Last active
January 14, 2024 04:30
-
-
Save thistleknot/83f17cf8fe9b917c6edc0923fa04d71a to your computer and use it in GitHub Desktop.
Gpt-neo Classify
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
from transformers import GPT2Tokenizer, GPTNeoForCausalLM | |
import torch | |
import torch.nn.functional as F | |
# Load the GPT-Neo 1.3B model and tokenizer | |
tokenizer = GPT2Tokenizer.from_pretrained("EleutherAI/gpt-neo-1.3B") | |
model = GPTNeoForCausalLM.from_pretrained("EleutherAI/gpt-neo-1.3B") | |
# Your question and prompt | |
question = "Is a bird a mammal?" | |
prompt = f""" | |
System: | |
Your role is to answer with a single character, Y for Yes, N for No, and ? for I don't know. | |
{question} | |
Y, N, or ? | |
Response: | |
""" | |
# Encode the prompt to a tensor | |
encoded_input = tokenizer.encode(prompt, return_tensors='pt') | |
# Get model predictions (logits) | |
with torch.no_grad(): | |
outputs = model(encoded_input) | |
predictions = outputs.logits | |
# Extract logits for 'Y', 'N', and '?' | |
logit_y = predictions[:, -1, tokenizer.encode('Y')[0]] | |
logit_n = predictions[:, -1, tokenizer.encode('N')[0]] | |
logit_q = predictions[:, -1, tokenizer.encode('?')[0]] | |
# Apply softmax to get probabilities | |
probs = F.softmax(torch.tensor([logit_y, logit_n, logit_q]), dim=0) | |
# Identify the maximum probability and its index | |
max_prob, max_index = torch.max(probs, 0) | |
# Map the index to the corresponding answer | |
answers = {0: 'Yes', 1: 'No', 2: 'I don\'t know'} | |
selected_answer = answers[max_index.item()] | |
print(f'Probability Yes: {probs[0].item()}') | |
print(f'Probability No: {probs[1].item()}') | |
print(f'Probability I don\'t know: {probs[2].item()}') | |
print(f'Is a bird a mammal? {selected_answer}') |
Author
thistleknot
commented
Jan 14, 2024
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment