rxnn 0.2.21__py3-none-any.whl → 0.2.22__py3-none-any.whl

rxnn/training/rl.py

@@ -24,8 +24,6 @@ class RlAlgorithm(ABC):
         return self.critic_loss(rewards, values)
 
 class PPOConfig(TypedDict):
-    gae_gamma: float
-    gae_lambda: float
     clip_eps: float
 
 class PPOAlgorithm(RlAlgorithm):
@@ -33,8 +31,6 @@ class PPOAlgorithm(RlAlgorithm):
         super(PPOAlgorithm, self).__init__()
 
         # PPO Config
-        self.gae_gamma = config.get('gae_gamma', 0.99)
-        self.gae_lambda = config.get('gae_lambda', 0.95)
         self.clip_eps = config.get('clip_eps', 0.2)
 
     def policy_loss(self, query: TokenizedDict, answer: TokenizedDict, logits: torch.Tensor,
@@ -86,15 +82,6 @@ class PPOAlgorithm(RlAlgorithm):
 
         return policy_loss
 
-    # def _compute_gae(self, rewards: torch.Tensor, values: torch.Tensor, next_value: torch.Tensor) -> torch.Tensor:
-    #     advantages = torch.zeros_like(rewards, device=values.device)
-    #     last_advantage = 0
-    #     for t in reversed(range(rewards.size(0))):
-    #         delta = rewards[t] + self.gae_gamma * next_value - values[t]
-    #         advantages[t] = delta + self.gae_gamma * self.gae_lambda * last_advantage
-    #         last_advantage = advantages[t]
-    #     return advantages
-
     def calculate_advantages(self, rewards: torch.Tensor, values: torch.Tensor) -> torch.Tensor:
         advantages = rewards - values
         normalized_advantages = (advantages - advantages.mean()) / (advantages.std() + 1e-8)

{rxnn-0.2.21.dist-info → rxnn-0.2.22.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: rxnn
-Version: 0.2.21
+Version: 0.2.22
 Summary: RxNN: Reactive Neural Networks Platform
 License: Apache-2.0
 Keywords: deep-learning,ai,machine-learning

{rxnn-0.2.21.dist-info → rxnn-0.2.22.dist-info}/RECORD

@@ -18,7 +18,7 @@ rxnn/training/dataset.py,sha256=7hTilFWPpqUEc6zNcMqBPjxFKxCfvTKKF3E8tVlwccQ,5125
 rxnn/training/models.py,sha256=wf98gYKKm9-ZY3zwdX9NIeJ-lvh7Ro1SoAijmQxYM28,5599
 rxnn/training/mrl.py,sha256=zk4m1JFuX0y82J0tG2XkY0Pz6Uy2did9cngOXqR9lMk,43326
 rxnn/training/reward.py,sha256=7MTVdNm5HnWmt6zFDi3TAYmnVSL_-24riOoY2F7z4x8,11290
-rxnn/training/rl.py,sha256=DHFwnPUlnq2JVj6CS6DwifnC_eMeBAUVp36UCAWNMis,3934
+rxnn/training/rl.py,sha256=j-KNLoZjhaEKasYNOc8DxHtwvknAgAJFwvXKot6otFA,3272
 rxnn/training/scheduler.py,sha256=LcjU35mEwz2U5x3U6tLfeeYlBqMxbFSxYzJYuXkWbSY,1408
 rxnn/training/tokenizer.py,sha256=umaLByMBx_NMrQElA45HLm9gkuzyKWDTFaKVd-CjXl0,8344
 rxnn/training/utils.py,sha256=Bw8nZLKIt7NQpUVCYkb_79kWKChVFOYgYXwODo4SvNc,5718
@@ -32,7 +32,7 @@ rxnn/transformers/moe.py,sha256=j6jEx6Ip0zttlUZKKn82azxo95lkLZs-H2GLSMD88hY,5859
 rxnn/transformers/positional.py,sha256=1PjcJybUzeQlIKJI4tahAGZcYgCRCL0otxs7mpsNuzM,4410
 rxnn/transformers/sampler.py,sha256=t6iiQTdLQ0TakUWnnhKkb5DKF2F_9-thXHBydDF3fxg,17389
 rxnn/utils.py,sha256=ihb6OTyDtPiocB_lOvnq7eOkjjpCkgs8wxvXUBNQ7mM,996
-rxnn-0.2.21.dist-info/LICENSE,sha256=C8coDFIUYuOcke4JLPwTqahQUCyXyGq6WOaigOkx8tY,11275
-rxnn-0.2.21.dist-info/METADATA,sha256=XXf_qBMs2dOwWyAN5oNEg1W1-oPVIAQPy0FkNcO7QZQ,25960
-rxnn-0.2.21.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
-rxnn-0.2.21.dist-info/RECORD,,
+rxnn-0.2.22.dist-info/LICENSE,sha256=C8coDFIUYuOcke4JLPwTqahQUCyXyGq6WOaigOkx8tY,11275
+rxnn-0.2.22.dist-info/METADATA,sha256=KYQSidSUXwKxzuQj77h-jT0DUimKZVW_XTl-7PcQG3o,25960
+rxnn-0.2.22.dist-info/WHEEL,sha256=b4K_helf-jlQoXBBETfwnf4B04YC67LOev0jo4fX5m8,88
+rxnn-0.2.22.dist-info/RECORD,,