renpas22 commited on
Commit ·
7bff7cb
1
Parent(s): 714d05d
Make train_prm and train_rl placeholders - dataset loading needs HF integration
Browse files
src/reasoning/step_level_cot.py
CHANGED
|
@@ -370,9 +370,10 @@ class StepLevelCoTTrainer:
|
|
| 370 |
if kwargs:
|
| 371 |
logger.info(f"PRM config: {kwargs}")
|
| 372 |
|
| 373 |
-
#
|
| 374 |
-
|
| 375 |
-
|
|
|
|
| 376 |
|
| 377 |
prm_trainer = PRMTrainer(
|
| 378 |
model=self.prm,
|
|
@@ -470,11 +471,10 @@ class StepLevelCoTTrainer:
|
|
| 470 |
if kwargs:
|
| 471 |
logger.info(f"PPO config: {kwargs}")
|
| 472 |
|
| 473 |
-
#
|
| 474 |
-
|
| 475 |
-
|
| 476 |
-
|
| 477 |
-
num_iterations = max_steps // self.config.train_batch_size
|
| 478 |
|
| 479 |
self.rl_trainer.train(
|
| 480 |
train_dataset=train_dataset,
|
|
|
|
| 370 |
if kwargs:
|
| 371 |
logger.info(f"PRM config: {kwargs}")
|
| 372 |
|
| 373 |
+
# Placeholder for PRM training - needs HuggingFace dataset integration
|
| 374 |
+
logger.warning("PRM training placeholder - requires dataset loading implementation")
|
| 375 |
+
logger.info("Skipping PRM phase - method needs full implementation with HuggingFace datasets")
|
| 376 |
+
return
|
| 377 |
|
| 378 |
prm_trainer = PRMTrainer(
|
| 379 |
model=self.prm,
|
|
|
|
| 471 |
if kwargs:
|
| 472 |
logger.info(f"PPO config: {kwargs}")
|
| 473 |
|
| 474 |
+
# Placeholder for RL training - needs HuggingFace dataset integration
|
| 475 |
+
logger.warning("RL training placeholder - requires dataset loading implementation")
|
| 476 |
+
logger.info("Skipping RL phase - method needs full implementation with HuggingFace datasets")
|
| 477 |
+
return
|
|
|
|
| 478 |
|
| 479 |
self.rl_trainer.train(
|
| 480 |
train_dataset=train_dataset,
|