robo-lib 0.0.9__tar.gz → 0.0.11__tar.gz

{robo_lib-0.0.9 → robo_lib-0.0.11}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.3
+Metadata-Version: 2.4
 Name: robo_lib
-Version: 0.0.9
+Version: 0.0.11
 Summary: A package to create, configure, and train transformer models.
 Project-URL: Homepage, https://github.com/hamburgerfish/robo_pack
 Project-URL: Issues, https://github.com/hamburgerfish/robo_pack/issues
@@ -124,7 +124,7 @@ robo = rl.RoboConstructor(
     enc_block_size=100
 )
 
-robo.train(
+robo.train_robo(
     max_iters=20000,
     eval_interval=200,
     batch_size=128,

{robo_lib-0.0.9 → robo_lib-0.0.11}/README.md

@@ -107,7 +107,7 @@ robo = rl.RoboConstructor(
     enc_block_size=100
 )
 
-robo.train(
+robo.train_robo(
     max_iters=20000,
     eval_interval=200,
     batch_size=128,

{robo_lib-0.0.9 → robo_lib-0.0.11}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "robo_lib"
-version = "0.0.9"
+version = "0.0.11"
 authors = [
   { name="Erik Papp", email="erik3papp@gmail.com" },
 ]

{robo_lib-0.0.9 → robo_lib-0.0.11}/robo_lib/components.py

@@ -615,6 +615,7 @@ class RoboConstructor(nn.Module):
                  enc_vocab_size:int=None,
                  enc_block_size:int=None,
                  enc_expansion_factor:int=4,
+                 enc_positional_encoding:bool=True,
                  dropout:float=0.1,
                  device:str=None
                  ) -> None:
@@ -635,6 +636,7 @@ class RoboConstructor(nn.Module):
         self.dec_positional_embedding_table = nn.Embedding(dec_block_size, n_embed)
 
         if enc_n_blocks != 0:
+            self.enc_positional_encoding = enc_positional_encoding
             self.enc_n_blocks = enc_n_blocks
             self.enc_n_head = enc_n_head
             self.enc_expansion_factor = enc_expansion_factor
@@ -642,7 +644,8 @@ class RoboConstructor(nn.Module):
             self.enc_block_size = enc_block_size
             self.cross_attention = True
             self.enc_token_embedding_table = nn.Embedding(enc_vocab_size, n_embed)
-            self.enc_positional_embedding_table = nn.Embedding(enc_block_size, n_embed)
+            if enc_positional_encoding:
+                self.enc_positional_embedding_table = nn.Embedding(enc_block_size, n_embed)
             self.encoder_blocks = MySequential(*[EncoderBlock(n_embed, enc_n_head, enc_expansion_factor, dropout=dropout) for _ in range(enc_n_blocks)])
         else:
             self.cross_attention = False
@@ -685,8 +688,11 @@ class RoboConstructor(nn.Module):
 
         if self.cross_attention:
             enc_tok_emb = self.enc_token_embedding_table(enc_in)
-            enc_pos_emb = self.enc_positional_embedding_table(torch.arange(enc_T, device=self.device))
-            enc_x = enc_tok_emb + enc_pos_emb
+            if self.enc_positional_encoding:
+                enc_pos_emb = self.enc_positional_embedding_table(torch.arange(enc_T, device=self.device))
+                enc_x = enc_tok_emb + enc_pos_emb
+            else:
+                enc_x = enc_tok_emb
 
             enc_out, enc_mask = self.encoder_blocks(enc_x, enc_mask)
         else: