plancraft 0.3.7__py3-none-any.whl → 0.3.9__py3-none-any.whl

plancraft/config.py

@@ -64,7 +64,9 @@ class LocalEnvSettings(BaseSettings):
     openai_api_key: str = ""
     wandb_api_key: str = ""
 
-    model_config = SettingsConfigDict(env_file=".env", env_file_encoding="utf-8")
+    model_config = SettingsConfigDict(
+        env_file=".env", env_file_encoding="utf-8", extra="ignore"
+    )
 
 
 class EvalConfig(BaseModel):

plancraft/evaluator.py

@@ -194,7 +194,6 @@ class Evaluator:
         """Given the loaded model and an example from Plancraft
         run the episode until success or termination."""
         success = False
-        num_non_env_actions = 0
         self.reset(example)
         action = None
 
@@ -209,24 +208,18 @@ class Evaluator:
                 success = example.impossible
                 break
             # action is external tool then it is str
-            # limit the number of consecutive non-env actions to 3
-            elif isinstance(action, str) and num_non_env_actions < 3:
-                observation = {"message": action}
-                num_non_env_actions += 1
+            if isinstance(action, str):
+                observation = self.environment.step()
+                observation["target"] = example.target
+                observation["message"] = action
             # action is environment action
             else:
-                if isinstance(action, str):
-                    observation = self.environment.step()
-                else:
-                    observation = self.environment.step(action)
-
+                observation = self.environment.step(action)
                 # convert inventory observation to text message
                 observation["target"] = example.target
                 observation["message"] = self.convert_observation_to_message(
                     observation
                 )
-                num_non_env_actions = 0
-
                 # check if the episode is done
                 success = self.check_done(observation["inventory"], example.target)
             # exit if success

{plancraft-0.3.7.dist-info → plancraft-0.3.9.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: plancraft
-Version: 0.3.7
+Version: 0.3.9
 Summary: Plancraft: an evaluation dataset for planning with LLM agents
 License: MIT License
         

{plancraft-0.3.7.dist-info → plancraft-0.3.9.dist-info}/RECORD

@@ -1,6 +1,6 @@
 plancraft/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-plancraft/config.py,sha256=Ppkps-E8xDNYEP9prOVxW2zEG9MpWVzcLJi4tmGLjuQ,4285
-plancraft/evaluator.py,sha256=zWk3k1EiqGOIJkSWGL5Qk27xqwLowUMwRRVc6dm0gGo,11037
+plancraft/config.py,sha256=ShsFRlJ7plsl3ToD9fiO_4LDQuXdbjNV6Xp6o3Yk2Yg,4315
+plancraft/evaluator.py,sha256=FogXAjnldK2K1gD02f2WW1YWvjWokKGUOJxwtwzGNjk,10804
 plancraft/generate_dataset.py,sha256=DlrU-PmvWqSNJD1g1-8Lpb8n3N-Ogw3rje1nrRzjGKs,2382
 plancraft/utils.py,sha256=phaHzbIS85YZrBPaGG9TStHY8ZBKR1LKfuN1exfVy1U,6889
 plancraft/data/test.json,sha256=4jWfYMAVuZCFmGB4iZJAjlh9_8jXECdaGp8xn7_tAM4,1317131
@@ -1920,7 +1920,7 @@ plancraft/models/generators.py,sha256=F76_iPiqxUjDIrQwF58tzM0bLM91OkZJ0sBqBuki5w
 plancraft/models/oracle.py,sha256=jDCE6zVFvbwFpDzQZTkHIlRwMud1yMJ4LVIdfpt5ddU,8449
 plancraft/models/utils.py,sha256=E-sZohvolWgGbpHQKgAgkgIfUJoVnT5pMt6JP8xLHKg,4034
 plancraft/train/dataset.py,sha256=oFqEd4LG9oEQ-71teh0Wf7-jJbtybT2ZibfM2bBdBkM,5474
-plancraft-0.3.7.dist-info/METADATA,sha256=fZOUxkStOAD_MOeokSk-MNcsFIf5m2DmrJnALuNrA9Q,11147
-plancraft-0.3.7.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
-plancraft-0.3.7.dist-info/licenses/LICENSE,sha256=YGR8ehDB4t-T-lOQKMfKNR-2zsOU7E3E5NA8t25HKE0,1070
-plancraft-0.3.7.dist-info/RECORD,,
+plancraft-0.3.9.dist-info/METADATA,sha256=zXrxilbtrVK05xX0jjFbDAub_tToabPX0wX1TULyUjA,11147
+plancraft-0.3.9.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
+plancraft-0.3.9.dist-info/licenses/LICENSE,sha256=YGR8ehDB4t-T-lOQKMfKNR-2zsOU7E3E5NA8t25HKE0,1070
+plancraft-0.3.9.dist-info/RECORD,,