app update

app.py

@@ -41,8 +41,9 @@ model.apply(_init_weights)
 model.load_state_dict
 
 
-checkpoint_dir = Path("out/redpajama/intermediate-ckpt-3_9.pth")
+checkpoint_dir = Path("out/redpajama/final-gpt-model-ckpt.pth")
 strategy = "auto"
+quantize = None
 devices = 1
 precision = None
 
@@ -61,14 +62,7 @@ model = fabric.setup_module(model)
 load_checkpoint(fabric, model, checkpoint_dir)
 
 tokenizer = Tokenizer(Path('tokenizer_config'))
-encoded = tokenizer.encode(prompt, device=fabric.device)
-prompt_length = encoded.size(0)
-max_returned_tokens = prompt_length + max_new_tokens
 
-with fabric.init_tensor():
-    # set the max_seq_length to limit the memory usage to what we need
-    model.max_seq_length = max_returned_tokens
-    
 @torch.inference_mode()
 def generate(
     model: GPT,