b[32m    114\u001b[39m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34mdecorate_context\u001b[39m(*args, **kwargs):\n\u001b[32m    115\u001b[39m     \u001b[38;5;28;01mwith\u001b[39;00m ctx_factory():\n\u001b[32m--> \u001b[39m\u001b[32m116\u001b[39m         \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mfunc\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43margs\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43mkwargs\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n",
+      "\u001b[36mCell\u001b[39m\u001b[36m \u001b[39m\u001b[32mIn[9]\u001b[39m\u001b[32m, line 14\u001b[39m, in \u001b[36mestimate_loss\u001b[39m\u001b[34m()\u001b[39m\n\u001b[32m     10\u001b[39m     \u001b[38;5;28;01mfor\u001b[39;00m split \u001b[38;5;28;01min\u001b[39;00m [\u001b[33m'train'\u001b[39m, \u001b[33m'val'\u001b[39m]:\n\u001b[32m     11\u001b[39m         losses = torch.zeros(eval_iters)\n\u001b[32m     12\u001b[39m         \u001b[38;5;28;01mfor\u001b[39;00m k \u001b[38;5;28;01min\u001b[39;00m range(eval_iters):\n\u001b[32m     13\u001b[39m             x, y = get_batch(split)\n\u001b[32m---> \u001b[39m\u001b[32m14\u001b[39m             _, loss = model(x, y)\n\u001b[32m     15\u001b[39m             losses[k] = loss.item()\n\u001b[32m     16\u001b[39m         output[split] = losses.mean().item()\n\u001b[32m     17\u001b[39m     model.train()\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\modules\\module.py:1736\u001b[39m, in \u001b[36mModule._wrapped_call_impl\u001b[39m\u001b[34m(self, *args, **kwargs)\u001b[39m\n\u001b[32m   1734\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m._compiled_call_impl(*args, **kwargs)  \u001b[38;5;66;03m# type: ignore[misc]\u001b[39;00m\n\u001b[32m   1735\u001b[39m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[32m-> \u001b[39m\u001b[32m1736\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43m_call_impl\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43margs\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43mkwargs\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\modules\\module.py:1747\u001b[39m, in \u001b[36mModule._call_impl\u001b[39m\u001b[34m(self, *args, **kwargs)\u001b[39m\n\u001b[32m   1742\u001b[39m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[32m   1743\u001b[39m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[32m   1744\u001b[39m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m._backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._forward_pre_hooks\n\u001b[32m   1745\u001b[39m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[32m   1746\u001b[39m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[32m-> \u001b[39m\u001b[32m1747\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mforward_call\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43margs\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43mkwargs\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n\u001b[32m   1749\u001b[39m result = \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[32m   1750\u001b[39m called_always_called_hooks = \u001b[38;5;28mset\u001b[39m()\n",
+      "\u001b[36mFile \u001b[39m\u001b[32md:\\Train_Your_Language_Model_Course-clean\\notebooks\\..\\transformer\\model.py:138\u001b[39m, in \u001b[36mGPTLanguageModel.forward\u001b[39m\u001b[34m(self, input_tokens, targets)\u001b[39m\n\u001b[32m    136\u001b[39m x = \u001b[38;5;28mself\u001b[39m.blocks(x)\n\u001b[32m    137\u001b[39m x = \u001b[38;5;28mself\u001b[39m.final_layer_norm(x)\n\u001b[32m--> \u001b[39m\u001b[32m138\u001b[39m logits = \u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mfinal_linear_layer\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43mx\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n\u001b[32m    140\u001b[39m \u001b[38;5;28;01mif\u001b[39;00m targets \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m:\n\u001b[32m    141\u001b[39m     loss = \u001b[38;5;28;01mNone\u001b[39;00m\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\modules\\module.py:1736\u001b[39m, in \u001b[36mModule._wrapped_call_impl\u001b[39m\u001b[34m(self, *args, **kwargs)\u001b[39m\n\u001b[32m   1734\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[38;5;28mself\u001b[39m._compiled_call_impl(*args, **kwargs)  \u001b[38;5;66;03m# type: ignore[misc]\u001b[39;00m\n\u001b[32m   1735\u001b[39m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[32m-> \u001b[39m\u001b[32m1736\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43m_call_impl\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43margs\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43mkwargs\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\modules\\module.py:1747\u001b[39m, in \u001b[36mModule._call_impl\u001b[39m\u001b[34m(self, *args, **kwargs)\u001b[39m\n\u001b[32m   1742\u001b[39m \u001b[38;5;66;03m# If we don't have any hooks, we want to skip the rest of the logic in\u001b[39;00m\n\u001b[32m   1743\u001b[39m \u001b[38;5;66;03m# this function, and just call forward.\u001b[39;00m\n\u001b[32m   1744\u001b[39m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m (\u001b[38;5;28mself\u001b[39m._backward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._forward_hooks \u001b[38;5;129;01mor\u001b[39;00m \u001b[38;5;28mself\u001b[39m._forward_pre_hooks\n\u001b[32m   1745\u001b[39m         \u001b[38;5;129;01mor\u001b[39;00m _global_backward_pre_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_backward_hooks\n\u001b[32m   1746\u001b[39m         \u001b[38;5;129;01mor\u001b[39;00m _global_forward_hooks \u001b[38;5;129;01mor\u001b[39;00m _global_forward_pre_hooks):\n\u001b[32m-> \u001b[39m\u001b[32m1747\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mforward_call\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43margs\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43m*\u001b[39;49m\u001b[30;43mkwargs\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n\u001b[32m   1749\u001b[39m result = \u001b[38;5;28;01mNone\u001b[39;00m\n\u001b[32m   1750\u001b[39m called_always_called_hooks = \u001b[38;5;28mset\u001b[39m()\n",
+      "\u001b[36mFile \u001b[39m\u001b[32mc:\\Users\\dglan\\AppData\\Local\\Programs\\Python\\Python312\\Lib\\site-packages\\torch\\nn\\modules\\linear.py:125\u001b[39m, in \u001b[36mLinear.forward\u001b[39m\u001b[34m(self, input)\u001b[39m\n\u001b[32m    124\u001b[39m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34mforward\u001b[39m(\u001b[38;5;28mself\u001b[39m, \u001b[38;5;28minput\u001b[39m: Tensor) -> Tensor:\n\u001b[32m--> \u001b[39m\u001b[32m125\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[30;43mF\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mlinear\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43minput\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mweight\u001b[39;49m\u001b[30;43m,\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mbias\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n",
+      "\u001b[31mKeyboardInterrupt\u001b[39m: "
+     ]
+    }
+   ],
    "source": [
+    "import torch\n",
+    "\n",
+    "if 'model' not in globals():\n",
+    "    from transformer.model import GPTLanguageModel\n",
+    "\n",
+    "    block_size = 512\n",
+    "    n_embd = 128\n",
+    "    n_head = 8\n",
+    "    n_layer = 1\n",
+    "    dropout = 0.2\n",
+    "    batch_size = 128\n",
+    "    vocab_size = get_vocab_size(tokenizer)\n",
+    "    device = 'cuda' if torch.cuda.is_available() else 'cpu'\n",
+    "\n",
+    "    model = GPTLanguageModel(\n",
+    "        vocab_size=vocab_size,\n",
+    "        block_size=block_size,\n",
+    "        n_embd=n_embd,\n",
+    "        n_head=n_head,\n",
+    "        n_layer=n_layer,\n",
+    "        dropout=dropout,\n",
+    "        device=device\n",
+    "    ).to(device)\n",
+    "\n",
     "max_iters = 1000\n",
     "eval_interval = 10\n",
     "learning_rate = 1e-4\n",
@@ -265,12 +357,12 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 12,
    "metadata": {},
    "outputs": [
     {
      "data": {
-      "image/png": "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",
+      "image/png": "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",
       "text/plain": [
        "<Figure size 1000x500 with 1 Axes>"
       ]
@@ -296,9 +388,23 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 13,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "ename": "KeyError",
+     "evalue": "1031",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[31m---------------------------------------------------------------------------\u001b[39m",
