falcon-sentiment

This Falcon 7B was fined-tuned on 672k tweets from twitter/X for text classification task. The classification accuracy obtained is 98%.
The number of labels is 3: {0: Negative, 1: Neutral, 2: Positive}
Warning: You need enough GPU to use Falcon.

This is an example to use it, it worked on 8 GB VRAM Nvidia RTX-4060

from transformers import AutoTokenizer
from transformers import pipeline
from transformers import AutoModelForSequenceClassification
import torch

checkpoint = 'kumo24/falcon-sentiment'
tokenizer=AutoTokenizer.from_pretrained(checkpoint)
id2label = {0: "negative", 1: "neutral", 2: "positive"}
label2id = {"negative": 0, "neutral": 1, "positive": 2}
    

if tokenizer.pad_token is None:
    tokenizer.add_special_tokens({'pad_token': '[PAD]'})

model = AutoModelForSequenceClassification.from_pretrained(checkpoint, 
                                                       num_labels=3,
                                                       id2label=id2label, 
                                                       label2id=label2id,
                                                       device_map='auto')

sentiment_task = pipeline("sentiment-analysis", 
                          model=model, 
                          tokenizer=tokenizer)

print(sentiment_task("Michigan Wolverines are Champions, Go Blue!"))