textpolicy 0.1.0__py3-none-any.whl → 0.1.2__py3-none-any.whl

textpolicy/__init__.py

@@ -30,7 +30,7 @@ from .validate import validate_installation
 
 # Export core reward functions and the reward decorator
 from .rewards.basic import length_reward, keyword_reward, perplexity_reward, accuracy_reward
-from .rewards.registry import reward
+from .rewards.registry import reward, verifier
 
 # Build __all__ combining submodule __all__ lists and additional symbols
 __all__ = (
@@ -48,5 +48,6 @@ __all__ = (
         "perplexity_reward",
         "accuracy_reward",
         "reward",
+        "verifier",
     ]
 )

textpolicy/environment/text_generation.py

@@ -630,24 +630,31 @@ class TextGenerationEnv(Environment):
         reward_fn: Callable[[str, str, dict], float],
         max_tokens: int = 25,
         seed: int = 42,
-        tokenizer: Any = None
+        tokenizer: Any = None,
+        examples: Optional[List[dict]] = None
     ):
         """
         Initialize simple text generation environment.
-        
+
         Args:
             prompts: List of prompts to cycle through
             reward_fn: Function that computes reward from (prompt, completion, example)
             max_tokens: Maximum tokens to generate per response
             seed: Random seed for reproducible behavior
             tokenizer: Tokenizer for converting prompts to tokens (required for MLX compatibility)
+            examples: Optional list of example dicts to pass to reward function. If provided,
+                      must have same length as prompts. examples[i] is passed when prompts[i] is used.
         """
         super().__init__()
         
         if tokenizer is None:
             raise ValueError("tokenizer is required for TextGenerationEnv to work with MLX rollout system")
-        
+
+        if examples is not None and len(examples) != len(prompts):
+            raise ValueError(f"examples length ({len(examples)}) must match prompts length ({len(prompts)})")
+
         self.prompts = prompts
+        self.examples = examples if examples is not None else [{} for _ in prompts]
         self.reward_fn = reward_fn
         self.max_tokens = max_tokens
         self.tokenizer = tokenizer
@@ -735,10 +742,11 @@ class TextGenerationEnv(Environment):
         
         # Compute reward using provided reward function
         # Pass tokenizer for EOS token detection and truncation detection
+        prompt_index = self.current_episode % len(self.prompts)
         reward = self.reward_fn(
             prompt=self.current_prompt,
             completion=response_text,
-            example={},
+            example=self.examples[prompt_index],
             tokenizer=self.tokenizer,  # Pass tokenizer for EOS detection
             truncated=truncated        # Pass truncation flag from environment
         )

{textpolicy-0.1.0.dist-info → textpolicy-0.1.2.dist-info}/METADATA

@@ -1,13 +1,23 @@
 Metadata-Version: 2.4
 Name: textpolicy
-Version: 0.1.0
-Summary: MLX-optimized reward and verification system for text generation RL
+Version: 0.1.2
+Summary: Reinforcement learning for text generation on MLX (Apple Silicon): GRPO/GSPO, environments, rollout, rewards, LoRA/QLoRA
+Project-URL: Homepage, https://github.com/teilomillet/textpolicy
+Project-URL: Repository, https://github.com/teilomillet/textpolicy
+Project-URL: Documentation, https://github.com/teilomillet/textpolicy#readme
+Project-URL: Changelog, https://github.com/teilomillet/textpolicy/blob/main/CHANGELOG.md
+Keywords: reinforcement-learning,text-generation,mlx,apple-silicon,lora,qlora,grpo,gspo,rlhf
+Classifier: Programming Language :: Python :: 3
+Classifier: Operating System :: MacOS
+Classifier: Intended Audience :: Developers
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: License :: OSI Approved :: MIT License
 Requires-Python: >=3.12
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: numpy>=2.3.2
-Requires-Dist: mlx>=0.21.0
-Requires-Dist: mlx-lm>=0.21.0
+Requires-Dist: mlx>=0.22.0
+Requires-Dist: mlx-lm>=0.22.0
 Requires-Dist: gymnasium>=0.29.0
 Requires-Dist: psutil>=7.0.0
 Requires-Dist: wandb>=0.21.1