+      "\u001b[31mKeyError\u001b[39m                                  Traceback (most recent call last)",
+      "\u001b[36mCell\u001b[39m\u001b[36m \u001b[39m\u001b[32mIn[13]\u001b[39m\u001b[32m, line 9\u001b[39m\n\u001b[32m      5\u001b[39m model.eval()\n\u001b[32m      6\u001b[39m \u001b[38;5;28;01mwith\u001b[39;00m torch.no_grad():\n\u001b[32m      7\u001b[39m     output = model.generate(input_tokens=input_tokens, max_new_tokens=\u001b[32m50\u001b[39m)\n\u001b[32m      8\u001b[39m \n\u001b[32m----> \u001b[39m\u001b[32m9\u001b[39m print(tokenizer.decode(output[\u001b[32m0\u001b[39m].tolist()))\n",
+      "\u001b[36mFile \u001b[39m\u001b[32md:\\Train_Your_Language_Model_Course-clean\\notebooks\\..\\minbpe\\basic.py:55\u001b[39m, in \u001b[36mBasicTokenizer.decode\u001b[39m\u001b[34m(self, ids)\u001b[39m\n\u001b[32m     53\u001b[39m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34mdecode\u001b[39m(\u001b[38;5;28mself\u001b[39m, ids):\n\u001b[32m     54\u001b[39m     \u001b[38;5;66;03m# given ids (list of integers), return Python string\u001b[39;00m\n\u001b[32m---> \u001b[39m\u001b[32m55\u001b[39m     text_bytes = \u001b[30;43mb\u001b[39;49m\u001b[30;43m\"\u001b[39;49m\u001b[30;43m\"\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mjoin\u001b[39;49m\u001b[30;43m(\u001b[39;49m\u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mvocab\u001b[39;49m\u001b[30;43m[\u001b[39;49m\u001b[30;43midx\u001b[39;49m\u001b[30;43m]\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43;01mfor\u001b[39;49;00m\u001b[30;43m \u001b[39;49m\u001b[30;43midx\u001b[39;49m\u001b[30;43m \u001b[39;49m\u001b[30;43;01min\u001b[39;49;00m\u001b[30;43m \u001b[39;49m\u001b[30;43mids\u001b[39;49m\u001b[30;43m)\u001b[39;49m\n\u001b[32m     56\u001b[39m     text = text_bytes.decode(\u001b[33m\"\u001b[39m\u001b[33mutf-8\u001b[39m\u001b[33m\"\u001b[39m, errors=\u001b[33m\"\u001b[39m\u001b[33mreplace\u001b[39m\u001b[33m\"\u001b[39m)\n\u001b[32m     57\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m text\n",
+      "\u001b[36mFile \u001b[39m\u001b[32md:\\Train_Your_Language_Model_Course-clean\\notebooks\\..\\minbpe\\basic.py:55\u001b[39m, in \u001b[36m<genexpr>\u001b[39m\u001b[34m(.0)\u001b[39m\n\u001b[32m     53\u001b[39m \u001b[38;5;28;01mdef\u001b[39;00m\u001b[38;5;250m \u001b[39m\u001b[34mdecode\u001b[39m(\u001b[38;5;28mself\u001b[39m, ids):\n\u001b[32m     54\u001b[39m     \u001b[38;5;66;03m# given ids (list of integers), return Python string\u001b[39;00m\n\u001b[32m---> \u001b[39m\u001b[32m55\u001b[39m     text_bytes = \u001b[33mb\u001b[39m\u001b[33m\"\u001b[39m\u001b[33m\"\u001b[39m.join(\u001b[30;43mself\u001b[39;49m\u001b[30;43m.\u001b[39;49m\u001b[30;43mvocab\u001b[39;49m\u001b[30;43m[\u001b[39;49m\u001b[30;43midx\u001b[39;49m\u001b[30;43m]\u001b[39;49m \u001b[38;5;28;01mfor\u001b[39;00m idx \u001b[38;5;129;01min\u001b[39;00m ids)\n\u001b[32m     56\u001b[39m     text = text_bytes.decode(\u001b[33m\"\u001b[39m\u001b[33mutf-8\u001b[39m\u001b[33m\"\u001b[39m, errors=\u001b[33m\"\u001b[39m\u001b[33mreplace\u001b[39m\u001b[33m\"\u001b[39m)\n\u001b[32m     57\u001b[39m     \u001b[38;5;28;01mreturn\u001b[39;00m text\n",
+      "\u001b[31mKeyError\u001b[39m: 1031"
+     ]
+    }
+   ],
    "source": [
     "input_tokens = tokenizer.encode(\"Salam labas\")\n",
     "input_tokens = torch.tensor(\n",
@@ -314,7 +420,7 @@
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "vincent",
+   "display_name": "Python 3",
    "language": "python",
    "name": "python3"
   },
@@ -328,7 +434,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.11.10"
+   "version": "3.12.10"
   }
  },
  "nbformat": 4,
diff --git a/notebooks/5_FineTuningDataset.ipynb b/notebooks/5_FineTuningDataset.ipynb
index d795eb1..55e5487 100644
--- a/notebooks/5_FineTuningDataset.ipynb
+++ b/notebooks/5_FineTuningDataset.ipynb
@@ -9,12 +9,28 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 2,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "28015"
+      ]
+     },
+     "execution_count": 2,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
-    "file_path = \"../data/private/fine_tuning.txt\"\n",
-    "with open(file_path, 'r', encoding='utf-8') as f:\n",
+    "from pathlib import Path\n",
