diff --git a/Gemma/Gemma_2_for_Japan_using_Transformers_and_PyTorch.ipynb b/Gemma/Gemma_2_for_Japan_using_Transformers_and_PyTorch.ipynb
new file mode 100644
index 0000000..5a04412
--- /dev/null
+++ b/Gemma/Gemma_2_for_Japan_using_Transformers_and_PyTorch.ipynb
@@ -0,0 +1,1991 @@
+{
+ "cells": [
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "id": "b-FX_tt-s8sU"
+ },
+ "source": [
+ "##### Copyright 2024 Google LLC."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": null,
+ "metadata": {
+ "cellView": "form",
+ "id": "JA485iBMs7ub"
+ },
+ "outputs": [],
+ "source": [
+ "# @title Licensed under the Apache License, Version 2.0 (the \"License\");\n",
+ "# you may not use this file except in compliance with the License.\n",
+ "# You may obtain a copy of the License at\n",
+ "#\n",
+ "# https://www.apache.org/licenses/LICENSE-2.0\n",
+ "#\n",
+ "# Unless required by applicable law or agreed to in writing, software\n",
+ "# distributed under the License is distributed on an \"AS IS\" BASIS,\n",
+ "# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.\n",
+ "# See the License for the specific language governing permissions and\n",
+ "# limitations under the License."
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "id": "igbYjdwhvioG"
+ },
+ "source": [
+ "This notebook provides a practical guide to working with Gemma 2 for Japan, the latest variant of Google's open language models.\n",
+ "\n",
+ "The model itself is available on both Kaggle and Hugging Face.\n",
+ "\n",
+ "As the model is currently only available in a 2B size, no special hardware is required.\n",
+ "\n",
+ ""
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "id": "y5ZzHdAb9sNW"
+ },
+ "source": [
+ "### Downloading prerequisites\n",
+ "The **KaggleHub** library will be used to retrieve the model from Kaggle, and the **Transformers** framework from Hugging Face will be used with **PyTorch** for inference.\n",
+ "\n",
+ "Note: the `%%capture` magic keyword suppresses output of a given cell."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 1,
+ "metadata": {
+ "id": "YYPGnh709SJR"
+ },
+ "outputs": [],
+ "source": [
+ "%%capture\n",
+ "!pip install kagglehub --upgrade\n",
+ "!pip install transformers --upgrade\n",
+ "!pip install torch --upgrade"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "id": "mEjYLz9l9eeg"
+ },
+ "source": [
+ "### Authenticating with Kaggle\n",
+ "\n",
+ "Next, Kaggle will need to be authenticated with.\n",
+ "\n",
+ "Note: the call to `login()` can be skipped if running in a Kaggle notebook."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 2,
+ "metadata": {
+ "id": "8aucYnW5_f1S"
+ },
+ "outputs": [
+ {
+ "data": {
+ "application/vnd.jupyter.widget-view+json": {
+ "model_id": "b2f1c08705234b9fad50a04c03229ade",
+ "version_major": 2,
+ "version_minor": 0
+ },
+ "text/plain": [
+ "VBox(children=(HTML(value=' str:\n",
+ " return f\"user {instruction}model\"\n",
+ "\n",
+ "def unwrap_gemma_response(query: str, response: str) -> str:\n",
+ " end_sequence = ''\n",
+ "\n",
+ " start_idx = 0\n",
+ " query_idx = response.find(query)\n",
+ "\n",
+ " if query_idx >= 0:\n",
+ " start_idx = query_idx + len(query)\n",
+ "\n",
+ " trim = response[start_idx:]\n",
+ "\n",
+ " end_idx = len(trim) - 1\n",
+ " endseq_idx = trim.find(end_sequence)\n",
+ "\n",
+ " if endseq_idx >= 0:\n",
+ " end_idx = endseq_idx\n",
+ "\n",
+ " return trim[:end_idx].strip()"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "id": "a7rqMiZ4vlnm"
+ },
+ "source": [
+ "First, you will send a prompt in Japanese to the model, asking it to write us a poem in Japanese."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 10,
+ "metadata": {
+ "id": "s2CFT7fOFaMy"
+ },
+ "outputs": [
+ {
+ "name": "stderr",
+ "output_type": "stream",
+ "text": [
+ "The 'max_batch_size' argument of HybridCache is deprecated and will be removed in v4.46. Use the more precisely named 'batch_size' argument instead.\n",
+ "Starting from v4.46, the `logits` model output will have the same type as the model (except at train time, where it will always be FP32)\n"
+ ]
+ },
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "機械学習の波、広がる未来の光、\n",
+ "データの海、複雑な知識を導く。\n",
+ "複雑なパターン、隠された法則を見つける、\n",
+ "予測と改善、未来を形作る。\n",
+ "\n",
+ "ニューラルネットワーク、複雑な脳の像、\n",
+ "学習と進化、無限の可能性を秘める。\n",
+ "教師あり学習、教師なし学習、\n",
+ "データの力、人間の知恵を融合する。\n",
+ "\n",
+ "機械学習の力、無限の可能性を秘める、\n",
+ "複雑な問題を解き明かす、未来を拓く。\n",
+ "AIの進化、人類の未来を左右する、\n",
+ "新たな時代を迎えよう。\n"
+ ]
+ }
+ ],
+ "source": [
+ "# Our prompt will be, \"Write me a poem about machine learning.\" in Japanese.\n",
+ "input_text = \"マシーンラーニングについての詩を書いてください。\"\n",
+ "\n",
+ "input_formatted = format_gemma_instruction(input_text)\n",
+ "\n",
+ "input_ids = tokenizer(input_formatted, return_tensors=\"pt\").to(device)\n",
+ "\n",
+ "outputs = model.generate(**input_ids, max_new_tokens=1024)\n",
+ "formatted = unwrap_gemma_response(input_formatted, tokenizer.decode(outputs[0]))\n",
+ "\n",
+ "print(formatted)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "id": "8H1FS_A7vuDI"
+ },
+ "source": [
+ "Next, you can feed that poem back into the model, asking it to translate it into English."
