titans-pytorch 0.1.1__tar.gz → 0.1.2__tar.gz

{titans_pytorch-0.1.1 → titans_pytorch-0.1.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: titans-pytorch
-Version: 0.1.1
+Version: 0.1.2
 Summary: Titans
 Project-URL: Homepage, https://pypi.org/project/titans-pytorch/
 Project-URL: Repository, https://github.com/lucidrains/titans-pytorch

{titans_pytorch-0.1.1 → titans_pytorch-0.1.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "titans-pytorch"
-version = "0.1.1"
+version = "0.1.2"
 description = "Titans"
 authors = [
     { name = "Phil Wang", email = "lucidrains@gmail.com" }

{titans_pytorch-0.1.1 → titans_pytorch-0.1.2}/tests/test_titans.py

@@ -50,10 +50,12 @@ def test_titans_attn_memory():
 
     assert seq.shape == retrieved.shape
 
+@pytest.mark.parametrize('seq_len', (1023, 17))
 @pytest.mark.parametrize('num_persist_mem_tokens', (0, 16))
 @pytest.mark.parametrize('num_longterm_mem_tokens', (0, 16))
 @pytest.mark.parametrize('neural_mem_gate_attn_output', (False, True))
 def test_mac(
+    seq_len,
     num_persist_mem_tokens,
     num_longterm_mem_tokens,
     neural_mem_gate_attn_output
@@ -70,13 +72,15 @@ def test_mac(
         neural_mem_gate_attn_output = neural_mem_gate_attn_output
     )
 
-    x = torch.randint(0, 256, (1, 1023))
+    x = torch.randint(0, 256, (1, seq_len))
 
     logits = transformer(x)
-    assert logits.shape == (1, 1023, 256)
+    assert logits.shape == (1, seq_len, 256)
 
+@pytest.mark.parametrize('seq_len', (1023, 17))
 @pytest.mark.parametrize('sliding', (True, False))
 def test_flex(
+    seq_len,
     sliding
 ):
     if not (torch.cuda.is_available() and exists(flex_attention)):
@@ -91,7 +95,7 @@ def test_flex(
         sliding = sliding
     ).cuda()
 
-    seq = torch.randn(1, 1019, 512).cuda()
+    seq = torch.randn(1, seq_len, 512).cuda()
 
     out_flex, _ = attn(seq)
     out_non_flex, _ = attn(seq, disable_flex_attn = True)

{titans_pytorch-0.1.1 → titans_pytorch-0.1.2}/titans_pytorch/mac_transformer.py

@@ -528,7 +528,8 @@ class MemoryAsContextTransformer(Module):
         filter_fn: Callable = min_p_filter,
         filter_kwargs: dict = dict(
             min_p = 0.1,
-        )
+        ),
+        show_progress = True
     ):
         was_training = self.training
         self.eval()
@@ -536,7 +537,9 @@ class MemoryAsContextTransformer(Module):
         prompt_seq_len, out = prompt.shape[-1], prompt.clone()
         sample_num_times = max(0, seq_len - prompt_seq_len)
 
-        for _ in tqdm.tqdm(range(sample_num_times)):
+        iter_wrap = tqdm.tqdm if show_progress else identity
+
+        for _ in iter_wrap(range(sample_num_times)):
             logits = self.forward(out, disable_flex_attn = True)
             logits = logits[:, -1]