| --- |
| library_name: peft |
| license: mit |
| base_model: FacebookAI/roberta-base |
| tags: |
| - generated_from_trainer |
| metrics: |
| - accuracy |
| - f1 (macro) |
| - top3 accuracy |
| model-index: |
| - name: roberta-base-with-tweet-eval-emoji-full |
| results: [] |
| --- |
| |
| # Model description |
|
|
| This model is a fine-tuned version of [roberta-base](https://huggingface.co/roberta-base) on the tweet_eval/emoji dataset. |
| It achieves the following results on the evaluation set: |
| - Loss: 1.9516 |
| - Accuracy: 0.4252 |
| - F1: 0.3314 |
| - Top3 Accuracy: 0.6504 |
| |
| ## Example of classification |
| |
| ```python |
| import torch |
| from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline |
| from peft import PeftModel |
| |
| # Specify the same model ID pushed or trained locally |
| MODEL_ID = "roberta-base-tweet-emoji-lora" |
|
|
| # Load tokenizer + model |
| device = 0 if torch.cuda.is_available() else -1 |
| tok = AutoTokenizer.from_pretrained(MODEL_ID) |
| base_model = AutoModelForSequenceClassification.from_pretrained( |
| "FacebookAI/roberta-base", |
| num_labels=20, |
| ignore_mismatched_sizes=True, |
| ) |
| model = PeftModel.from_pretrained(base_model, MODEL_ID).eval() |
| |
| pipe = pipeline( |
| task="text-classification", |
| model=model, |
| tokenizer=tok, |
| return_all_scores=True, |
| function_to_apply="softmax", |
| device=device, |
| ) |
| |
| # Map label IDs to emojis |
| id2label = { |
| 0: "❤", |
| 1: "😍", |
| 2: "😂", |
| 3: "💕", |
| 4: "🔥", |
| 5: "😊", |
| 6: "😎", |
| 7: "✨", |
| 8: "💙", |
| 9: "😘", |
| 10: "📷", |
| 11: "🇺🇸", |
| 12: "☀", |
| 13: "💜", |
| 14: "😉", |
| 15: "💯", |
| 16: "😁", |
| 17: "🎄", |
| 18: "📸", |
| 19: "😜", |
| } |
| |
|
|
| def predict_emojis(text, top_k=2): |
| """ |
| Predict top k emojis for the given text. |
| |
| Args: |
| text (str): Input string. |
| k (int): Number of top emojis to return. |
| |
| Returns: |
| str: Space-separated top k emojis. |
| """ |
| probs = pipe(text)[0] |
| top = sorted(probs, key=lambda x: x["score"], reverse=True)[:top_k] |
| return " ".join(id2label[int(d["label"].split("_")[-1])] for d in top) |
| |
| print(predict_emojis("Sunny days!")) |
| ``` |
| |
| Output: |
| |
| ``` |
| 😎 ☀ |
| ``` |
| |
| |
| ### Training hyperparameters |
| |
| The following hyperparameters were used during training: |
| - learning_rate: 0.0005 |
| - train_batch_size: 128 |
| - eval_batch_size: 256 |
| - seed: 42 |
| - optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments |
| - lr_scheduler_type: linear |
| - lr_scheduler_warmup_ratio: 0.05 |
| - num_epochs: 4 |
| - mixed_precision_training: Native AMP |
| - label_smoothing_factor: 0.1 |
|
|
| ### Training results |
|
|
| | Training Loss | Epoch | Step | Validation Loss | Accuracy | F1 (Macro) | Top3 Accuracy | |
| |:-------------:|:-----:|:----:|:---------------:|:--------:|:------:|:-------------:| |
| | 2.1291 | 1.0 | 352 | 2.4306 | 0.219 | 0.2126 | 0.405 | |
| | 2.1083 | 2.0 | 704 | 2.3812 | 0.2356 | 0.2411 | 0.43 | |
| | 2.0068 | 3.0 | 1056 | 2.3611 | 0.2456 | 0.2492 | 0.4442 | |
| | 1.9326 | 4.0 | 1408 | 2.3694 | 0.2524 | 0.2536 | 0.4472 | |
|
|
|
|
| ### Framework versions |
|
|
| - PEFT 0.14.0 |
| - Transformers 4.51.3 |
| - Pytorch 2.6.0+cu124 |
| - Datasets 3.6.0 |
| - Tokenizers 0.21.1 |
