Logits are raw, unnormalized model outputs that represent evidence for possible classes before conversion into probabilities. Their values can be compared directly to select the highest-scoring class. section 3.2.3 section 6.2.2
Reading sentiment logits
A sequence-classification model returns logits with shape (1, 2): one batch item and two classes. For [-4.3428, 4.6955], argmax selects class 1, interpreted here as positive sentiment. section 3.2.3
Listing 6.12 Using a multimodal model to simultaneously process visual and textual data
from transformers import CLIPProcessor, CLIPModel
model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32") #1
processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
labels = ["cat", "dog", "tiger", "train"]
inputs = processor(text = labels, #2
images = image,
return_tensors = "pt",
padding=True)
model.eval()
with torch.no_grad():
outputs = model(**inputs) #3
logits_per_image = outputs.logits_per_image #4
probs = logits_per_image.softmax(dim=1) #5
most_likely_index = torch.argmax(probs, dim=1).item() #6
most_likely_object = labels[most_likely_index]
print(f"The most likely object is: {most_likely_object}")
Run a CLIP image-label comparison that extracts logits, applies softmax, and selects the most likely label. section 6.2.2
