This model is a fine-tuned version of GPT‑2, trained for tweet sentiment classification (negative, neutral, positive) using the MTEB Tweet Sentiment Extraction dataset.

Model Overview

Base Model: GPT‑2 (decoder-only transformer)
Dataset: MTEB Tweet Sentiment Extraction. A dataset containing ~27481 training tweets and 3534 test tweets, labeled as negative (0), neutral (1), or positive (2)
Label Distribution (test set):
- Negative (0): 1001 tweets
- Neutral (1): 1430 tweets
- Positive (2): 1103 tweets

Evaluation Metric

These results reflect robust performance for a GPT‑2-based classifier on this tweet sentiment task.

Metric	Value
Loss	0.6867
Accuracy	0.7937
F1 Score	0.7934
Epochs	5

Training Details

Epochs: 5
Evaluation Metrics: Accuracy, F1 Score, Loss
Tokenizer: GPT-2 tokenizer (GPT2Tokenizer) loaded from "gpt2" and modified to use eos_token as the pad_token

Dataset Details

Name: MTEB Tweet Sentiment Extraction
Source: MTEB - Massive Text Embedding Benchmark (dataset link)
Split Sizes:
- Train: 27481 samples
- Test: 3534 samples
Text Length:
- Average: ~68 characters
- Range: from 4 to ~142 characters
Classes: 3 (0 = negative, 1 = neutral, 2 = positive)

License

This model is licensed under the Apache 2.0 License.

Author

Fine-tuned and published by @herrerovir using Hugging Face Transformers.

Usage Example

Loading and using the model via Hugging Face transformers:

python

from transformers import pipeline

classifier = pipeline("text-classification", model="herrerovir/gpt2-sentiment-tweets")

output = classifier("I can't believe how amazing this is!")
print(output)  # e.g., [{'label': 'positive', 'score': 0.94}]

herrerovir/gpt2-tweet-sentiment-model

GPT‑2 Tweet Sentiment Classifier