agilerl 2.7.0.dev2__tar.gz → 2.7.1__tar.gz

agilerl-2.7.1/.envrc

@@ -0,0 +1,5 @@
+export UV_PROJECT_ENVIRONMENT=".venv313"
+
+if [[ -d .venv313 ]]; then
+  source .venv313/bin/activate
+fi

agilerl-2.7.1/.gitattributes

@@ -0,0 +1,3 @@
+# Large test fixtures: treat as generated/binary so they don't bloat diffs,
+# blame, or PR reviews. This is a minified HuggingFace tokenizer used by tests.
+tests/assets/tiny_llm/tokenizer.json linguist-generated=true -diff

{agilerl-2.7.0.dev2 → agilerl-2.7.1}/.github/workflows/linux-tests.yml

@@ -33,7 +33,7 @@ jobs:
                 python-version: ['3.10', '3.11', '3.12', '3.13']
 
         container:
-            image: pytorch/pytorch:2.7.1-cuda12.6-cudnn9-devel
+            image: pytorch/pytorch:2.11.0-cuda13.0-cudnn9-devel
             options: --user root
 
         # Workspace (/__w) is ~1GB with little free space; root (/) has plenty. Put cache and venv on /.

{agilerl-2.7.0.dev2 → agilerl-2.7.1}/.pre-commit-config.yaml

@@ -35,7 +35,7 @@ repos:
                 - --skip=*.css,*.js,*.map,*.scss,*.svg
                 - --ignore-words-list=magent,pres,roate
     - repo: https://github.com/astral-sh/ruff-pre-commit
-      rev: v0.15.12
+      rev: v0.15.18
       hooks:
           - id: ruff
             name: Ruff Linter
@@ -50,7 +50,7 @@ repos:
 
     - repo: https://github.com/astral-sh/uv-pre-commit
       # uv version.
-      rev: 0.11.8
+      rev: 0.11.23
       hooks:
           - id: uv-lock
 

