tidy

app.py

@@ -78,17 +78,10 @@ def generate(
     # Generate without threading
     with torch.no_grad():
         outputs = model.generate(**generation_kwargs)
-    print(outputs.sequences.shape, input_ids.shape)
     generated_tokens = outputs.sequences[0, input_ids.shape[1]:]
-    print("Generated tokens:", generated_tokens, generated_tokens.shape)
     generated_text = tokenizer.decode(generated_tokens, skip_special_tokens=True)
-    print("Generated text:", generated_text)
     # hidden states
     hidden = outputs.hidden_states  # list of tensors, one for each token, then (batch size, sequence length, hidden size)
-    print(len(hidden)) 
-    print(len(hidden[1]))     # layers
-    print(hidden[1][0].shape)  # (sequence length, hidden size)
-    # stack token embeddings 
 
     # TODO do this loop on the fly instead of waiting for the whole generation
     highlighted_text = ""