{"slug": "your-gmail-can-write-code", "title": "Your 'Gmail' can write code.", "summary": "Google's Gemini AI in Gmail drafted a fully functional Python script to fine-tune Microsoft's DeBERTa-v3-large model for 3-label sentiment classification, including data loading, training loops, and evaluation, demonstrating the AI's capability to generate complex code within email drafts.", "body_md": "Gemini in Gmail drafted Python code to fine-tune a DeBERTa model for sentiment classification, complete with data loading, training loops, and evaluation.\n\nGmail drafter is powered by a language model, and it can do almost anything your chat assistant can. For example here I ask it to write python code to train a sentiment classification model.\n\n```\nwrite code for train.py to fine-tune deberta v3 large for sentiment classification into 3 labels: positive, neutral and negative data.csv as source headers: text,positive,neutral,negative (values for labels are 0 or 1) use tqdm and wandb for logging per step\n```\n\nIt drafted the following email with a fully functional python code:\n\n---\n\nHey [Name],\n\nHere is the code for train.py to fine-tune DeBERTa-v3-large on your sentiment data. It reads from data.csv, sets up a 3-label classification, and handles logging via tqdm and Weights & Biases per step.\n\n``` python\nimport pandas as pd\nimport torch\nfrom torch.utils.data import Dataset, DataLoader\nfrom transformers import AutoTokenizer, AutoModelForSequenceClassification, AdamW, get_scheduler\nfrom sklearn.model_selection import train_test_split\nimport wandb\nfrom tqdm import auto as tqdm_auto\nimport numpy as np\n\n# Initialize wandb\nwandb.init(project=\"deberta-sentiment-classification\")\n\n# Load data\ndf = pd.read_csv('data.csv')\n\n# Map one-hot labels to a single target index\n# Assumes headers are text, positive, neutral, negative\ndef get_label(row):\n if row['positive'] == 1: return 0\n if row['neutral'] == 1: return 1\n if row['negative'] == 1: return 2\n return -1\n\ndf['label'] = df.apply(get_label, axis=1)\ndf = df[df['label'] != -1].reset_index(drop=True)\n\ntrain_df, val_df = train_test_split(df, test_size=0.1, random_state=42)\n\nclass SentimentDataset(Dataset):\n def __init__(self, texts, labels, tokenizer, max_len=128):\n self.texts = texts\n self.labels = labels\n self.tokenizer = tokenizer\n self.max_len = max_len\n \n def __len__(self):\n return len(self.texts)\n \n def __getitem__(self, idx):\n text = str(self.texts[idx])\n label = self.labels[idx]\n encoding = self.tokenizer(\n text,\n add_special_tokens=True,\n max_length=self.max_len,\n padding='max_length',\n truncation=True,\n return_tensors='pt'\n )\n return {\n 'input_ids': encoding['input_ids'].flatten(),\n 'attention_mask': encoding['attention_mask'].flatten(),\n 'labels': torch.tensor(label, dtype=torch.long)\n }\n\n# Model and Tokenizer setup\nMODEL_NAME = \"microsoft/deberta-v3-large\"\ntokenizer = AutoTokenizer.pretrained(MODEL_NAME)\nmodel = AutoModelForSequenceClassification.pretrained(MODEL_NAME, num_labels=3)\n\ntrain_dataset = SentimentDataset(train_df['text'].values, train_df['label'].values, tokenizer)\nval_dataset = SentimentDataset(val_df['text'].values, val_df['label'].values, tokenizer)\n\ntrain_loader = DataLoader(train_dataset, batch_size=8, shuffle=True)\nval_loader = DataLoader(val_dataset, batch_size=16)\n\ndevice = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\nmodel.to(device)\n\noptimizer = AdamW(model.parameters(), lr=2e-5)\nnum_epochs = 3\nnum_training_steps = num_epochs * len(train_loader)\nlr_scheduler = get_scheduler(\"linear\", optimizer=optimizer, num_warmup_steps=0, num_training_steps=num_training_steps)\n\nwandb.config.update({\n \"model\": MODEL_NAME,\n \"epochs\": num_epochs,\n \"batch_size\": 8,\n \"lr\": 2e-5\n})\n\nglobal_step = 0\nfor epoch in range(num_epochs):\n model.train()\n train_bar = tqdm_auto.tqdm(train_loader, desc=f\"Epoch {epoch+1}/{num_epochs} [Train]\")\n \n for batch in train_bar:\n optimizer.zero_grad()\n input_ids = batch['input_ids'].to(device)\n attention_mask = batch['attention_mask'].to(device)\n labels = batch['labels'].to(device)\n \n outputs = model(input_ids=input_ids, attention_mask=attention_mask, labels=labels)\n loss = outputs.loss\n loss.backward()\n \n optimizer.step()\n lr_scheduler.step()\n \n global_step += 1\n wandb.log({\"train_loss\": loss.item(), \"learning_rate\": lr_scheduler.get_last_lr()[0]}, step=global_step)\n train_bar.set_postfix(loss=loss.item())\n\n # Validation loop\n model.eval()\n val_loss = 0\n correct = 0\n total = 0\n val_bar = tqdm_auto.tqdm(val_loader, desc=f\"Epoch {epoch+1}/{num_epochs} [Val]\")\n \n with torch.no_grad():\n for batch in val_bar:\n input_ids = batch['input_ids'].to(device)\n attention_mask = batch['attention_mask'].to(device)\n labels = batch['labels'].to(device)\n \n outputs = model(input_ids=input_ids, attention_mask=attention_mask, labels=labels)\n val_loss += outputs.loss.item()\n \n preds = torch.argmax(outputs.logits, dim=1)\n correct += (preds == labels).sum().item()\n total += labels.size(0)\n\n avg_val_loss = val_loss / len(val_loader)\n accuracy = correct / total\n wandb.log({\"val_loss\": avg_val_loss, \"val_accuracy\": accuracy}, step=global_step)\n print(f\"Epoch {epoch+1} - Val Loss: {avg_val_loss:.4f}, Val Acc: {accuracy:.4f}\")\n\nmodel.save_pretrained(\"./sentiment_model\")\ntokenizer.save_pretrained(\"./sentiment_model\")\nwandb.finish()\n```\n\nLet me know if you need to adjust the batch size or max length constraints.\n\n---\n\n```\n\n \n \n \n \n \n \n \n\n \n \n \n \n \n \n DEJAN\n \n\n```\n\nPS: Cute, but I think I'll keep the one we have for now though ;-)", "url": "https://wpnews.pro/news/your-gmail-can-write-code", "canonical_source": "https://dejan.ai/blog/gmail-code/", "published_at": "2026-08-31 05:35:48+00:00", "updated_at": "2026-08-31 05:53:19.220473+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-products", "ai-tools"], "entities": ["Google", "Gemini", "Gmail", "Microsoft", "DeBERTa-v3-large"], "alternates": {"html": "https://wpnews.pro/news/your-gmail-can-write-code", "markdown": "https://wpnews.pro/news/your-gmail-can-write-code.md", "text": "https://wpnews.pro/news/your-gmail-can-write-code.txt", "jsonld": "https://wpnews.pro/news/your-gmail-can-write-code.jsonld"}}