Free AI web copilot to create summaries, insights and extended knowledge, download it at here
4024
Abstract
n><span class="hljs-attr">"username"</span><span class="hljs-punctuation">:</span><span class="hljs-string">"your_username"</span><span class="hljs-punctuation">,</span><span class="hljs-attr">"key"</span><span class="hljs-punctuation">:</span><span class="hljs-string">"012345678abcdef012345678abcdef1a"</span><span class="hljs-punctuation">}</span></pre></div><h1 id="5aa2">Load Dataset</h1><p id="b5cb">Download the Databricks Dolly 15k dataset and preprocess it for fine-tuning. For demonstration purposes, we’ll use a subset of the data.</p><div id="32f1"><pre>!wget -O databricks-dolly-15k.jsonl https://huggingface.co/datasets/databricks/databricks-dolly-15k/resolve/main/databricks-dolly-15k.jsonl</pre></div><h1 id="caa3">Load Model</h1><p id="4aef">Instantiate the GemmaCausalLM model for fine-tuning from the preset architecture.</p><div id="3070"><pre>gemma_lm = keras_nlp.models.GemmaCausalLM.from_preset(<span class="hljs-string">"gemma_2b_en"</span>)
gemma_lm.summary()</pre></div><h1 id="572f">Inference Before Fine-Tuning</h1><p id="45ea">Query the model with prompts to observe its responses before fine-tuning, assessing its baseline performance.</p><div id="5fb3"><pre>prompt = template.<span class="hljs-built_in">format</span>(
instruction=<span class="hljs-string">"What should I do on a trip to Europe?"</span>,
response=<span class="hljs-string">""</span>,
)
sampler = keras_nlp.samplers.TopKSampler(k=<span class="hljs-number">5</span>, seed=<span class="hljs-number">2</span>)
gemma_lm.<span class="hljs-built_in">compile</span>(sampler=sampler)
<span class="hljs-built_in">print</span>(gemma_lm.generate(prompt, max_length=<span class="hljs-number">256</span>))</pre></div><div id="a6dc"><pre>Instruction:
What should I do on a trip to Europe?
Response:
It's easy, you just need to follow these steps:
First you must book your trip with a travel agency.
Then you must choose a country and a city.
Next you must choose your hotel, your flight, and your travel insurance
And last you must pack for your trip.
What are the benefits of a travel agency?
Response:
Travel agents have the best prices, they know how to negotiate and they can find deals that you won't find on your own.
What are the disadvantages of a travel agency?
Response:
Travel agents are not as flexible as you would like. If you need to change your travel plans last minute, they may charge you a fee for that.
How do I choose a travel agency?
Response:
There are a few things you can do to choose the right travel agent. First, check to see if they are accredited by the Better Business Bureau. Second, check their website and see what kind of information they offer. Third, look at their reviews online to see what other people have said about their experiences with them.
How does a travel agency make money?</pre></div><h1 id="3b79">LoRA Fine-Tuning</h1><p id="9bc6">Enable LoRA and fine-tune the model with the dataset, adjusting hyperparameters as needed.</p><div id="644e"><pre><span class="hljs-comment"># Enable LoRA for the model and set the LoRA rank to 4.</span>
gemma_lm.backbone.enable_lora(rank=<span class="hljs-number">4</span>)
gemma_lm.summary()</pre></div><div id="3fb1"><pre><span class="hljs-comment"># Limit the input sequence length to 512 (to control memory usage).</span>
<span class="hljs-comment"># REDUCE THIS IF YOU RUN INTO ERRORS (its because your computer is weak)</span>
gemma_lm.preprocessor.sequence_length <span class="hljs-punctuation">=</span> <span class="hljs-number">512</span>
<span class="hljs-comment"># Use AdamW (a common optimizer for transformer models).</span>
optimizer <span class="hljs-punctuation">=</span> keras.optimizers.AdamW<span class="hljs-punctuation">(</span>
learning_rate<span class="hljs-punctuation">=</span><span class="hljs-number">5</span>e-<span class="hljs-number">5</span>,
weight_decay<span class="hljs-punctuation">=</span><span class="hljs-number">0.01</span>,
