agilerl 2.3.5.dev0__tar.gz → 2.4.0.dev0__tar.gz

{agilerl-2.3.5.dev0 → agilerl-2.4.0.dev0}/LICENSE

@@ -199,3 +199,16 @@
    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    See the License for the specific language governing permissions and
    limitations under the License.
+
+---
+
+THIRD-PARTY SOFTWARE NOTICES AND INFORMATION
+
+This project incorporates code from the following third-party software:
+
+TRL (Transformer Reinforcement Learning)
+Copyright 2020-2025 The HuggingFace Team
+Licensed under Apache License 2.0
+https://github.com/huggingface/trl
+
+Portions of the GRPO implementation use code patterns adapted from TRL's GRPOTrainer.

{agilerl-2.3.5.dev0 → agilerl-2.4.0.dev0}/PKG-INFO

@@ -1,8 +1,9 @@
-Metadata-Version: 2.3
+Metadata-Version: 2.4
 Name: agilerl
-Version: 2.3.5.dev0
+Version: 2.4.0.dev0
 Summary: AgileRL is a deep reinforcement learning library focused on improving RL development through RLOps.
 License: Apache 2.0
+License-File: LICENSE
 Author: Nick Ustaran-Anderegg
 Author-email: dev@agilerl.com
 Requires-Python: >=3.10,<4.0
@@ -12,6 +13,7 @@ Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
 Requires-Dist: SuperSuit (>=3.9.0,<4.0.0)
 Requires-Dist: accelerate (>=1.7.0,<2.0.0)
 Requires-Dist: deepspeed (>=0.17.1,<0.18.0)
@@ -275,7 +277,7 @@ agent_pop = create_population(
 Next, create the tournament, mutations and experience replay buffer objects that allow agents to share memory and efficiently perform evolutionary HPO.
 
 <details>
-<summary>Mutations and Tournament Seelection Example</summary>
+<summary>Mutations and Tournament Selection Example</summary>
 
 ```python
 from agilerl.components.replay_buffer import ReplayBuffer

{agilerl-2.3.5.dev0 → agilerl-2.4.0.dev0}/README.md

@@ -228,7 +228,7 @@ agent_pop = create_population(
 Next, create the tournament, mutations and experience replay buffer objects that allow agents to share memory and efficiently perform evolutionary HPO.
 
 <details>
-<summary>Mutations and Tournament Seelection Example</summary>
+<summary>Mutations and Tournament Selection Example</summary>
 
 ```python
 from agilerl.components.replay_buffer import ReplayBuffer

{agilerl-2.3.5.dev0 → agilerl-2.4.0.dev0}/agilerl/algorithms/__init__.py

@@ -1,6 +1,7 @@
 from .bc_lm import BC_LM, BC_Evaluator, BC_Policy
 from .cqn import CQN
 from .ddpg import DDPG
+from .dpo import DPO
 from .dqn import DQN
 from .dqn_rainbow import RainbowDQN
 from .grpo import GRPO
@@ -30,4 +31,5 @@ __all__ = [
     "PPO",
     "TD3",
     "GRPO",
+    "DPO",
 ]

{agilerl-2.3.5.dev0 → agilerl-2.4.0.dev0}/agilerl/algorithms/bc_lm.py

@@ -1,4 +1,4 @@
-from typing import Any, Callable, Optional, Tuple, Union
+from typing import Any, Callable, Optional, Union
 
 import numpy as np
 import torch
@@ -167,7 +167,7 @@ class BC_LM(nn.Module):
         temp: float = 1.0,
         top_k: Optional[int] = None,
         top_p: Optional[float] = None,
-    ) -> Tuple[torch.Tensor, Any]:
+    ) -> tuple[torch.Tensor, Any]:
         prepared_inputs = self.prepare_inputs(items)
         tokens = prepared_inputs["tokens"]
         scores, model_outputs = self.score(
@@ -189,7 +189,7 @@ class BC_LM(nn.Module):
         temp: float = 1.0,
         top_k: Optional[int] = None,
         top_p: Optional[float] = None,
-    ) -> Tuple[torch.Tensor, Any]:
+    ) -> tuple[torch.Tensor, Any]:
         scores, model_outputs = self.score(
             (
                 tokens.unsqueeze(1),