{agilerl-2.7.0.dev2 → agilerl-2.7.1}/PKG-INFO

@@ -1,59 +1,58 @@
 Metadata-Version: 2.4
 Name: agilerl
-Version: 2.7.0.dev2
+Version: 2.7.1
 Summary: AgileRL is a deep reinforcement learning library focused on improving RL development through RLOps.
 Author-email: Nick Ustaran-Anderegg <dev@agilerl.com>
 License-Expression: Apache-2.0
 License-File: LICENSE
 Requires-Python: <3.14,>=3.10
-Requires-Dist: accelerate~=1.7.0
-Requires-Dist: dill~=0.3.7
-Requires-Dist: fastrand~=1.3.0
-Requires-Dist: flatten-dict~=0.4.2
-Requires-Dist: google-cloud-storage~=2.5.0
-Requires-Dist: gymnasium~=1.0.0
-Requires-Dist: h5py~=3.15.0
-Requires-Dist: hydra-core~=1.3.2
-Requires-Dist: jax[cpu]~=0.4.31
-Requires-Dist: matplotlib<3.10,~=3.9.4
-Requires-Dist: minari[all]==0.5.2
+Requires-Dist: accelerate>=1.7.0
+Requires-Dist: dill>=0.3.7
+Requires-Dist: fastrand>=1.3.0
+Requires-Dist: flatten-dict>=0.4.2
+Requires-Dist: google-cloud-storage>=2.5.0
+Requires-Dist: gymnasium>=1.0.0
+Requires-Dist: h5py>=3.15.0
+Requires-Dist: hydra-core>=1.3.2
+Requires-Dist: jax[cpu]>=0.4.31
+Requires-Dist: matplotlib>=3.9.4
+Requires-Dist: minari[all]>=0.5.2
 Requires-Dist: numpy<3.0,>=2.0.0
-Requires-Dist: omegaconf~=2.3.0
+Requires-Dist: omegaconf>=2.3.0
 Requires-Dist: packaging>=20.0
-Requires-Dist: pandas~=2.2.3
-Requires-Dist: pettingzoo~=1.23.1
-Requires-Dist: pre-commit~=3.8.0
-Requires-Dist: pygame~=2.6.0
-Requires-Dist: pymunk~=6.2.0
-Requires-Dist: redis~=4.4.4
-Requires-Dist: supersuit~=3.9.0
-Requires-Dist: tensordict~=0.8
-Requires-Dist: termcolor~=1.1.0
-Requires-Dist: torch==2.8.0; sys_platform == 'darwin'
-Requires-Dist: torch==2.9.0; sys_platform != 'darwin'
+Requires-Dist: pandas>=2.2.3
+Requires-Dist: pettingzoo>=1.23.1
+Requires-Dist: pre-commit>=3.8.0
+Requires-Dist: pygame>=2.6.0
+Requires-Dist: pymunk<7.3,>=6.2
+Requires-Dist: redis<8.1.0,>=4.4.4
+Requires-Dist: supersuit>=3.9.0
+Requires-Dist: tensordict==0.13.0
+Requires-Dist: termcolor>=1.1.0
+Requires-Dist: torch>=2.11.0
 Requires-Dist: tqdm>=4.66.4
-Requires-Dist: wandb~=0.18.0
+Requires-Dist: wandb>=0.18.0
 Provides-Extra: all
-Requires-Dist: datasets==4.4.1; extra == 'all'
-Requires-Dist: deepspeed~=0.17.1; (sys_platform != 'win32') and extra == 'all'
-Requires-Dist: gymnasium[box2d]~=1.0.0; (sys_platform != 'win32') and extra == 'all'
-Requires-Dist: liger-kernel==0.7.0; (sys_platform == 'linux') and extra == 'all'
-Requires-Dist: peft~=0.18.0; extra == 'all'
-Requires-Dist: swig~=4.4.1; (sys_platform != 'win32') and extra == 'all'
-Requires-Dist: transformers~=4.57.1; extra == 'all'
-Requires-Dist: vllm==0.11.0; (sys_platform == 'darwin') and extra == 'all'
-Requires-Dist: vllm==0.13.0; (sys_platform == 'linux') and extra == 'all'
+Requires-Dist: bitsandbytes==0.49.2; (sys_platform == 'linux') and extra == 'all'
+Requires-Dist: datasets==5.0.0; extra == 'all'
+Requires-Dist: deepspeed>=0.19.1; (sys_platform == 'linux') and extra == 'all'
+Requires-Dist: gymnasium[box2d]>=1.0.0; (sys_platform != 'win32') and extra == 'all'
+Requires-Dist: liger-kernel==0.8.0; (sys_platform == 'linux') and extra == 'all'
+Requires-Dist: peft==0.19.1; extra == 'all'
+Requires-Dist: swig>=4.4.1; (sys_platform != 'win32') and extra == 'all'
+Requires-Dist: transformers==5.11.0; extra == 'all'
+Requires-Dist: vllm>=0.23; (sys_platform == 'linux') and extra == 'all'
 Provides-Extra: box2d
-Requires-Dist: gymnasium[box2d]~=1.0.0; (sys_platform != 'win32') and extra == 'box2d'
-Requires-Dist: swig~=4.4.1; (sys_platform != 'win32') and extra == 'box2d'
+Requires-Dist: gymnasium[box2d]>=1.0.0; (sys_platform != 'win32') and extra == 'box2d'
+Requires-Dist: swig>=4.4.1; (sys_platform != 'win32') and extra == 'box2d'
 Provides-Extra: llm
-Requires-Dist: datasets==4.4.1; extra == 'llm'
-Requires-Dist: deepspeed~=0.17.1; (sys_platform != 'win32') and extra == 'llm'
-Requires-Dist: liger-kernel==0.7.0; (sys_platform == 'linux') and extra == 'llm'
-Requires-Dist: peft~=0.18.0; extra == 'llm'
-Requires-Dist: transformers~=4.57.1; extra == 'llm'
-Requires-Dist: vllm==0.11.0; (sys_platform == 'darwin') and extra == 'llm'
-Requires-Dist: vllm==0.13.0; (sys_platform == 'linux') and extra == 'llm'
+Requires-Dist: bitsandbytes==0.49.2; (sys_platform == 'linux') and extra == 'llm'
+Requires-Dist: datasets==5.0.0; extra == 'llm'
+Requires-Dist: deepspeed>=0.19.1; (sys_platform == 'linux') and extra == 'llm'
+Requires-Dist: liger-kernel==0.8.0; (sys_platform == 'linux') and extra == 'llm'
+Requires-Dist: peft==0.19.1; extra == 'llm'
+Requires-Dist: transformers==5.11.0; extra == 'llm'
+Requires-Dist: vllm>=0.23; (sys_platform == 'linux') and extra == 'llm'
 Description-Content-Type: text/markdown
 
