plancraft 0.3.18__py3-none-any.whl → 0.3.19__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
plancraft/evaluator.py CHANGED
@@ -4,7 +4,6 @@ from typing import Optional
4
4
  from copy import deepcopy
5
5
 
6
6
  import imageio
7
- from loguru import logger
8
7
  from tqdm import tqdm
9
8
 
10
9
  import wandb
@@ -293,6 +292,8 @@ class Evaluator:
293
292
  # Get observations for all active environments
294
293
  observations = []
295
294
  active_indices = []
295
+ obs_mapping = {} # Map active_indices to their observation index
296
+
296
297
  for i, (env, action, active) in enumerate(
297
298
  zip(environments, actions, active_mask)
298
299
  ):
@@ -311,7 +312,6 @@ class Evaluator:
311
312
  "example_id": examples[i].id,
312
313
  "images": histories[i].images,
313
314
  }
314
- logger.info("STOP")
315
315
  continue
316
316
 
317
317
  active_indices.append(i)
@@ -342,6 +342,9 @@ class Evaluator:
342
342
  }
343
343
  continue
344
344
 
345
+ obs_mapping[i] = len(
346
+ observations
347
+ ) # Map active index to observation index
345
348
  observations.append(obs)
346
349
  histories[i].add_observation_to_history(obs)
347
350
  histories[i].add_message_to_history(content=obs["message"], role="user")
@@ -358,13 +361,14 @@ class Evaluator:
358
361
 
359
362
  # Process actions for each active environment
360
363
  for idx, raw_action in zip(active_indices, raw_actions):
361
- logger.info(f"{histories[idx].num_steps}, {raw_action}")
362
364
  histories[idx].add_message_to_history(
363
365
  content=raw_action, role="assistant"
364
366
  )
365
367
  actions[idx] = self.parse_raw_model_response(
366
368
  raw_action,
367
- observation=observations[active_indices.index(idx)],
369
+ observation=observations[
370
+ obs_mapping[idx]
371
+ ], # Use mapping to get correct observation
368
372
  history=histories[idx],
369
373
  )
370
374
 
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: plancraft
3
- Version: 0.3.18
3
+ Version: 0.3.19
4
4
  Summary: Plancraft: an evaluation dataset for planning with LLM agents
5
5
  License: MIT License
6
6
 
@@ -1,6 +1,6 @@
1
1
  plancraft/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
2
2
  plancraft/config.py,sha256=ShsFRlJ7plsl3ToD9fiO_4LDQuXdbjNV6Xp6o3Yk2Yg,4315
3
- plancraft/evaluator.py,sha256=v8itX8buduqTZdR39gtLwdhKGEnSX3rJv9Yd13EzNgQ,16395
3
+ plancraft/evaluator.py,sha256=SXg7UoZ9KqjxgLkBo53BMw9XdR2nwMx7mDEXVwVhBiM,16544
4
4
  plancraft/generate_dataset.py,sha256=DlrU-PmvWqSNJD1g1-8Lpb8n3N-Ogw3rje1nrRzjGKs,2382
5
5
  plancraft/utils.py,sha256=67UUDMSv8TqX_I0fL5-yG_vkHvTZlnhSLkktWAg5p34,5712
6
6
  plancraft/data/test.json,sha256=4jWfYMAVuZCFmGB4iZJAjlh9_8jXECdaGp8xn7_tAM4,1317131
@@ -1920,7 +1920,7 @@ plancraft/models/generators.py,sha256=F76_iPiqxUjDIrQwF58tzM0bLM91OkZJ0sBqBuki5w
1920
1920
  plancraft/models/oracle.py,sha256=f-0KWlBuHy6wcxmDsxM3MQ_QwfBstzfbA26mlk1MgLA,1657
1921
1921
  plancraft/models/utils.py,sha256=E-sZohvolWgGbpHQKgAgkgIfUJoVnT5pMt6JP8xLHKg,4034
1922
1922
  plancraft/train/dataset.py,sha256=oFqEd4LG9oEQ-71teh0Wf7-jJbtybT2ZibfM2bBdBkM,5474
1923
- plancraft-0.3.18.dist-info/METADATA,sha256=p_Ln_3jx77ygBZG6yjuLhVs883PysUXUCi1sK67QvJs,11148
1924
- plancraft-0.3.18.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
1925
- plancraft-0.3.18.dist-info/licenses/LICENSE,sha256=YGR8ehDB4t-T-lOQKMfKNR-2zsOU7E3E5NA8t25HKE0,1070
1926
- plancraft-0.3.18.dist-info/RECORD,,
1923
+ plancraft-0.3.19.dist-info/METADATA,sha256=mA9CYITxxP37FICRAeVEq9j1WyBEeaWnVAPJsQnA1Ho,11148
1924
+ plancraft-0.3.19.dist-info/WHEEL,sha256=qtCwoSJWgHk21S1Kb4ihdzI2rlJ1ZKaIurTj_ngOhyQ,87
1925
+ plancraft-0.3.19.dist-info/licenses/LICENSE,sha256=YGR8ehDB4t-T-lOQKMfKNR-2zsOU7E3E5NA8t25HKE0,1070
1926
+ plancraft-0.3.19.dist-info/RECORD,,