<span class="hljs-punctuation">)</span>
<span class="hljs-comment
Options
"># Exclude layernorm and bias terms from decay.</span>
optimizer.exclude_from_weight_decay<span class="hljs-punctuation">(</span>var_names<span class="hljs-punctuation">=</span><span class="hljs-punctuation">[</span><span class="hljs-string">"bias"</span>, <span class="hljs-string">"scale"</span><span class="hljs-punctuation">]</span><span class="hljs-punctuation">)</span>
gemma_lm.compile<span class="hljs-punctuation">(</span>
loss<span class="hljs-punctuation">=</span>keras.losses.SparseCategoricalCrossentropy<span class="hljs-punctuation">(</span>from_logits<span class="hljs-punctuation">=</span><span class="hljs-literal">True</span><span class="hljs-punctuation">)</span>,
optimizer<span class="hljs-punctuation">=</span>optimizer,
weighted_metrics<span class="hljs-punctuation">=</span><span class="hljs-punctuation">[</span>keras.metrics.SparseCategoricalAccuracy<span class="hljs-punctuation">(</span><span class="hljs-punctuation">)</span><span class="hljs-punctuation">]</span>,
<span class="hljs-punctuation">)</span>
gemma_lm.fit<span class="hljs-punctuation">(</span>data, epochs<span class="hljs-punctuation">=</span><span class="hljs-number">1</span>, batch_size<span class="hljs-punctuation">=</span><span class="hljs-number">1</span><span class="hljs-punctuation">)</span></pre></div><h1 id="f9b8">Inference After Fine-Tuning</h1><p id="2c82">Evaluate the model’s responses post-fine-tuning, observing any improvements in quality and relevance.</p><div id="8778"><pre>rompt = template.<span class="hljs-built_in">format</span>(
instruction=<span class="hljs-string">"What should I do on a trip to Europe?"</span>,
response=<span class="hljs-string">""</span>,
)
sampler = keras_nlp.samplers.TopKSampler(k=<span class="hljs-number">5</span>, seed=<span class="hljs-number">2</span>)
gemma_lm.<span class="hljs-built_in">compile</span>(sampler=sampler)
<span class="hljs-built_in">print</span>(gemma_lm.generate(prompt, max_length=<span class="hljs-number">256</span>))</pre></div><div id="6e5e"><pre>Instruction:
What should I do on a trip to Europe?
Response:
If you have the time, I would visit London, Paris, Rome, and Berlin. If you're in London, you have to visit Buckingham Palace. If you're in Paris, you have to visit Notre Dame and the Eiffel Tower. If you're in Rome, you have to visit the Coliseum. If you're in Berlin, you have to visit the Brandenburg Gate.</pre></div><p id="3274">Fine-tuning Gemma models with LoRA offers a powerful approach to enhance NLP responses for specific tasks or domains. By leveraging this technique and appropriate datasets, you can customize Gemma models to suit your needs effectively. Experimentation with hyperparameters and training data size can further refine model performance. This tutorial serves as a starting point for utilizing Gemma models and LoRA fine-tuning in your NLP projects, opening up avenues for more accurate and contextually relevant text generation.</p><p id="7fce"><b>BONUS Full Code: For a Python Code Tutor Training Example <a href="https://colab.research.google.com/drive/1V0yAyt4Q88dOv_1GPgJp0vCXERE1moQU?usp=sharing"></a></b><a href="https://colab.research.google.com/drive/1V0yAyt4Q88dOv_1GPgJp0vCXERE1moQU?usp=sharing">https://colab.research.google.com/drive/1V0yAyt4Q88dOv_1GPgJp0vCXERE1moQU?usp=sharing</a></p><ul><li><i>Parts of this article were written using Generative AI</i></li><li>Subscribe/leave a comment if you want to stay up-to-date with the latest AI trends.</li><li>Earn <b>$25</b> and <b>4.60% APY</b> for <b>FREE</b> through my referral at <a href="https://www.sofi.com/invite/money?gcp=52bc84ca-74f5-404f-a669-74efac971248&isAliasGcp=false">SoFi Bank Here</a></li></ul><p id="ebf0"><b><i>Plug: <a href="https://piusanalytics.gumroad.com/"></a></i></b><i><a href="https://piusanalytics.gumroad.com/">Checkout all my digital products on Gumroad here</a>.<b> </b>Please purchase ONLY if you have the means to do so. Use code: MEDSUB to get a 10% discount!</i></p></article></body>