+    "\n",
+    "file_path = Path(\"../data/private/Direct Messages - Buncha retards [1411801796439183370].txt\")\n",
+    "if not file_path.exists():\n",
+    "    raise FileNotFoundError(f\"Could not find expected data file: {file_path}\")\n",
+    "\n",
+    "with file_path.open('r', encoding='utf-8') as f:\n",
     "    lines = f.readlines()\n",
     "\n",
     "len(lines)"
@@ -29,14 +45,22 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 4,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Lines removed: 862\n"
+     ]
+    }
+   ],
    "source": [
     "import re\n",
     "\n",
     "encryption_message = \"Messages and calls are end-to-end encrypted. No one outside of this chat, not even WhatsApp, can read or listen to them. Tap to learn more.\"\n",
-    "media_pattern = \"<Media omitted>\"\n",
+    "media_pattern = \"{Embed}\"\n",
     "email_pattern = r'[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\\.[A-Z|a-z]{2,}'\n",
     "url_pattern = r'http[s]?://(?:[a-zA-Z]|[0-9]|[$-_@.&+]|[!*\\(\\),]|(?:%[0-9a-fA-F][0-9a-fA-F]))+'\n",
     "edited_message = \"<This message was edited>\"\n",
@@ -110,9 +134,20 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 5,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "13568"
+      ]
+     },
+     "execution_count": 5,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "grouped_messages = []\n",
     "\n",
@@ -147,9 +182,20 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 6,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "13568"
+      ]
+     },
+     "execution_count": 6,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
    "source": [
     "# Define special tokens\n",
     "start_of_text_token = \"<|startoftext|>\"\n",
@@ -176,21 +222,38 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 6,
+   "execution_count": null,
    "metadata": {},
-   "outputs": [],
+   "outputs": [
+    {
+     "ename": "FileNotFoundError",
+     "evalue": "[Errno 2] No such file or directory: '../output/fine_tuning/data/fine_tuning.json'",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[31m---------------------------------------------------------------------------\u001b[39m",
+      "\u001b[31mFileNotFoundError\u001b[39m                         Traceback (most recent call last)",
+      "\u001b[36mCell\u001b[39m\u001b[36m \u001b[39m\u001b[32mIn[7]\u001b[39m\u001b[32m, line 4\u001b[39m\n\u001b[32m      1\u001b[39m \u001b[38;5;28;01mimport\u001b[39;00m json\n\u001b[32m      2\u001b[39m \n\u001b[32m      3\u001b[39m save_path = \u001b[33m\"../output/fine_tuning/data/fine_tuning.json\"\u001b[39m\n\u001b[32m----> \u001b[39m\u001b[32m4\u001b[39m \u001b[38;5;28;01mwith\u001b[39;00m open(save_path, \u001b[33m'w'\u001b[39m, encoding=\u001b[33m'utf-8'\u001b[39m) \u001b[38;5;28;01mas\u001b[39;00m f:\n\u001b[32m      5\u001b[39m     json.dump(fine_tuning_data, f, ensure_ascii=\u001b[38;5;28;01mFalse\u001b[39;00m, indent=\u001b[32m4\u001b[39m)\n",
+      "\u001b[31mFileNotFoundError\u001b[39m: [Errno 2] No such file or directory: '../output/fine_tuning/data/fine_tuning.json'"
+     ]
+    }
+   ],
    "source": [
     "import json\n",
+    "from pathlib import Path\n",
+    "\n",
+    "save_path = Path(\"../output/fine_tuning/data/fine_tuning.json\")\n",
+    "save_path.parent.mkdir(parents=True, exist_ok=True)\n",
+    "\n",
+    "with save_path.open('w', encoding='utf-8') as f:\n",
+    "    json.dump(fine_tuning_data, f, ensure_ascii=False, indent=4)\n",
     "\n",
-    "save_path = \"../output/fine_tuning/data/fine_tuning.json\"\n",
-    "with open(save_path, 'w', encoding='utf-8') as f:\n",
-    "    json.dump(fine_tuning_data, f, ensure_ascii=False, indent=4)"
+    "save_path"
    ]
   }
  ],
  "metadata": {
   "kernelspec": {
-   "display_name": "vincent",
+   "display_name": "Python 3",
    "language": "python",
    "name": "python3"
   },
@@ -204,7 +267,7 @@
    "name": "python",
    "nbconvert_exporter": "python",
    "pygments_lexer": "ipython3",
-   "version": "3.11.10"
+   "version": "3.12.10"
   }
  },
  "nbformat": 4,
diff --git a/requirements.txt b/requirements.txt
index 55dd2f0..82be6ab 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -3,3 +3,4 @@ numpy==2.2.4
 tiktoken==0.9.0
 torch==2.5.0
 tqdm==4.66.5
+pandas==2.2.3