{textpolicy-0.1.0.dist-info → textpolicy-0.1.2.dist-info}/RECORD

@@ -1,4 +1,4 @@
-textpolicy/__init__.py,sha256=u4u0fIHfAvXFN2ATHCsG0Tx4xGfOcfuOITBTmKbGhrw,1576
+textpolicy/__init__.py,sha256=vDAHJ826gKuTZUjcAftzz-RTX8KuOjH50Uj1RMhjTIQ,1606
 textpolicy/__main__.py,sha256=IlGmjJaW-DJUC7yhxUhbwNOZA3GxkeQGkVbFdS3_wBI,136
 textpolicy/cli.py,sha256=3CcJzrRlin1pgd6Mh312Xp3-EihHtTSvhakyYpdfacs,2107
 textpolicy/validate.py,sha256=lxmegz83B_c-PS3cFHaaL3c9fgWrEaLsDLkpPFtSj8Y,3780
@@ -16,7 +16,7 @@ textpolicy/environment/environment.py,sha256=o8-RY6wj5xrzDBp77HoY2At3XlBwvreF3DK
 textpolicy/environment/factory.py,sha256=pebQo1_M3sMF8Pdc9yvpdXzRXfIDllKJoAQAjQbif0E,3124
 textpolicy/environment/gym.py,sha256=P8Bi8PlDtcWWa9uLuCjkhZnYRVs-mg6iSJVSBkG99f8,3186
 textpolicy/environment/task_suites.py,sha256=ssPnw2Y3eGYaskWf8dUab4rNu_Bx5L284b3VdhgvSPM,1544
-textpolicy/environment/text_generation.py,sha256=BXSJS_05Q89cPFfdXcUKxOXSZm3HBR3KMi55BnOdoLY,31258
+textpolicy/environment/text_generation.py,sha256=Jql0pEfrPp9tqNsPOAdIP-UYoAUsfV969TMR2uPkUp4,31837
 textpolicy/environment/vectorized.py,sha256=ZROtpmdbh1Oi8c0b0D_vmVzqI16Cp2WZTmkjkRbMoDg,9932
 textpolicy/generation/__init__.py,sha256=J3dc0SPAZChJTsRn47tz8FfIp3XwNgZ-8_H9VBpQYvQ,1266
 textpolicy/generation/lora.py,sha256=xSKRczJY20BrkkU1SSgBtDc30tZjdFE7FhEZPUEoiyg,13747
@@ -58,9 +58,9 @@ textpolicy/utils/logging/tensorboard.py,sha256=aY9YMReSJkWEhy6SdAAUlHSB4lzDecivB
 textpolicy/utils/logging/wandb.py,sha256=U4pxuZNOz2l8XiymK8OFbCpiRTBOLNtnZakC_udttfQ,2206
 textpolicy/validation/__init__.py,sha256=KcyppNi91w0bF51gZ0ykUIKEiF7z6TT37uuavMFScnA,328
 textpolicy/validation/logprob_validation.py,sha256=G_CCy5NRDUTmo7WZIChhNVM3NtP1VmWAjdd5z6TIvos,11749
-textpolicy-0.1.0.dist-info/licenses/LICENSE,sha256=AYDHSNRbiqZt4HHH1gaOoQ2hjYjK4bqw4Vd9UyKzx18,1065
-textpolicy-0.1.0.dist-info/METADATA,sha256=XdyIh8e2IIRymRf31vu1MuVM2aaut2qsZ5PcsjHrl9Y,3199
-textpolicy-0.1.0.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-textpolicy-0.1.0.dist-info/entry_points.txt,sha256=d0Cj5boT6k_l_beVPWPt9LZMllsN4kbIUmsNsn1BANE,51
-textpolicy-0.1.0.dist-info/top_level.txt,sha256=Ww6_QEF71dI-AYCaugiGeGcgMoFAixSOszSoRsyX-E0,11
-textpolicy-0.1.0.dist-info/RECORD,,
+textpolicy-0.1.2.dist-info/licenses/LICENSE,sha256=AYDHSNRbiqZt4HHH1gaOoQ2hjYjK4bqw4Vd9UyKzx18,1065
+textpolicy-0.1.2.dist-info/METADATA,sha256=HXAh6fGcTtNez86WFNlr6OnIQZNcswptUXPnBSmXQHM,3895
+textpolicy-0.1.2.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
+textpolicy-0.1.2.dist-info/entry_points.txt,sha256=d0Cj5boT6k_l_beVPWPt9LZMllsN4kbIUmsNsn1BANE,51
+textpolicy-0.1.2.dist-info/top_level.txt,sha256=Ww6_QEF71dI-AYCaugiGeGcgMoFAixSOszSoRsyX-E0,11
+textpolicy-0.1.2.dist-info/RECORD,,

{textpolicy-0.1.0.dist-info → textpolicy-0.1.2.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (80.9.0)
+Generator: setuptools (80.10.2)
 Root-Is-Purelib: true
 Tag: py3-none-any