sum the kl div loss across number of actions by default for action embedder .kl_div

dreamer4/dreamer4.py

@@ -72,20 +72,22 @@ TokenizerLosses = namedtuple('TokenizerLosses', ('recon', 'lpips'))
 
 WorldModelLosses = namedtuple('WorldModelLosses', ('flow', 'rewards', 'discrete_actions', 'continuous_actions'))
 
+MaybeTensor = Tensor | None
+
 @dataclass
 class Experience:
     latents: Tensor
-    video: Tensor | None = None
-    proprio: Tensor | None = None
-    agent_embed: Tensor | None = None
+    video: MaybeTensor = None
+    proprio: MaybeTensor = None
+    agent_embed: MaybeTensor = None
     rewards: Tensor | None = None
-    actions: tuple[Tensor, Tensor] | None = None
-    log_probs: tuple[Tensor, Tensor] | None = None
-    old_action_unembeds: tuple[Tensor, Tensor] | None = None
-    values: Tensor | None = None
+    actions: tuple[MaybeTensor, MaybeTensor] | None = None
+    log_probs: tuple[MaybeTensor, MaybeTensor] | None = None
+    old_action_unembeds: tuple[MaybeTensor, MaybeTensor] | None = None
+    values: MaybeTensor = None
     step_size: int | None = None
-    lens: Tensor | None = None
-    is_truncated: Tensor | None = None
+    lens: MaybeTensor = None
+    is_truncated: MaybeTensor = None
     agent_index: int = 0
     is_from_world_model: bool = True
 
@@ -850,9 +852,10 @@ class ActionEmbedder(Module):
 
     def kl_div(
         self,
-        src: tuple[Tensor | None, Tensor | None],
-        tgt: tuple[Tensor | None, Tensor | None]
-    ) -> tuple[Tensor | None, Tensor | None]:
+        src: tuple[MaybeTensor, MaybeTensor],
+        tgt: tuple[MaybeTensor, MaybeTensor],
+        reduce_across_num_actions = True
+    ) -> tuple[MaybeTensor, MaybeTensor]:
 
         src_discrete, src_continuous = src
         tgt_discrete, tgt_continuous = tgt
@@ -894,6 +897,15 @@ class ActionEmbedder(Module):
 
             continuous_kl_div = kl.kl_divergence(src_normal, tgt_normal)
 
+        # maybe reduce
+
+        if reduce_across_num_actions:
+            if exists(discrete_kl_div):
+                discrete_kl_div = discrete_kl_div.sum(dim = -1)
+
+            if exists(continuous_kl_div):
+                continuous_kl_div = continuous_kl_div.sum(dim = -1)
+
         return discrete_kl_div, continuous_kl_div
 
     def forward(

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "dreamer4"
-version = "0.0.94"
+version = "0.0.95"
 description = "Dreamer 4"
 authors = [
     { name = "Phil Wang", email = "lucidrains@gmail.com" }

tests/test_dreamer.py

@@ -352,8 +352,8 @@ def test_action_embedder():
 
     discrete_kl_div, continuous_kl_div = embedder.kl_div((discrete_logits, continuous_mean_log_var), (discrete_logits_tgt, continuous_mean_log_var_tgt))
 
-    assert discrete_kl_div.shape == (2, 3, 2)
-    assert continuous_kl_div.shape == (2, 3, 2)
+    assert discrete_kl_div.shape == (2, 3)
+    assert continuous_kl_div.shape == (2, 3)
 
     # return discrete split by number of actions