 <p align="center">
@@ -153,7 +152,7 @@ We are constantly updating our tutorials to showcase the latest features of Agil
 | [Hierarchical curriculum learning](https://docs.agilerl.com/en/latest/tutorials/skills/index.html) | Shows how to teach agents Skills and combine them to achieve an end goal. | [PPO - Lunar Lander](https://docs.agilerl.com/en/latest/tutorials/skills/index.html) |
 | [Contextual multi-arm bandits](https://docs.agilerl.com/en/latest/tutorials/bandits/index.html) | Learn to make the correct decision in environments that only have one timestep. | [NeuralUCB - Iris Dataset](https://docs.agilerl.com/en/latest/tutorials/bandits/agilerl_neural_ucb_tutorial.html) <br> [NeuralTS - PenDigits](https://docs.agilerl.com/en/latest/tutorials/bandits/agilerl_neural_ts_tutorial.html) |
 | [Custom Modules & Networks](https://docs.agilerl.com/en/latest/tutorials/custom_networks/index.html) | Learn how to create custom evolvable modules and networks for RL algorithms. | [Dueling Distributional Q Network](https://docs.agilerl.com/en/latest/tutorials/custom_networks/agilerl_rainbow_tutorial.html) <br> [EvolvableSimBa](https://docs.agilerl.com/en/latest/tutorials/custom_networks/agilerl_simba_tutorial.html) |
-| [LLM Finetuning](https://docs.agilerl.com/en/latest/tutorials/llm_finetuning/index.html) | Learn how to finetune an LLM using AgileRL. | [GRPO](https://docs.agilerl.com/en/latest/tutorials/llm_finetuning/index.html) |
+| [LLM Finetuning](https://docs.agilerl.com/en/latest/tutorials/llm_finetuning/index.html) | Learn how to finetune an LLM using AgileRL. | [GRPO](https://docs.agilerl.com/en/latest/tutorials/llm_finetuning/grpo_finetuning.html) <br> [GRPO with HPO](https://docs.agilerl.com/en/latest/tutorials/llm_finetuning/grpo_hpo.html) <br> [SFT & DPO](https://docs.agilerl.com/en/latest/tutorials/llm_finetuning/sft_dpo_finetuning.html) <br> [Multi-turn LLMPPO, LLMREINFORCE & GRPO](https://docs.agilerl.com/en/latest/tutorials/llm_finetuning/multiturn_grpo_ppo.html) |
 
 ## Evolvable algorithms (more coming soon!)
 
@@ -179,10 +178,11 @@ We are constantly updating our tutorials to showcase the latest features of Agil
 
   ### LLM Fine-tuning Algorithms
 
-  | RL         | Algorithm |
-  | ---------- | --------- |
-  | [On-Policy](https://docs.agilerl.com/en/latest/llm_finetuning/index.html) | [Group Relative Policy Optimization (GRPO)](https://docs.agilerl.com/en/latest/api/algorithms/grpo.html)
-  | [Off-Policy](https://docs.agilerl.com/en/latest/llm_finetuning/index.html) | [Direct Preference Optimization (DPO)](https://docs.agilerl.com/en/latest/api/algorithms/dpo.html)
+  | Type         | Algorithm |
+  | ------------ | --------- |
+  | [On-Policy RL](https://docs.agilerl.com/en/latest/llm_finetuning/index.html) | [Group Relative Policy Optimization (GRPO)](https://docs.agilerl.com/en/latest/api/algorithms/grpo.html) <br> [Clipped IS-weight Policy Optimization (CISPO)](https://docs.agilerl.com/en/latest/api/algorithms/cispo.html) <br> [Group Sequence Policy Optimization (GSPO)](https://docs.agilerl.com/en/latest/api/algorithms/gspo.html) <br> [LLM Proximal Policy Optimization (LLM PPO)](https://docs.agilerl.com/en/latest/api/algorithms/llmppo.html) <br> [LLM REINFORCE](https://docs.agilerl.com/en/latest/api/algorithms/llmreinforce.html) |
+  | [Preference Optimization](https://docs.agilerl.com/en/latest/llm_finetuning/index.html) | [Direct Preference Optimization (DPO)](https://docs.agilerl.com/en/latest/api/algorithms/dpo.html) |
+  | [Supervised Fine-Tuning](https://docs.agilerl.com/en/latest/llm_finetuning/index.html) | [Supervised Fine-Tuning (SFT)](https://docs.agilerl.com/en/latest/api/algorithms/sft.html) |
 
 
 ## Train an Agent to Beat a Gym Environment

{agilerl-2.7.0.dev2 → agilerl-2.7.1}/README.md

@@ -95,7 +95,7 @@ We are constantly updating our tutorials to showcase the latest features of Agil
 | [Hierarchical curriculum learning](https://docs.agilerl.com/en/latest/tutorials/skills/index.html) | Shows how to teach agents Skills and combine them to achieve an end goal. | [PPO - Lunar Lander](https://docs.agilerl.com/en/latest/tutorials/skills/index.html) |
 | [Contextual multi-arm bandits](https://docs.agilerl.com/en/latest/tutorials/bandits/index.html) | Learn to make the correct decision in environments that only have one timestep. | [NeuralUCB - Iris Dataset](https://docs.agilerl.com/en/latest/tutorials/bandits/agilerl_neural_ucb_tutorial.html) <br> [NeuralTS - PenDigits](https://docs.agilerl.com/en/latest/tutorials/bandits/agilerl_neural_ts_tutorial.html) |
 | [Custom Modules & Networks](https://docs.agilerl.com/en/latest/tutorials/custom_networks/index.html) | Learn how to create custom evolvable modules and networks for RL algorithms. | [Dueling Distributional Q Network](https://docs.agilerl.com/en/latest/tutorials/custom_networks/agilerl_rainbow_tutorial.html) <br> [EvolvableSimBa](https://docs.agilerl.com/en/latest/tutorials/custom_networks/agilerl_simba_tutorial.html) |
-| [LLM Finetuning](https://docs.agilerl.com/en/latest/tutorials/llm_finetuning/index.html) | Learn how to finetune an LLM using AgileRL. | [GRPO](https://docs.agilerl.com/en/latest/tutorials/llm_finetuning/index.html) |
+| [LLM Finetuning](https://docs.agilerl.com/en/latest/tutorials/llm_finetuning/index.html) | Learn how to finetune an LLM using AgileRL. | [GRPO](https://docs.agilerl.com/en/latest/tutorials/llm_finetuning/grpo_finetuning.html) <br> [GRPO with HPO](https://docs.agilerl.com/en/latest/tutorials/llm_finetuning/grpo_hpo.html) <br> [SFT & DPO](https://docs.agilerl.com/en/latest/tutorials/llm_finetuning/sft_dpo_finetuning.html) <br> [Multi-turn LLMPPO, LLMREINFORCE & GRPO](https://docs.agilerl.com/en/latest/tutorials/llm_finetuning/multiturn_grpo_ppo.html) |
 
 ## Evolvable algorithms (more coming soon!)
 
@@ -121,10 +121,11 @@ We are constantly updating our tutorials to showcase the latest features of Agil
 
