Loading a test set when extracting a d-vector - model wants the training set size

So I have to extract a d-vector (d-vector is the averaged activation from the last hidden layer of this DNN) of the taught model. I pass in a checkpoint from the last epoch and want to create a list of vectors. My training set is 1000 but my testing one is 20.

x = torch.stack(array_of_test_items)
checkpoint = torch.load(path, map_location=torch.device('cpu'))

model = Model(length_of_the_test_item_array) #20
model.load_state_dict(checkpoint['state_dict']) #a .tar from torch.save
model.eval()
with torch.no_grad():
    dvectors = model.extractTheVector(x)

but it throws an error of :

size mismatch for classifier.proj.weight: copying a param with shape torch.Size([1000, 1024]) from checkpoint, the shape in current model is torch.Size([20, 1024]). size mismatch for classifier.proj.bias: copying a param with shape torch.Size([1000]) from checkpoint, the shape in current model is torch.Size([20]).

In my Model class:

def extractTheVector(self, x):
    batch_size, num_chunks, features  = x.shape
    x = x.reshape((batch_size * num_chunks, 1, features))
    embeddings = self.encoder(x)
    embeddings = embeddings.reshape(batch_size, num_chunks, -1)
    return self.classifier.generateVector(embeddings) #dvectors

in Classifier:

class Classifier(nn.Module):
    def __init__(self, num_speakers): #num speakers = 1000 or 20
        super(Classifier, self).__init__()
        self.fc1 = nn.Linear(1024, 1024) #inner layer
        self.proj = nn.Linear(1024, num_speakers)

    def generateVector(self, x):
        x = self.encode(x)
        x = torch.nn.functional.normalize(x, dim=-1)
        x = torch.mean(x, 1)
        return x

    def encode(self, x):
        return Fun.relu(self.fc1(x))

    def forward(self, x):
        x = self.encode(x)
        x = Fun.softmax(self.proj(x), dim=1)
        return x

Why is it acting as if it's telling it to keep on learning?