+ ]
+ },
+ {
+ "cell_type": "code",
+ "execution_count": 16,
+ "metadata": {
+ "id": "DDK4sYMaBSb8"
+ },
+ "outputs": [
+ {
+ "name": "stdout",
+ "output_type": "stream",
+ "text": [
+ "Here's the translation of the Japanese poem:\n",
+ "\n",
+ "**The wave of machine learning, spreading the light of the future,\n",
+ "A vast ocean of data, guiding complex knowledge.\n",
+ "Finding complex patterns, uncovering hidden laws,\n",
+ "Prediction and improvement, shaping the future.\n",
+ "\n",
+ "A neural network, a complex image of the brain,\n",
+ "Learning and evolution, holding infinite possibilities.\n",
+ "Supervised and unsupervised learning,\n",
+ "The power of data, fusing human wisdom.\n",
+ "\n",
+ "The power of machine learning, holding infinite possibilities,\n",
+ "Unraveling complex problems, opening up the future.\n",
+ "The evolution of AI, shaping the future of humanity,\n",
+ "We are ushering in a new era.**\n",
+ "\n",
+ "\n",
+ "\n",
+ "**Explanation of the Poem's Themes:**\n",
+ "\n",
+ "* **Machine Learning's Impact:** The poem highlights the growing influence of machine learning, emphasizing its ability to shape the future.\n",
+ "* **Data as the Foundation:** The poem emphasizes the role of data as the foundation for machine learning, suggesting that it is the key to unlocking its potential.\n",
+ "* **Complexity and Pattern Recognition:** The poem focuses on the ability of machine learning to identify complex patterns and uncover hidden laws within data.\n",
+ "* **Prediction and Improvement:** The poem highlights the role of prediction and improvement in shaping the future, suggesting that machine learning can be used to make better decisions and create a better world.\n",
+ "* **AI's Role in the Future:** The poem suggests that AI, powered by machine learning, will play a crucial role in shaping the future of humanity.\n",
+ "\n",
+ "\n",
+ "\n",
+ "Let me know if you'd like any further clarification on the poem's meaning or symbolism.\n"
+ ]
+ }
+ ],
+ "source": [
+ "translation_input_text = \"Translate the following poem from Japanese to English. \\n\\n\" + formatted\n",
+ "translation_input_formatted = format_gemma_instruction(translation_input_text)\n",
+ "\n",
+ "translation_input_ids = tokenizer(translation_input_formatted, return_tensors=\"pt\").to(device)\n",
+ "\n",
+ "translation_output = model.generate(**translation_input_ids, max_new_tokens=2048)\n",
+ "translation_formatted = unwrap_gemma_response(translation_input_formatted, tokenizer.decode(translation_output[0]))\n",
+ "\n",
+ "print(translation_formatted)"
+ ]
+ },
+ {
+ "cell_type": "markdown",
+ "metadata": {
+ "id": "r2uMH5c4v5e3"
+ },
+ "source": [
+ "And we're done!"
+ ]
+ }
+ ],
+ "metadata": {
+ "accelerator": "GPU",
+ "colab": {
+ "name": "Gemma_2_for_Japan_using_Transformers_and_PyTorch.ipynb",
+ "toc_visible": true
+ },
+ "kernelspec": {
+ "display_name": "Python 3",
+ "name": "python3"
+ }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 0
+}
![](\"https://www.tensorflow.org/images/colab_logo_32px.png\"){kind=logo}
Run in Google Colab\n",
+ "