   ### LLM Fine-tuning Algorithms
 
-  | RL         | Algorithm |
-  | ---------- | --------- |
-  | [On-Policy](https://docs.agilerl.com/en/latest/llm_finetuning/index.html) | [Group Relative Policy Optimization (GRPO)](https://docs.agilerl.com/en/latest/api/algorithms/grpo.html)
-  | [Off-Policy](https://docs.agilerl.com/en/latest/llm_finetuning/index.html) | [Direct Preference Optimization (DPO)](https://docs.agilerl.com/en/latest/api/algorithms/dpo.html)
+  | Type         | Algorithm |
+  | ------------ | --------- |
+  | [On-Policy RL](https://docs.agilerl.com/en/latest/llm_finetuning/index.html) | [Group Relative Policy Optimization (GRPO)](https://docs.agilerl.com/en/latest/api/algorithms/grpo.html) <br> [Clipped IS-weight Policy Optimization (CISPO)](https://docs.agilerl.com/en/latest/api/algorithms/cispo.html) <br> [Group Sequence Policy Optimization (GSPO)](https://docs.agilerl.com/en/latest/api/algorithms/gspo.html) <br> [LLM Proximal Policy Optimization (LLM PPO)](https://docs.agilerl.com/en/latest/api/algorithms/llmppo.html) <br> [LLM REINFORCE](https://docs.agilerl.com/en/latest/api/algorithms/llmreinforce.html) |
+  | [Preference Optimization](https://docs.agilerl.com/en/latest/llm_finetuning/index.html) | [Direct Preference Optimization (DPO)](https://docs.agilerl.com/en/latest/api/algorithms/dpo.html) |
+  | [Supervised Fine-Tuning](https://docs.agilerl.com/en/latest/llm_finetuning/index.html) | [Supervised Fine-Tuning (SFT)](https://docs.agilerl.com/en/latest/api/algorithms/sft.html) |
 
 
 ## Train an Agent to Beat a Gym Environment

{agilerl-2.7.0.dev2 → agilerl-2.7.1}/agilerl/__init__.py

@@ -16,19 +16,6 @@ def get_extra_dependencies(package: str, extra: str) -> list[str]:
     return deps
 
 
-def _is_dist_installed(dist_name: str) -> bool:
-    """Check if a distribution is installed by its PyPI name.
-
-    Uses distribution metadata instead of find_spec because PyPI names
-    can differ from import names.
-    """
-    try:
-        metadata(dist_name)
-        return True
-    except PackageNotFoundError:
-        return False
-
-
 LLM_PACKAGES = get_extra_dependencies("agilerl", "llm")
 
 
@@ -40,7 +27,7 @@ def _is_distribution_installed(distribution: str) -> bool:
     return True
 
 
-# Use these flags for laxzy import checks
+# Use these flags for lazy import checks
 HAS_LLM_DEPENDENCIES = all(_is_distribution_installed(pkg) for pkg in LLM_PACKAGES)
 HAS_LIGER_KERNEL = _is_distribution_installed("liger-kernel")
 HAS_VLLM = _is_distribution_installed("vllm")

{agilerl-2.7.0.dev2 → agilerl-2.7.1}/agilerl/algorithms/cispo.py

@@ -2,25 +2,19 @@
 
 from __future__ import annotations
 
-from functools import partial
 from typing import Any
 
-from agilerl.algorithms.grpo import GRPO, _signatures_without_loss_type
+from agilerl.algorithms.grpo import GRPO
+from agilerl.utils.algo_utils import inherit_init_signature
 
 
+@inherit_init_signature(GRPO, fixed={"loss_type"})
 class CISPO(GRPO):
     """CISPO loss variant of :class:`agilerl.algorithms.grpo.GRPO`
 
     Paper: https://arxiv.org/abs/2506.13585
     """
 
-    _init_with_cispo = partial(GRPO.__init__, loss_type="cispo")
-
     def __init__(self, *args: Any, **kwargs: Any) -> None:
         """Initialize a CISPO agent with fixed ``loss_type``."""
-        self._init_with_cispo(self, *args, **kwargs)
-
-
-_CISPO_CLASS_SIG, _CISPO_INIT_SIG = _signatures_without_loss_type()
-CISPO.__signature__ = _CISPO_CLASS_SIG
-CISPO.__init__.__signature__ = _CISPO_INIT_SIG
+        super().__init__(*args, loss_type="cispo", **kwargs)

{agilerl-2.7.0.dev2 → agilerl-2.7.1}/agilerl/algorithms/core/__init__.py

@@ -1,7 +1,14 @@
-from .base import EvolvableAlgorithm, LLMAlgorithm, MultiAgentRLAlgorithm, RLAlgorithm
+from .base import (
+    ActionResult,
+    EvolvableAlgorithm,
+    LLMAlgorithm,
+    MultiAgentRLAlgorithm,
+    RLAlgorithm,
+)
 from .optimizer_wrapper import OptimizerWrapper
 
 __all__ = [
+    "ActionResult",
     "EvolvableAlgorithm",
     "LLMAlgorithm",
     "MultiAgentRLAlgorithm",