Fixes an issue where the LitGPT Python API was consuming too much mem…

…ory (#1590)

litgpt/api.py

@@ -163,8 +163,9 @@ def load(
         with fabric.init_module(empty_init=(num_devices > 1)):
             model = GPT(config)
 
-        with fabric.init_tensor():
-            model.set_kv_cache(batch_size=1)
+        # This should be set if we add a compile feature later
+        # with fabric.init_tensor():
+        #     model.set_kv_cache(batch_size=1)
 
         model.eval()
         model = fabric.setup_module(model)
@@ -178,6 +179,7 @@ def load(
             prompt_style=prompt_style, checkpoint_dir=checkpoint_dir, fabric=fabric,
         )
 
+    @torch.inference_mode()
     def generate(
         self,
         prompt: str,
@@ -221,6 +223,11 @@ def generate(
         prompt_length = input_ids.size(0)
         max_returned_tokens = prompt_length + max_new_tokens
 
+        first_turn = self.model.mask_cache is None
+        if first_turn or max_returned_tokens > self.model.max_seq_length:
+            self.model.max_seq_length = max_returned_tokens
+            self.model.set_kv_cache(batch_size=1, device=self.fabric.device)
+
         self.model.eval()
 
         if calculate_number_of_devices(self.devices) > 1:

tests/test_api.py

@@ -95,8 +95,24 @@ def test_llm_load_random_init(tmp_path):
         init="random",
         tokenizer_dir=Path(tmp_path/"EleutherAI/pythia-14m")
     )
-    text = llm.generate("text", max_new_tokens=10)
-    assert len(text.split(" ")) > 5
+
+    input_text = "some text text"
+    output_text = llm.generate(input_text, max_new_tokens=15)
+    ln = len(llm.preprocessor.tokenizer.encode(output_text)) - len(llm.preprocessor.tokenizer.encode(input_text))
+    assert ln <= 15
+
+    # The following below tests that generate works with different prompt lengths
+    # after the kv cache was set
+
+    input_text = "some text"
+    output_text = llm.generate(input_text, max_new_tokens=15)
+    ln = len(llm.preprocessor.tokenizer.encode(output_text)) - len(llm.preprocessor.tokenizer.encode(input_text))
+    assert ln <= 15
+
+    input_text = "some text text text"
+    output_text = llm.generate(input_text, max_new_tokens=15)
+    ln = len(llm.preprocessor.tokenizer.encode(output_text)) - len(llm.preprocessor.tokenizer.encode(input_text))
+    assert ln <= 15
 
 
 def test_llm_load_hub_init(tmp_path):